Intelligent puncture positioning system for neuroendoscopic surgery based on image navigation

Abstract

The invention relates to the technical field of neurosurgical intelligent operations, and discloses a neuroendoscopic surgery intelligent puncture positioning system based on image navigation, and the system comprises a preoperative modeling module which constructs a preoperative tissue model and a corresponding elastic tensor field based on a three-dimensional medical image of a patient; the image fusion module is used for receiving the data of the preoperative modeling module, collecting intraoperative endoscopic images, predicting the tensor disturbance of tissues through a deep learning model, and feeding back the updated tensor disturbance to the tensor updating module; and the tensor updating module is in two-way communication with the image fusion module and is used for superposing tensor disturbance obtained by processing the intraoperative image with the preoperative tensor. Through fusion of an image navigation technology and preoperative modeling, a three-dimensional positioning frame based on neuroendoscopy operation requirements is constructed, intelligent planning and real-time correction of a puncture path are realized, important brain regions and vascular structures are effectively avoided, intra-operative risks are reduced, and positioning precision and safety are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent neurosurgery, and in particular to an intelligent puncture positioning system for neuroendoscopic surgery based on image navigation. Background Art

[0002] With the development of minimally invasive neurosurgery, neuroendoscopy, as an important surgical method, is widely used in delicate operations such as ventricular system lesions, intracranial cysts, and tumor resection. In such surgeries, the selection of preoperative puncture paths and the accuracy of intraoperative navigation are directly related to the safety and efficacy of the surgery.

[0003] Although some navigation systems currently incorporate preoperative imaging for assisted positioning, most still rely on the surgeon's subjective experience to judge image information, and puncture path planning lacks an intelligent optimization mechanism. Furthermore, existing systems often struggle to efficiently integrate imaging and endoscopic vision during surgery, making it difficult to provide real-time feedback of positioning information to the surgical process, which in turn affects operational accuracy. Especially during the puncture process, once a path deviation occurs, existing technologies struggle to identify and dynamically adjust it in a timely manner. The lack of effective closed-loop control increases the risk of accidental injury to functional areas or blood vessels.

[0004] Therefore, existing technologies still have obvious limitations in terms of precision control, intelligent path decision-making, and dynamic feedback during surgery. A system with real-time navigation and intelligent path control capabilities is urgently needed to solve these problems. Summary of the Invention

[0005] In response to the deficiencies of the existing technology, the present invention provides an intelligent puncture positioning system for neuroendoscopic surgery based on image navigation, which solves the problems of inaccurate positioning, unstable path and insufficient feedback mechanism in existing surgery.

[0006] To achieve the above objectives, the present invention is implemented through the following technical solutions: an intelligent puncture positioning system for neuroendoscopic surgery based on image navigation, comprising:

[0007] Preoperative modeling module, which constructs preoperative tissue models and corresponding elastic tensor fields based on the patient's 3D medical images;

[0008] An image fusion module receives data from the preoperative modeling module and collects intraoperative endoscopic images, predicts the tensor perturbation of the tissue through a deep learning model, and feeds the updated tensor perturbation back to the tensor update module;

[0009] a tensor update module, communicating bidirectionally with the image fusion module, for superimposing the tensor perturbation obtained by intraoperative image processing with the preoperative tensor to generate a three-dimensional elastic tensor field of the current tissue, and outputting the updated tensor field to the path optimization module;

[0010] a path optimization module, connected to the tensor update module and in bidirectional communication with the strategy guidance module, for calculating the puncture path based on the updated three-dimensional elastic tensor field and the multi-objective cost function, and sending the puncture path information to the strategy guidance module for further decision adjustment;

[0011] a strategy guidance module, communicating bidirectionally with the path optimization module, for guiding path decisions using a reinforcement learning model, providing feedback information to the path optimization module based on path adjustment suggestions, and generating corresponding path adjustment suggestions;

[0012] A feedback control module, which communicates bidirectionally with the strategy guidance module, the path optimization module, and the tensor update module, is used to trigger path updates according to intraoperative image changes or path status, and transmit path adjustment control signals to each module for adaptive correction;

[0013] The navigation display module communicates bidirectionally with the feedback control module, is used to receive path adjustment information and risk prompts, and synchronously display the updated path and real-time risk prompts on the intraoperative navigation interface, and at the same time feed back the display effect to the feedback control module.

[0014] Preferably, the preoperative modeling module includes:

[0015] Anatomical structure segmentation unit, used to automatically segment neural structures, blood vessels, and lesion areas in 3D medical images;

[0016] The elastic tensor generating unit is used to calculate the symmetric positive definite elastic tensor of each voxel based on the anatomical structure segmentation result, so as to represent the local anisotropy of the tissue.

[0017] Preferably, the image fusion module includes:

[0018] Endoscopic image acquisition unit, used to obtain intraoperative image streams in real time;

[0019] Image feature extraction network, used to extract deformation edges, structural boundaries and grayscale texture information;

[0020] Tensor perturbation prediction network for estimating voxel-level tensor perturbation distribution based on image features.

[0021] Preferably, the tensor perturbation prediction network adopts a convolutional neural network with an encoding-decoding structure, and improves the prediction accuracy by jointly encoding the preoperative tensor and image features.

[0022] Preferably, the tensor update module updates the elastic tensor field through a tensor fusion formula, and the tensor fusion formula is in the form of:

[0023] T(x,y,z,t)=T0(x,y,z)+ΔD(x,y,z,t)

[0024] Where T0(x, y, z) is the reference tissue position mapping at the spatial position (x, y, z) at the initial time, ΔD(x, y, z, t) is the incremental displacement of the tissue at the position (x, y, z) from the initial time to time t, and T(x, y, z, t) is the current tissue position mapping at the spatial position (x, y, z) at time t.

[0025] Preferably, the path optimization module constructs a multi-objective cost function for path planning, and the multi-objective cost function is in the form of:

[0026]

[0027] Among them, d i is the path segment length; r i is the organizational risk coefficient; δ i is the local tensor perturbation value; w d 、w r and w t is an adjustable weight coefficient.

[0028] Preferably, the policy guidance module provides feedback information and generates corresponding path adjustment suggestions through the following steps, specifically including:

[0029] Construct a state space model that includes the current position of the path, the current tensor perturbation, and risk information;

[0030] Use reinforcement learning models to evaluate the current state and generate path adjustment suggestions;

[0031] Based on the generated path adjustment suggestions, feedback information is provided to the path optimization module to adjust the current path decision;

[0032] Based on the feedback information, the path planning is adjusted to optimize the final decision of the puncture path.

[0033] Preferably, the reinforcement learning model in the strategy guidance module is trained by setting a reward and punishment function, and the form of the reward and punishment function is:

[0034]

[0035] Among them, R t The reward value adjusted for the current path; D t is the path length; is the risk value of the path; is the error of the path reaching the target area; α, β and γ are weight coefficients.

[0036] Preferably, the feedback control module triggers the path update operation based on one of the following conditions:

[0037] The total amount of current path disturbance exceeds the set threshold;

[0038] The rising rate of the path cost function exceeds the set slope;

[0039] The strategy guidance module outputs a suggestion that deviates from the current path direction by more than a preset tolerance.

[0040] Preferably, the navigation display module is used to display the current puncture path, surrounding risk tissue and tensor disturbance distribution in a three-dimensional superposition manner, and provide graphic annotation prompts for path adjustment suggestions.

[0041] The present invention provides an intelligent puncture positioning system for neuroendoscopic surgery based on image navigation. It has the following beneficial effects:

[0042] 1. This invention integrates image navigation technology with preoperative modeling to construct a three-dimensional positioning framework based on the needs of neuroendoscopic operations, realizes intelligent planning and real-time correction of the puncture path, effectively avoids important brain areas and vascular structures, reduces intraoperative risks, and improves positioning accuracy and safety.

[0043] 2. The present invention dynamically aligns intraoperative endoscopic images with preoperative imaging data, and achieves real-time synchronization between the operating field of view and the three-dimensional positioning space through a spatial recognition algorithm, enabling doctors to accurately grasp the relative relationship between endoscopic displacement and anatomical structure during surgery, thereby improving navigation intuitiveness and operational efficiency.

[0044] 3. The present invention introduces a path optimization algorithm based on a learning model, which comprehensively analyzes image features, tissue density, and spatial channel patency, intelligently recommends the optimal puncture path, assists doctors in making preoperative strategic decisions, and effectively reduces the dependence of human experience on positioning results.

[0045] 4. The present invention designs an intraoperative feedback module that combines sensors and image recognition results to monitor puncture deviations in real time, and provides intelligent warnings or automatic correction suggestions when the path deviates, ensuring that the operation process is always performed within the planned channel, thereby enhancing the stability and controllability of the surgical process.

[0046] 5. Based on an adjustable imaging parameter and navigation strategy configuration framework, the present invention supports the construction of personalized puncture positioning solutions according to lesion type, location, and patient characteristics, enabling flexible adaptation to different neuroendoscopic surgery scenarios and enhancing the versatility and clinical application value of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 Schematic diagram of the system architecture of the present invention. DETAILED DESCRIPTION

[0048] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0049] Please see the attached Figure 1 The present invention provides an intelligent puncture positioning system for neuroendoscopic surgery based on image navigation, the system comprising:

[0050] Preoperative modeling module, which constructs preoperative tissue models and corresponding elastic tensor fields based on the patient's 3D medical images;

[0051] An image fusion module receives data from the preoperative modeling module and collects intraoperative endoscopic images, predicts the tensor perturbation of the tissue through a deep learning model, and feeds the updated tensor perturbation back to the tensor update module;

[0052] a tensor update module, communicating bidirectionally with the image fusion module, for superimposing the tensor perturbation obtained by intraoperative image processing with the preoperative tensor to generate a three-dimensional elastic tensor field of the current tissue, and outputting the updated tensor field to the path optimization module;

[0053] a path optimization module, connected to the tensor update module and in bidirectional communication with the strategy guidance module, for calculating the puncture path based on the updated three-dimensional elastic tensor field and the multi-objective cost function, and sending the puncture path information to the strategy guidance module for further decision adjustment;

[0054] a strategy guidance module, communicating bidirectionally with the path optimization module, for guiding path decisions using a reinforcement learning model, providing feedback information to the path optimization module based on path adjustment suggestions, and generating corresponding path adjustment suggestions;

[0055] A feedback control module, which communicates bidirectionally with the strategy guidance module, the path optimization module, and the tensor update module, is used to trigger path updates according to intraoperative image changes or path status, and transmit path adjustment control signals to each module for adaptive correction;

[0056] The navigation display module communicates bidirectionally with the feedback control module, is used to receive path adjustment information and risk prompts, and synchronously display the updated path and real-time risk prompts on the intraoperative navigation interface, and at the same time feed back the display effect to the feedback control module.

[0057] The following is a detailed description of each component in the system of the present invention.

[0058] As for the preoperative modeling module, in this embodiment, the preoperative modeling module is used to perform tissue structure modeling and parametric expression of elastic properties of the patient's target area before neuroendoscopic surgery, aiming to provide stable and continuous basic tissue reference data for intraoperative path planning and tensor perturbation identification.

[0059] This module performs processing based on the patient's three-dimensional medical imaging data, which may be preferably magnetic resonance imaging (MRI) data or computed tomography (CT) data. The image data should have sufficient resolution to support the segmentation of fine anatomical structures.

[0060] In this embodiment, the input medical image is first processed by the anatomical structure segmentation unit. Based on an image semantic segmentation network, this unit preferably employs a multi-scale feature extraction and contextual information fusion strategy to accurately segment key structures in the image, such as neural bundles, ventricular systems, vascular structures, and lesion tissue. By introducing a spatial attention mechanism, the model's ability to identify regions with blurred boundaries is further enhanced, ensuring clear structural boundaries and accurate topological relationships.

[0061] After image segmentation, the elasticity modeling of the tissue within the segmented region is performed using an elastic tensor generation unit. This unit uses the anatomical classification of the tissue as a criterion and incorporates biomechanical assumptions from the literature to construct a tensor field using voxel-level parameter mapping. A three-dimensional symmetric positive definite tensor is constructed at each voxel to describe the stress response directionality and elastic anisotropy characteristics of the tissue within the local region.

[0062] Tensor construction follows the following tensor parameter expression model:

[0063]

[0064] Where T0(x,y,z) is the tensor value of the tensor field at voxel (x,y,z) before surgery; λ k is the tensor eigenvalue corresponding to the principal axis direction, indicating the elastic response strength in that direction; T is the vector transpose operation; v k is the eigenvector, indicating the principal axis direction in this direction. This tensor satisfies symmetry and positive definiteness, which is consistent with the characteristics of the local continuum mechanics model of biological tissue.

[0065] The tensor generation process not only considers the anatomical classification after segmentation but also introduces a method for estimating the microstructural direction based on the image grayscale texture. Preferably, principal component analysis (PCA) is combined with the regional grayscale covariance matrix to extract the main local structural directions of the tissue, thereby placing a priori constraints on the tensor feature directions and improving the anatomical rationality and continuity of the tensor field.

[0066] The generated preoperative tensor field serves as the initial baseline data for fusion with the tensor perturbations output by the intraoperative image fusion module. This tensor field is stored as a 3D voxel grid and serves as the basic input data for subsequent tensor updates and path planning modules.

[0067] It is worth noting that the output results of the preoperative modeling module not only have tissue structure topological information, but also carry the directional mechanical characteristics of each tissue voxel, which is one of the core inputs supporting the puncture path planning and risk assessment algorithm of the present invention.

[0068] In summary, the preoperative modeling module establishes a three-dimensional tissue model that conforms to physiological anatomy and elastic properties through precise structural segmentation and tensor modeling methods, and plays a stable, continuous, and updateable structural benchmark role in intraoperative navigation and path reconstruction.

[0069] As for the image fusion module, in this embodiment, the image fusion module is used to collect endoscopic images in real time during neuroendoscopic surgery, and dynamically compare and analyze the image with the preoperative tensor model to predict the tensor disturbance of the current tissue area for subsequent tensor update and path optimization processing.

[0070] The image fusion module includes an endoscopic image acquisition unit, an image feature extraction network, and a tensor perturbation prediction network. The entire module operates in a streaming data mode, ensuring continuous and real-time image data processing and model inference.

[0071] During system operation, the endoscopic image acquisition unit is preferably integrated into the neuroendoscopic device, acquiring intraoperative image sequences in real time through the image acquisition interface. The acquired images are stored in RGB or grayscale format and sequentially labeled according to timestamps for subsequent deep model processing.

[0072] The image feature extraction network receives intraoperative images from the image acquisition unit and jointly extracts low-level texture and mid-level structural features. This network preferably employs a deep convolutional architecture and, through a multi-scale feature fusion strategy, extracts deformation edges, tissue structure boundaries, and grayscale texture information from the image, ensuring that the downstream prediction network can fully capture the potential manifestations of tensor perturbations.

[0073] To improve the coupling accuracy between image features and the preoperative tensor model, this embodiment inputs the voxel features of the preoperative tensor field as a 3D tensor and co-encodes them with the image features during the encoding phase. This approach achieves semantic correspondence between image features and the preoperative model through a spatial alignment mechanism, facilitating the inference of the impact of tissue deformation on tensor features during the prediction phase.

[0074] After joint encoding, the tensor perturbation prediction network inputs the joint image-tensor features into the decoder structure. Through spatial upsampling and feature fusion, it predicts the tensor perturbation increment ΔD(x, y, z, t) at each voxel position. This perturbation tensor is a symmetric second-order tensor that reflects the mechanical property deviation of the tissue caused by surgical manipulation or changes in physiological state during surgery.

[0075] In this embodiment, the tensor perturbation prediction network can adopt an encoder-decoder symmetric structure, where the encoder is used to extract abstract semantic features and the decoder is used to reconstruct the spatial distribution of the perturbation tensor. To improve the continuity and stability of perturbation prediction, a residual connection mechanism is introduced into the network, and a tensor field smoothness constraint term is introduced into the loss function. The specific form is as follows:

[0076]

[0077] in, is the mean squared error between the perturbation prediction tensor and the true perturbation; is the gradient norm constraint of the tensor perturbation field, which is used to suppress local noise; λ is the balance factor.

[0078] After the tensor perturbation prediction is completed, the output ΔD(x, y, z, t) will be transmitted to the tensor update module through bidirectional communication, and fused with the preoperative tensor field to construct the dynamic tensor field of the tissue at the current moment.

[0079] This image fusion module achieves effective connection between intraoperative image information and preoperative elastic models, enabling the system to perceive the real-time deformation and tensor disturbance of tissues during surgery, providing a dynamic update basis for path planning and risk assessment, thereby enhancing the adaptability and safety of the surgery.

[0080] Regarding the tensor update module, in this embodiment, the tensor update module is used to update the three-dimensional elastic tensor field constructed before surgery in real time according to the tensor disturbance information output by the image fusion module during the intraoperative stage, thereby reflecting the dynamic changes of the current tissue state and providing a timeliness basis for subsequent path optimization and risk assessment.

[0081] The tensor update module maintains a bidirectional communication connection with the image fusion module. After processing the intraoperative images, the image fusion module outputs a voxel-level perturbation tensor field, representing the impact of intraoperative tissue deformation on local elastic directionality and strength. The tensor update module receives this perturbation data and fuses it with the preoperative tensor field to generate the new current tensor field data.

[0082] In this embodiment, the tensor update process follows the linear perturbation superposition model, and the mathematical expression of the tensor update operation is:

[0083] T(x,y,z,t)=T0(x,y,z)+ΔD(x,y,z,t);

[0084] Among them, T0(x, y, z) is the static tensor field output by the preoperative modeling module, which describes the elastic properties of the tissue at the voxel (x, y, z) in the undisturbed state; ΔD(x, y, z, t) is the disturbance tensor estimated by the image fusion module based on the intraoperative image, which reflects the tensor change caused by intraoperative deformation; T(x, y, z, t) is the current tensor after real-time update.

[0085] This updated model, based on the premise that tissue elastic behavior is approximately linearly additive at small scales, is suitable for describing the strain response of neural tissue during minimally invasive interventions. The updated tensor retains its symmetric positive definiteness, ensuring its physical meaning and numerical stability.

[0086] During actual system operation, the tensor update module continuously receives the perturbation tensor data from the image fusion module in a time series manner and periodically updates the original tensor field. To avoid the accumulation of perturbation errors, this embodiment preferably introduces an exponential sliding average strategy to smooth the perturbation time series. Its expression is:

[0087]

[0088] Where α is the disturbance response coefficient; ΔD (t) The perturbation tensor predicted for the current frame; is the smoothing result of the previous frame.

[0089] Through this strategy, potential transient noise in image perturbation prediction can be suppressed, achieving more stable and continuous tensor field updates.

[0090] The output of the tensor update module is expressed as a three-dimensional tensor grid, with each voxel corresponding to a complete second-order tensor object. This is stored as a symmetric matrix and transmitted to the path optimization module in a structured data format. This structured tensor field provides continuous support for the local elastic features involved in the path cost calculation.

[0091] In addition, the tensor update module in this embodiment can establish an interface with the feedback control module to notify the system to enter the path reconstruction process when a significant abnormality in the tensor disturbance is detected, thereby achieving real-time adaptive adjustment of the puncture path.

[0092] In summary, the tensor update module realizes the real-time transition and fusion between the preoperative static elastic model and the intraoperative dynamic tissue state, ensuring that the tensor data based on path optimization has high timeliness and biophysical consistency. It is an important component module for realizing the adaptive navigation mechanism of the present invention.

[0093] In this embodiment, the path optimization module is used to plan the neuroendoscopic puncture path during the intraoperative phase based on real-time updated tissue tensor field information and a multi-objective optimization strategy. The module's design goal is to construct a feasible path with minimal anatomical risk, low path disturbance, and reasonable geometry, thereby assisting surgeons in completing high-precision surgical navigation.

[0094] The path optimization module receives the current three-dimensional tissue elastic tensor field T(x, y, z, t) output from the tensor update module. This tensor field provides mechanical directional information of the tissue at different spatial locations in voxels, which can be used to evaluate the biomechanical response of the puncture path in different regions.

[0095] In order to fully reflect the multiple constraints of the path, a multi-objective path cost function is introduced in this embodiment. For any path Perform a comprehensive evaluation. The path cost function is specifically defined as follows:

[0096]

[0097] Among them, d i is the length of the i-th path segment in the path, which is used to constrain the total length of the path from unnecessary extension;

[0098] r i The tissue risk value of the area where the path segment is located is derived from the segmentation results of important anatomical structures (such as blood vessels and nerve bundles) in the preoperative modeling stage, combined with the risk map formed by the doctor's annotations;

[0099] δ i The tensor perturbation value at the corresponding position of the path segment is extracted from the tensor field provided by the tensor update module, reflecting the current elastic uncertainty or deformation degree of the tissue area where the path segment is located;

[0100] w d 、w r and w t It is an adjustable cost weight used to assign different importance to different target dimensions according to the technique or user preference.

[0101] The path optimization module preferably uses a graph search algorithm or a spatial sampling algorithm to complete path planning. Specifically, heuristic search (such as A*) or probabilistic sampling methods (such as RRT*) can be used. During the search process, a cost function serves as an evaluation criterion for path construction, guiding the algorithm to select the set of paths with the lowest cost in the candidate path space.

[0102] In order to further enhance the path smoothness and punctureability, this embodiment also introduces a curvature penalty term in the path evaluation process. Specifically, for a continuous path segment (pi ,p i+1 ,p i+2 ), calculate the angle change rate and include it in the overall cost:

[0103]

[0104] On this basis, the smoothing cost term is defined as:

[0105]

[0106] Where γ is the smoothness penalty weight; n is the total number of discrete points on the path; θ i is the angle at the i-th corner of the path, which is used to measure the degree of turning of the path at that point.

[0107] The final path cost function is updated as:

[0108]

[0109] in, For path The original price.

[0110] In this embodiment, a bidirectional communication channel is established between the path optimization module and the policy guidance module. The policy guidance module, based on a reinforcement learning mechanism, performs secondary evaluation and optimization of the path planning after completion. The feedback results are then fed back to the path optimization module for iterative correction, forming a dynamic optimization closed loop.

[0111] After the path optimization is completed, the final path The output is sent to the navigation display module and synchronously transmitted to the feedback control module for subsequent dynamic adjustment. The system supports a path recalculation trigger mechanism to address path failure caused by significant changes in tensor perturbations during surgery.

[0112] In summary, the path optimization module assumes the key responsibility of path decision-making in the system of the present invention, fully integrates the three-dimensional information of tissue structure risk, elastic disturbance and geometric cost, combines optimization strategy with control feedback, and realizes the dynamic adaptability construction of the puncture path at the spatial and physiological levels.

[0113] In this embodiment, the strategy guidance module uses a reinforcement learning algorithm to perform secondary optimization of the initial puncture path calculated by the path optimization module and real-time tensor perturbation information to enhance the adaptability and robustness of path planning in dynamic environments. This module continuously adjusts the path during surgery and triggers path recalculation when significant intraoperative changes occur, ensuring the accuracy and safety of the surgical path in practical applications.

[0114] The strategy guidance module receives the preliminary planned path from the path optimization module The core task of the module is to optimize the given path using a reinforcement learning framework to minimize biomechanical risks and surgical difficulty.

[0115] Construction of reinforcement learning framework

[0116] In this embodiment, the policy guidance module performs path optimization based on the reinforcement learning (RL) framework. The core idea of RL is to gradually improve the path planning strategy through trial and error based on feedback signals from the interaction between the current path and the environment. Specifically, the policy guidance module defines a Markov decision process (MDP) that includes the following elements:

[0117] State space S: The state at each moment represents the current path The relationship between the state and the environment (such as the tissue tensor perturbation field). The state includes the geometric information of the path, the tissue characteristics of the path segment, and the tensor perturbation information of the area where the path is located.

[0118] Action space A: path adjustment actions that can be taken at each moment, such as curvature adjustment, path offset, path replanning, etc.

[0119] Reward function R: Evaluates the path's performance in the current environment and assigns a corresponding reward or penalty to each action. The design of the reward function takes into account the path's length, risk, tensor perturbations, and path smoothness. The reward function is as follows:

[0120]

[0121] in, is the path length; is path risk; is the tensor perturbation of the path segment; C smooth is the path smoothness cost; w d 、w r 、w t and w smooth is a weight item used to adjust the priority of different goals.

[0122] Policy function π(a|s): represents the probability distribution of selecting action a in state s. The policy function is obtained through training and guides the module to select the optimal path adjustment strategy.

[0123] Value function V(s): represents the expected return of operating according to policy π in state s. The optimization goal of the value function is to maximize the expected reward.

[0124] Training and optimization of the strategy guidance module

[0125] The policy guidance module is trained using Deep Reinforcement Learning (DRL). Specifically, it uses a Deep Q-Network (DQN) or Proximal Policy Optimization (PPO) algorithm for policy learning. Through multiple iterations of training, the module gradually learns how to effectively adjust its path based on changes in the current path and the environment.

[0126] During reinforcement learning training, the module explores multiple simulated environments and continuously optimizes its strategy based on a reward function, ensuring that path planning remains optimal or near-optimal in the dynamically changing intraoperative environment. The goal of training is to learn an optimal path adjustment strategy using the policy function π, minimizing risk, tensor perturbations, and path curvature while maintaining path rationality and smoothness.

[0127] During training, the policy guidance module's feedback mechanism ensures the system can promptly respond to intraoperative changes. Specifically, if significant intraoperative tissue deformation causes the original path plan to fail or creates potential risks, the system triggers path replanning. Path replanning is based on the current state and perturbation tensor information, and the policy guidance module implements a new path optimization.

[0128] Collaboration between the Strategy Guidance Module and other modules

[0129] The policy guidance module works closely with the path optimization module and the tensor update module to ensure accurate and real-time path planning. During path planning, the path optimization module provides preliminary path estimates and assesses the path's rationality based on the current tensor perturbation information. Based on these inputs, the policy guidance module further adjusts the path to ensure feasibility and safety in real-time operation.

[0130] Furthermore, the strategy guidance module and the feedback control module are closely linked. When real-time tensor perturbation information indicates that the planned path fails to meet surgical requirements, the feedback control module initiates a path recalculation process. At this point, the strategy guidance module adjusts the optimization strategy based on the new perturbation information, ensuring that the planned path remains adaptable to the dynamic environment.

[0131] In summary, the strategy guidance module in the present system dynamically optimizes and adjusts the path using a reinforcement learning algorithm, combined with real-time tissue tensor perturbation data, to ensure the accuracy, robustness, and adaptability of path planning within the intraoperative environment. This module's design enhances the system's ability to cope with complex intraoperative changes, providing important support for precision medicine and surgical navigation.

[0132] In this embodiment, the feedback control module monitors the deviation between the planned path and the actual operation in real time during the surgery. Based on this real-time feedback, it adjusts the path or operation strategy to ensure the accuracy and safety of the path execution process. This module, through linkage with the path optimization module, the tensor update module, and the strategy guidance module, achieves full dynamic control, allowing for timely correction of the path plan if the intraoperative environment changes or if there are significant deviations.

[0133] Module function and working principle

[0134] The core task of the feedback control module is to analyze and respond to the feedback information of the intraoperative path execution in real time. Specifically, the module is based on the real-time tensor perturbation data provided by the tensor update module and the current path given by the path optimization module. Detect any deviations that may occur during path execution and make timely adjustments to the path.

[0135] The workflow of the feedback control module can be summarized into the following steps:

[0136] The feedback control module continuously monitors the actual execution of the path and compares it with the initially planned path. It is defined as the spatial distance between the planned path and the actual path, which is calculated as follows:

[0137]

[0138] in, is the actual position of the path at the current moment; The planned path provided by the path optimization module.

[0139] The path deviation reflects the error or drift during the path execution process. The feedback control module uses this as a basis to determine whether the path adjustment is needed.

[0140] Once the path deviation is detected to exceed the predetermined threshold, the feedback control module will trigger the path recalculation mechanism. The core of path recalculation is to re-evaluate whether the current path is still suitable for the intraoperative environment based on the tensor perturbation data of the current tissue and the path deviation information. The mathematical form of path recalculation can be expressed as:

[0141]

[0142] Among them, λ is the path adjustment weight factor, which is used to balance the relationship between path deviation and cost function; For path The cost function reflects the risk, complexity and other factors of the path.

[0143] Recalculated path It will be output as a new surgical navigation path for use by the surgical execution system.

[0144] The feedback control module not only triggers adjustments when path deviations occur, but also operates in closed-loop control with the policy guidance module to optimize path planning in real time. The policy guidance module uses a reinforcement learning algorithm to adjust the path planning strategy to adapt to changes during surgery. The feedback control module, based on real-time feedback signals and path deviations, dynamically assesses the feasibility of the current path and adjusts the optimization strategy as needed to ensure the continued effectiveness of the path at different stages.

[0145] The feedback control module assesses the impact of tissue elasticity changes on path planning based on the perturbation information provided by the tensor update module. When the tensor perturbation is large, the feedback control module adjusts the path appropriately to avoid or minimize disturbance to the tissue during surgery. This process involves locally optimizing the path based on the perturbation tensor field ΔD(x, y, z, t), with the goal of minimizing the risk of puncture in areas of high perturbation.

[0146] Feedback mechanism and module linkage

[0147] The feedback mechanism of the feedback control module closely cooperates with other modules. Specifically, the module interacts with the path optimization module, the tensor update module, and the policy guidance module as follows:

[0148] Path optimization module: provides preliminary path planning results and path cost evaluation. The feedback control module adjusts the path planning based on real-time feedback to reduce path deviation.

[0149] Tensor update module: provides real-time tissue elasticity tensor disturbance information. The feedback control module evaluates path deviation based on this disturbance information and adjusts the feasibility of the path.

[0150] Strategy guidance module: provides path optimization strategy based on reinforcement learning. The feedback control module adjusts the optimization strategy according to the path execution results to form a closed-loop feedback.

[0151] Through this multi-module synergy, the feedback control module ensures the continuity and accuracy of path execution. During surgery, if any environmental changes occur that affect path execution, the feedback control module adjusts the path based on real-time feedback to ensure a smooth surgical procedure.

[0152] Path adjustment and optimization goals

[0153] The goal of path adjustment and optimization is not only to minimize path deviation, but also to consider path smoothness, risk, and minimize organizational interference. To this end, the feedback control module comprehensively considers the following key factors when optimizing the path:

[0154] Optimize the length of the path to avoid it being too long or too short, and ensure the rationality of the path in the anatomical structure;

[0155] During the execution process, the path should avoid sharp turns or corners and maintain a certain degree of smoothness to reduce damage to the tissue;

[0156] Pathway planning should avoid high-risk areas, such as important blood vessels and nerve bundles, to reduce the risk of complications during surgery;

[0157] During execution, the path needs to be dynamically adjusted according to real-time tensor disturbance information to avoid high-disturbance areas, thereby reducing the impact of surgery on tissues.

[0158] In summary, the feedback control module in the system of this invention implements a dynamic closed-loop feedback mechanism between path execution and path planning. By real-time monitoring of path deviations and tensor perturbation information, it ensures the stability, accuracy, and safety of the surgical path during actual operation. This module effectively addresses changes in the intraoperative environment, enhances the system's adaptability, and provides strong technical support for precision medicine.

[0159] In this embodiment, the navigation display module is used to present real-time surgical path information, tissue status, tensor perturbations, and related feedback results to the operator in an intuitive and visual manner. The core function of this module is to integrate multi-dimensional data information and update it in real time to provide accurate surgical navigation support, helping doctors make efficient decisions and operations.

[0160] Module function and working principle

[0161] The Navigation Display Module interacts with the Path Optimization Module, the Tensor Update Module, the Feedback Control Module, and the Strategy Guidance Module to integrate the information processed by each module and generate a real-time updated visual output. This module primarily provides two functions: path navigation display and real-time feedback on the intraoperative status.

[0162] During the operation, the navigation display module will display the current path in real time The path information is generated by the path optimization module and the feedback control module and displayed on the user interface using real-time image rendering technology. The path can be displayed in 3D, using virtual reality or augmented reality technology to accurately display the path on an anatomical model within the patient's body. Changes in the path reflect intraoperative adjustments in real time, enabling precise navigation.

[0163] During the path display process, the system overlays the path with the current tissue status information to show the relationship between the path and the surrounding tissue. Specifically, the relative position of the path and the surrounding tissue can be displayed using different color coding, transparency, or annotation to help doctors identify the potential impact of the path on different tissue structures.

[0164] In addition to path information, the navigation display module also displays real-time updated tensor disturbance information. By working in conjunction with the tensor update module, the navigation display module can capture the current changes in the tissue's tensor field and display them as images or heat maps. Areas with large disturbance values are highlighted, displaying the tension changes in that area, thus alerting the operator to potential biomechanical risks.

[0165] The display of tensor perturbations is typically presented as a color scale, with low-perturbation areas shown in green or blue, and high-perturbation areas in red or orange. This information can help doctors identify high-risk areas of tissue and avoid excessive manipulation in these areas.

[0166] The navigation display module also dynamically updates the displayed path based on real-time path adjustment information from the feedback control module. Whenever the feedback control module makes a path adjustment, the navigation display module immediately updates the path display, ensuring the operator always sees the most up-to-date path information. Furthermore, path adjustment information is also provided via a graphical interface, indicating areas to focus on or avoid.

[0167] Through this real-time feedback mechanism, the navigation display module can provide more intuitive and real-time path adjustment guidance for surgery, thereby improving the accuracy and safety of the surgery.

[0168] The navigation display module also displays the overall progress of the surgery, such as the current puncture path progress and the estimated completion time. The system can adjust the estimated time of the surgery in real time based on the path execution deviation monitored by the feedback control module and automatically update the status display based on the path deviation.

[0169] At the same time, the system can remind doctors of potential dangers through dynamically updated path trajectories and current tissue status displays, and assist doctors in determining whether further adjustments are needed.

[0170] Fusion and rendering of display information

[0171] The navigation display module needs to fuse the data from different modules in a short time and display them in a visual way. To achieve this, this embodiment adopts a multi-source information fusion method, which is specifically implemented as follows:

[0172] The navigation display module acquires real-time data such as path planning, tensor perturbations, and surgical feedback through synchronization interfaces with various modules. All data is synchronized by timestamp to ensure timeliness and consistency. During data fusion, the module first aligns the spatial positions of the path planning and tensor perturbations to ensure accurate mapping of the different data sources in three-dimensional space.

[0173] Using 3D graphics rendering technology, the navigation display module visualizes information such as pathways, tissue structures, and tensor perturbations. Path information is presented as lines, curves, or 3D models, tissue structures are displayed through anatomical diagrams or CT / MRI image overlays, and tensor perturbations are represented through heat maps or color coding.

[0174] During the display process, the module can adjust the display angle according to the operator's needs, supporting interactive operations such as zooming and rotating, helping doctors to review the intraoperative situation from different angles. At the same time, the display interface automatically marks the relative position of the path and tissue structures, allowing doctors to quickly identify the relationship between the path and key anatomical structures.

[0175] Real-time updates and surgical adaptability

[0176] Another key feature of the navigation display module is its real-time update capability. Due to tissue deformation, tensor perturbations, and path deviations that may occur during surgery, the navigation display module dynamically adjusts the displayed content based on real-time input from the feedback control module. For example, if a path deviation occurs during intraoperative execution, the module immediately updates the displayed path based on the new path planning results and prompts the operator if path adjustments are necessary.

[0177] In addition, the navigation display module assesses the risk of the path based on the current tensor disturbance data. If the current path passes through a high-disturbance area, the system will pop up a real-time warning, prompting the doctor to avoid these high-risk areas to ensure surgical safety.

[0178] In summary, the navigation display module in this embodiment provides efficient and accurate visualization support for surgery through real-time path navigation display, tensor perturbation display, feedback path adjustment, and multidimensional data fusion. This module not only improves the accuracy of surgical navigation but also enhances the surgeon's ability to adapt to various dynamic changes during surgery, thus providing a strong guarantee for precision medicine and surgical safety.

[0179] In general, the workflow of the system of the present invention can be described as follows:

[0180] The system uses endoscopic equipment to collect intraoperative images in real time, aligns them with the preoperative model, extracts image features, and obtains real-time information about the tissue; based on the collected image data, it predicts the tensor perturbation of the intraoperative tissue, updates the preoperative elastic tensor field, and obtains real-time dynamic tensor information; the path optimization module calculates the optimal puncture path, comprehensively considers factors such as path length, tissue risk, and tensor perturbation, and plans a preliminary path; the strategy guidance module dynamically optimizes the path based on reinforcement learning, adjusts the path to adapt to intraoperative changes, and minimizes risks; the feedback control module monitors the path execution in real time, detects path deviations, and triggers path corrections to ensure the accuracy and safety of the surgical path; the navigation display module displays the optimized path, tensor perturbation information, and real-time status to the operator in a three-dimensional visual manner, providing surgical navigation support; the entire system continuously performs real-time data updates and path adjustments to ensure the adaptability and accuracy of the path to cope with changes in the intraoperative environment.

[0181] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. An intelligent puncture positioning system for neuroendoscopic surgery based on image navigation, characterized by: include: Preoperative modeling module, which constructs preoperative tissue models and corresponding elastic tensor fields based on the patient's 3D medical images; An image fusion module receives data from the preoperative modeling module and collects intraoperative endoscopic images, predicts the tensor perturbation of the tissue through a deep learning model, and feeds the updated tensor perturbation back to the tensor update module; a tensor update module, communicating bidirectionally with the image fusion module, for superimposing the tensor perturbation obtained by intraoperative image processing with the preoperative tensor to generate a three-dimensional elastic tensor field of the current tissue, and outputting the updated tensor field to the path optimization module; a path optimization module, connected to the tensor update module and in bidirectional communication with the strategy guidance module, for calculating the puncture path based on the updated three-dimensional elastic tensor field and the multi-objective cost function, and sending the puncture path information to the strategy guidance module for further decision adjustment; a strategy guidance module, communicating bidirectionally with the path optimization module, for guiding path decisions using a reinforcement learning model, providing feedback information to the path optimization module based on path adjustment suggestions, and generating corresponding path adjustment suggestions; A feedback control module, which communicates bidirectionally with the strategy guidance module, the path optimization module, and the tensor update module, is used to trigger path updates according to intraoperative image changes or path status, and transmit path adjustment control signals to each module for adaptive correction; The navigation display module communicates bidirectionally with the feedback control module, is used to receive path adjustment information and risk prompts, and synchronously display the updated path and real-time risk prompts on the intraoperative navigation interface, and at the same time feed back the display effect to the feedback control module.

2. The image-guided neuroendoscopic surgery intelligent puncture positioning system according to claim 1 is characterized in that: The preoperative modeling module includes: Anatomical structure segmentation unit, used to automatically segment neural structures, blood vessels, and lesion areas in 3D medical images; The elastic tensor generating unit is used to calculate the symmetric positive definite elastic tensor of each voxel based on the anatomical structure segmentation result, so as to represent the local anisotropy of the tissue.

3. The intelligent puncture positioning system for neuroendoscopic surgery based on image navigation according to claim 1 is characterized in that: The image fusion module includes: Endoscopic image acquisition unit, used to obtain intraoperative image streams in real time; Image feature extraction network, used to extract deformation edges, structural boundaries and grayscale texture information; Tensor perturbation prediction network for estimating voxel-level tensor perturbation distribution based on image features.

4. The image-guided neuroendoscopic surgery intelligent puncture positioning system according to claim 3 is characterized in that: The tensor perturbation prediction network adopts a convolutional neural network with an encoding-decoding structure, and improves the prediction accuracy by jointly encoding preoperative tensors and image features.

5. The intelligent puncture positioning system for neuroendoscopic surgery based on image navigation according to claim 1 is characterized in that: The tensor update module updates the elastic tensor field through the tensor fusion formula, and the tensor fusion formula is in the form of: T(x,y,z,t)=T0(x,y,z)+ΔD(x,y,z,t) Where T0(x, y, z) is the reference tissue position mapping at the spatial position (x, y, z) at the initial time, ΔD(x, y, z, t) is the incremental displacement of the tissue at the position (x, y, z) from the initial time to time t, and T(x, y, z, t) is the current tissue position mapping at the spatial position (x, y, z) at time t.

6. The intelligent puncture positioning system for neuroendoscopic surgery based on image navigation according to claim 1 is characterized in that: The path optimization module constructs a multi-objective cost function for path planning. The form of the multi-objective cost function is: Among them, d i is the path segment length; r i is the organizational risk coefficient; δ i is the local tensor perturbation value; w d 、w r and w t is an adjustable weight coefficient.

7. The intelligent puncture positioning system for neuroendoscopic surgery based on image navigation according to claim 1 is characterized in that: The strategy guidance module provides feedback information and generates corresponding path adjustment suggestions through the following steps, specifically including: Construct a state space model that includes the current position of the path, the current tensor perturbation, and risk information; Use reinforcement learning models to evaluate the current state and generate path adjustment suggestions; Based on the generated path adjustment suggestions, feedback information is provided to the path optimization module to adjust the current path decision; Based on the feedback information, the path planning is adjusted to optimize the final decision of the puncture path.

8. The intelligent puncture positioning system for neuroendoscopic surgery based on image navigation according to claim 7 is characterized in that: The reinforcement learning model in the strategy guidance module is trained by setting a reward and punishment function, and the form of the reward and punishment function is: Among them, R t The reward value adjusted for the current path; D t is the path length; is the risk value of the path; is the error of the path reaching the target area; α, β and γ are weight coefficients.

9. The intelligent puncture positioning system for neuroendoscopic surgery based on image navigation according to claim 1, characterized in that: The feedback control module triggers the path update operation based on one of the following conditions: The total amount of current path disturbance exceeds the set threshold; The rising rate of the path cost function exceeds the set slope; The strategy guidance module outputs a suggestion that deviates from the current path direction by more than a preset tolerance.

10. The intelligent puncture positioning system for neuroendoscopic surgery based on image navigation according to claim 1, characterized in that: The navigation display module is used to display the current puncture path, surrounding risk tissue and tensor disturbance distribution in a three-dimensional superposition manner, and provide graphic annotation prompts for path adjustment suggestions.

Images