Intelligent planning system of electrode array, electronic device and storage medium

Abstract

Embodiments of the present application relate to the field of medical equipment, and disclose an intelligent planning system of electrode array, an electronic device and a storage medium.The system comprises: a probabilistic ablation effect prediction module which utilizes a pre-trained Gaussian process regression model describing the probability of any point in a target region being ablated, takes maximizing the expected ablation coverage volume of the target region and minimizing the uncertainty of the prediction result as the target, and predicts an initial electrode array configuration; a model self-adaptation module which acquires actual pose information of a current electrode during electrode placement into the target region based on a current electrode array configuration; and a constraint-aware dynamic optimization module which determines target pose information of a next electrode in an effective position domain satisfying a preset geometric constraint based on the current electrode array configuration and the actual pose information of the current electrode placed into the target region, so as to place the next electrode into the target region based on the target pose information. The problem of dynamically planning an electrode array is solved.

Description

TECHNICAL FIELD

[0001] Embodiments of the present application relate to the field of medical devices, and in particular to an intelligent planning system for electrode array, an electronic device and a storage medium. BACKGROUND

[0002] Irreversible electroporation (IRE) in steep pulse ablation is an emerging tumor ablation technology, which is often used to ablate solid tumors with complex location and irregular shape. This technology highly depends on the precise cooperative deployment of multiple electrodes, and through the application of high-intensity pulsed electric field in the tumor area, the cell membrane produces irreversible electroporation, thereby realizing non-thermal ablation of tumor cells. Irreversible electroporation technology requires multiple electrodes to work cooperatively to form an effective combined electric field. The physical principle of this technology determines that it has strict geometric constraints on the spatial configuration of the electrode array, which specifically embodies as follows: 1) all electrodes in the array need to maintain a high degree of mutual parallelism; 2) the Euclidean distance between any adjacent electrodes needs to be controlled within a preset effective interval. Any deviation from the above geometric constraints may lead to uneven distribution of combined electric field intensity or the occurrence of field strength "blind area", thereby affecting the thoroughness of subsequent treatment.

[0003] The existing IRE electrode array planning technology has the following defects:

[0004] (1) Limitation and non-adaptability of the model: the existing method is based on an idealized homogeneous tissue model for planning, which cannot adapt to individual differences and tissue heterogeneity of patients, resulting in a large deviation between actual ablation effect and preoperative planning.

[0005] (2) Accumulation and amplification of errors: electrodes are placed one by one, and the positioning error of the early electrodes will affect the optimal position selection of the subsequent electrodes, causing error accumulation and amplification, and finally leading to the geometric accuracy of the entire array not meeting the requirements.

[0006] (3) Lack of intraoperative adaptive ability: the existing system is an open-loop control, which cannot adjust the planning strategy according to the real-time feedback during the operation, and is difficult to cope with uncertain factors such as tissue deformation and electrode deviation.

[0007] (4) Low efficiency of constraint processing: the processing of distance constraints and parallelism constraints between electrodes depends on the experience judgment of the operator, and lacks efficient mathematical optimization methods. SUMMARY

[0008] The purpose of the present application is to at least provide an intelligent planning system for electrode array, an electronic device and a storage medium, which can at least solve the problem of dynamic planning of electrode array, and can at least improve the consistency between the ablation effect brought by the electrode array planning scheme and the expected ablation effect.

[0009] To solve the above technical problems, at least one embodiment of the present application provides an intelligent planning system of an electrode array, comprising: a probabilistic ablation effect prediction module configured to predict an initial electrode array configuration by using a pre-trained probabilistic ablation effect prediction model, with the goal of maximizing the expected ablation coverage volume of a target region and minimizing the uncertainty of the prediction result, wherein the probabilistic ablation effect prediction model comprises a Gaussian process regression model describing the probability of any point in the target region being ablated; a model self-adaption module configured to acquire actual pose information of a current electrode during the process of placing the electrode into the target region based on a current electrode array configuration; and a constraint-aware dynamic optimization module configured to determine target pose information of a next electrode in an effective position domain satisfying a preset geometric constraint based on the current electrode array configuration and the actual pose information of the current electrode placed into the target region, so as to place the next electrode into the target region based on the target pose information, wherein the preset geometric constraint comprises a parallel constraint and a distance constraint, the parallel constraint comprises that all electrodes in the electrode array are consistent with a reference reference direction, and the distance constraint comprises that the Euclidean distance value between the next electrode and at least one placed electrode is in a preset effective interval.

[0010] At least one embodiment of the present application also provides an electronic device, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform at least the following: predicting an initial electrode array configuration by using a pre-trained probabilistic ablation effect prediction model, with the goal of maximizing the expected ablation coverage volume of a target region and minimizing the uncertainty of the prediction result, wherein the probabilistic ablation effect prediction model comprises a Gaussian process regression model describing the probability of any point in the target region being ablated; acquiring actual pose information of a current electrode during the process of placing the electrode into the target region based on a current electrode array configuration; determining target pose information of a next electrode in an effective position domain satisfying a preset geometric constraint based on the current electrode array configuration and the actual pose information of the current electrode placed into the target region, so as to place the next electrode into the target region based on the target pose information; and the preset geometric constraint comprises a parallel constraint and a distance constraint, the parallel constraint comprises that all electrodes in the electrode array are consistent with a reference reference direction, and the distance constraint comprises that the Euclidean distance value between the next electrode and at least one placed electrode is in a preset effective interval.

[0011] At least one embodiment of this application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements: using a pre-trained probabilistic ablation effect prediction model, with the goal of maximizing the expected ablation coverage volume of the target area and minimizing the uncertainty of the prediction result, predicting an initial electrode array configuration, wherein the probabilistic ablation effect prediction model includes a Gaussian process regression model describing the probability of ablation at any point within the target area; during the process of placing electrodes into the target area based on the current electrode array configuration, acquiring the actual pose information of the current electrodes; based on the current electrode array configuration and the actual pose information of the current electrodes placed in the target area, determining the target pose information of the next electrode within an effective position domain that satisfies preset geometric constraints, and placing the next electrode into the target area based on the target pose information; wherein the preset geometric constraints include parallel constraints and distance constraints, the parallel constraints include that all electrodes in the electrode array are aligned with a reference reference direction, and the distance constraints include that the Euclidean distance between the next electrode and at least one already placed electrode is within a preset effective range.

[0012] The intelligent planning system, electronic device, and storage medium for electrode arrays provided in the embodiments of this application establish a probabilistic ablation effect prediction model through Gaussian process regression. This model can describe the combined electric field effect and quantify the uncertainty of the prediction results, enabling accurate prediction of the ablation effect of any electrode array. During the sequential placement of electrodes, based on the current electrode array configuration and the actual pose information of the currently placed electrodes, the target pose information of the next electrode is determined within the effective position domain that satisfies preset geometric constraints. This effectively compensates for the impact of the positioning error of the previous electrodes on the optimal position selection of subsequent electrodes, actively suppresses error accumulation in sequential operations, and effectively avoids the amplification of accumulated errors that ultimately lead to the geometric accuracy of the entire electrode array failing to meet requirements.

[0013] In some optional embodiments, the mean function of the Gaussian process regression model is a physical prior-based mean function constructed based on the cumulative probability model of the electric field effect, at any point within the target region. r The total probability of ablation arises from the sum of the contributions of each electrode; the contribution of a single electrode is proportional to its voltage and varies with any point. r When the distance increases and the value decreases, the expression for the mean function based on physical priors is as follows:

[0014]

[0015] In the formula, and They represent the first i The voltage and position of the root electrode. Indicates any point within the target area r The total probability of not being ablated, where N represents the number of electrodes in the electrode array. Represents any point within the target region under array configuration X. r The average probability of being ablated by each electrode.

[0016] In some optional embodiments, the kernel function of the Gaussian process regression model is a kernel function that includes an anisotropic component and an inter-field suppression component. The anisotropic component is used to describe the physical properties of the electric field extending further along the electrode axis, and the inter-field suppression component is used to describe the electric field extending further along the electrode axis based on any point within the target region. r The field strength is modulated by the distance to each electrode.

[0017] In some optional embodiments, the kernel function of the Gaussian process regression model is expressed as follows:

[0018]

[0019] In the formula, Represents the kernel function. r Indicates any point within the target area. Indicates the target area r Any point other than that, l Indicates the basic length scale. This represents the anisotropic distance metric matrix under array configuration X. Indicates the strength of the repulsion effect. express r The variance of the Euclidean distance to each electrode in array configuration X. express The variance of the Euclidean distance to each electrode in array configuration X.

[0020] In some optional embodiments, a pre-trained probabilistic ablation effect prediction model is used to predict the initial electrode array configuration using a first objective function, the expression of which is as follows:

[0021]

[0022] In the formula, This indicates the initial electrode array configuration. Used to maximize the target area The expected ablation coverage volume, Used to minimize the overall uncertainty of the prediction results, Represents any point within the target region under array configuration X. r The average probability of being ablated by each electrode. express r The variance of the point prediction results λ This represents a non-negative weight hyperparameter.

[0023] In some optional embodiments, the model adaptation module is further configured to condition the probabilistic ablation effect prediction model based on the actual pose information of the current electrode, so as to update the probabilistic ablation effect prediction model and the predicted electrode array configuration.

[0024] In some optional embodiments, the effective position domain satisfying the preset geometric constraint is a position domain determined by a constraint manifold calculated based on parallel constraints and distance constraints, the constraint manifold is a Cartesian product of a reference direction and the preset effective interval, and the reference direction is a direction calculated according to actual pose information of all electrodes implanted in the target region.

[0025] In some optional embodiments, based on the current electrode array configuration and the actual pose information of the current electrode implanted in the target region, a second objective function is used to determine the target pose information of the next electrode in the effective position domain satisfying the preset geometric constraint, so that the expected ablation volume gain is maximized and the uncertainty reduction of the prediction result is minimized, and the expression of the second objective function is as follows:

[0026]

[0027] In the formula, f represents the second objective function, represents the second objective function, represents the expected ablation volume gain, represents a non-negative weight coefficient, represents the uncertainty reduction of the prediction result, represents the effective position domain, represents any candidate position in the effective position domain.

[0028] In some optional embodiments, the model adaptation module is further configured to, after implanting at least one electrode in the target region based on the current electrode array configuration, acquire physical data of the patient in the ablation process, and correct the hyperparameters of the probabilistic ablation effect prediction model according to a difference between the physical data of the patient and the physical data predicted based on the hyperparameters of the current probabilistic ablation effect prediction model. BRIEF DESCRIPTION OF DRAWINGS

[0029] One or more embodiments are exemplified by the pictures in the drawings corresponding thereto, which do not constitute a limitation on the embodiments.

[0030] Figure 1 FIG. 1 is a structural schematic diagram of an intelligent planning system for an electrode array provided by an embodiment of the present application;

[0031] Figure 2 FIG. 2 is a structural example of an intelligent planning system for an electrode array provided by an embodiment of the present application;

[0032] Figure 3 Figure 1 is a flow chart of a method for intelligent planning of an electrode array according to an embodiment of the present application. DETAILED DESCRIPTION

[0033] To make the objectives, technical solutions, and advantages of the embodiments of the present application clearer, the embodiments of the present application will be described in detail below with reference to the drawings. However, it can be understood by those skilled in the art that, in the embodiments of the present application, many technical details are presented in order to make the readers better understand the present application. However, the technical solutions claimed by the present application can be implemented even without these technical details and various changes and modifications based on the following embodiments. The division of the following embodiments is for the convenience of description, and should not constitute any limitation on the specific implementation of the present application, and the embodiments can be combined and referenced with each other on the premise of not contradicting.

[0034] To facilitate understanding of the embodiments of the present application, the related content about the electrode array is introduced first.

[0035] Irreversible electroporation (IRE) is an emerging tumor ablation technology, which can achieve non-thermal ablation of tumor cells by applying a high-intensity pulsed electric field to the tumor area to cause irreversible electroporation of the cell membrane. The irreversible electroporation technology requires multiple electrodes to work cooperatively to form an effective combined electric field. The physical principle of the technology determines that it has strict geometric constraints on the spatial configuration of the electrode array, which are specifically embodied as: 1) all electrodes in the array need to maintain a high degree of mutual parallelism; 2) the Euclidean distance between any adjacent electrodes needs to be controlled within a preset effective interval. Any deviation from the above geometric constraints can lead to uneven distribution of the combined electric field intensity or the occurrence of a field strength "blind area", which in turn affects the thoroughness of the subsequent treatment.

[0036] The existing IRE electrode array planning technology has the following defects:

[0037] (1) Limitation and non-adaptability of the model: the existing method is based on an idealized homogeneous tissue model for planning, which cannot adapt to individual differences and tissue heterogeneity of patients, resulting in a large deviation between the actual ablation effect and the preoperative planning.

[0038] (2) Accumulation and amplification of errors: the electrodes are placed one by one, and the positioning errors of the previous electrodes will affect the optimal position selection of the subsequent electrodes, causing error accumulation and amplification, and finally resulting in the geometric accuracy of the entire array not meeting the requirements.

[0039] (3) Lack of intraoperative adaptive ability: the existing system belongs to open-loop control and cannot adjust the planning strategy according to the real-time feedback during the operation, making it difficult to cope with uncertain factors such as tissue deformation and electrode deviation.

[0040] (4) The processing efficiency of the distance constraint and the parallel constraint is low, and the processing depends on the experience of an operator, and lacks efficient mathematical optimization methods.

[0041] To solve the above technical problems, the present application provides an intelligent planning system for an electrode array, and the implementation details of the intelligent planning system for the electrode array of the present embodiment are described below. The following content is only provided for the implementation details for easy understanding, and is not necessary for implementing the present solution.

[0042] Embodiment One:

[0043] The intelligent planning system for the electrode array of the present embodiment can be applied to an electronic device with communication, calculation and data storage capabilities, and is used for intelligent planning of an electrode array for irreversible electroporation. The specific structure of the intelligent planning system for the electrode array can be as shown in Figure 1 The intelligent planning system for the electrode array of the present embodiment can be applied to an electronic device with communication, calculation and data storage capabilities, and is used for intelligent planning of an electrode array for irreversible electroporation. The specific structure of the intelligent planning system for the electrode array can be as shown in

[0044] The probabilistic ablation effect prediction module 101 is used to predict an initial electrode array configuration by using a pre-trained probabilistic ablation effect prediction model, with the target of maximizing the expected ablation coverage volume of the target region and minimizing the uncertainty of the prediction result. Specifically, the probabilistic ablation effect prediction model includes a Gaussian process regression model describing the probability of any point in the target region being ablated.

[0045] The model adaptive module 102 is used to obtain actual pose information of the current electrode during the process of placing the electrode into the target region based on the current electrode array configuration.

[0046] The constraint-aware dynamic optimization module 103 is used to determine target pose information of the next electrode in an effective position domain satisfying a preset geometric constraint based on the current electrode array configuration and the actual pose information of the current electrode placed in the target region, and place the next electrode into the target region based on the target pose information. Specifically, the preset geometric constraint includes a parallel constraint and a distance constraint. The parallel constraint includes that all electrodes in the electrode array are consistent with a reference reference direction, and the distance constraint includes that the Euclidean distance value between the next electrode and at least one placed electrode is in a preset effective interval.

[0047] In this embodiment, a probabilistic ablation effect prediction model is established through Gaussian process regression. This model can describe the combined electric field effect and quantify the uncertainty of the prediction results, enabling accurate prediction of the ablation effect of any electrode array. During the sequential placement of electrodes, the target pose information of the next electrode is determined within the effective position domain that satisfies preset geometric constraints, based on the current electrode array configuration and the actual pose information of the currently placed electrode. This effectively compensates for the impact of the positioning error of the previous electrodes on the optimal position selection of subsequent electrodes, actively suppresses error accumulation in sequential operations, and effectively avoids the amplification of accumulated errors that ultimately lead to the geometric accuracy of the entire electrode array failing to meet requirements.

[0048] In practice, the target area is obtained through real-time CT (Computed Tomography) or MRI (Magnetic Resonance Imaging) images of the patient provided by a medical imaging system. A probabilistic ablation effect prediction model based on Gaussian Process Regression (GPR), denoted as GP-Field, is constructed. This model determines the target area at any point within the target region. r Probability of being (successfully) dissolved Model it as a Gaussian regression process:

[0049]

[0050] in, This represents an array configuration consisting of N electrodes. For the first i The position vector of the root electrode, For the first i The direction vector of the root electrode; For any point within the target region under array configuration X r The mean probability of ablation represents the most likely ablation outcome (expected ablation effect). Let covariance function (or kernel function) have its diagonal elements represent r The variance of the predicted results for a point is a quantitative measure of uncertainty. Within the target area r Any point outside of that.

[0051] To overcome the limitations of traditional Gaussian process regression models when directly applied to complex physical fields, such as insufficient prediction accuracy and poor generalization ability due to inconsistencies between their general assumptions and specific physical laws, this embodiment improves upon the key components of the model—the mean function and the kernel function. To enhance model accuracy and data efficiency, a mean function based on physical priors is introduced. Instead of employing the traditional zero-mean assumption, this mean function can be constructed based on the principle of linear superposition of electric fields, thereby quickly calculating the basic distribution of the combined electric field.

[0052] In some examples, the mean function of the Gaussian process regression model described above is a physical prior-based mean function constructed from the cumulative probability model of the electric field effect, assuming any point within the target area... r The total probability of ablation arises from the sum of the contributions of each electrode; the contribution of a single electrode is proportional to its voltage and varies with any point. r The value decreases as the distance increases. The expression for the mean function based on physical priors is as follows:

[0053]

[0054] In the formula, and They represent the first i The voltage and position of the root electrode. Indicates any point within the target area r The total probability of not being ablated, where N represents the number of electrodes in the electrode array. Represents any point within the target region under array configuration X. r The average probability of being ablated by each electrode.

[0055] By designing the mean function as described above, the core task of the Gaussian process regression model is transformed from "learning the entire complex electric field from scratch" to "learning the complex residuals between the real electric field and the physical prior" (such as tissue heterogeneity, anisotropy, inter-field inhibition effect, and individual patient differences), which greatly reduces the need for training data and improves the convergence speed and robustness of the model.

[0056] Kernel function The model defines the understanding of the spatial correlation of the ablation field, encoding professional knowledge, specifically parameterized by a set of hyperparameters θ. To accurately capture the complex physical characteristics of the electric field in multi-electrode IREs, in some embodiments, a novel kernel function—the Anisotropic-Repulsive Kernel (ARK)—is proposed instead of the general kernel function. Specifically, the kernel function of the Gaussian process regression model is a kernel function that includes an anisotropic component and an inter-field suppression component. The anisotropic component is used to describe the physical characteristics of the electric field extending further along the electrode axis, while the inter-field suppression component is used to describe the electric field based on any point within the target region. r The field strength is modulated by the distance to each electrode.

[0057] In some examples, the kernel function of a Gaussian process regression model is expressed as follows:

[0058]

[0059] where, denotes the kernel function, r denotes any point in the target region, denotes any point in the target region r except the origin, l denotes the base length scale, denotes the anisotropic distance metric matrix under array configuration X, denotes the repulsion strength, denotes r the variance of the Euclidean distance to each electrode in array configuration X, denotes the variance of the Euclidean distance to each electrode in array configuration X.

[0060] Anisotropic component , through an anisotropic distance metric matrix related to the electrode configuration , precisely describes the physical property that the electric field extends further along the electrode axis.

[0061] Construction: built through a dynamically calculated rotation matrix R and a fixed diagonal scaling matrix S, in the form of: ; where the S matrix defines the scaling ratio along the principal axis, in three-dimensional space, in the form of , and are the scaling coefficients in the vertical and parallel directions, respectively. The matrix R rotates the standard coordinate system to align with the unified reference direction of the current electrode array, so that can be dynamically adjusted according to the actual electrode orientation X.

[0062] : provides a global, isotropic reference scale for distance measurement, and the matrix further introduces local, anisotropic directional correction on this reference.

[0063] Field suppression component , which modulates the field strength using the geometric relationship between the spatial point r in the target region and the distance to each electrode. Where:

[0064] : refers to the variance of the Euclidean distance from the spatial point r to each electrode in the electrode array, which is a geometric quantity that changes with the position r , and is not a hyperparameter. This design achieves field strength modulation through the following mechanism: at the geometric center of the electrode array, rClose to each electrode, Approaching zero, the component value is close to 1, thus suppressing the model response; conversely, near a single electrode, r Far from each electrode, Large value, the component value is much greater than 1, thus enhancing the model response.

[0065] η: This hyperparameter is used to adjust the intensity of the suppression and enhancement effects.

[0066] The key members in the above hyperparameter set θ include: the basic length scale , the strength of the repulsion effect η, and the relevant parameters of the distance metric matrix . The initial values of these hyperparameters can be determined by learning from the offline generated simulation data set.

[0067] After establishing the above Gaussian process regression model, in order to ensure that the model has good generalization ability in the entire high-dimensional parameter space (composed of electrode number, position, direction, voltage, etc.) and reduce the dependence on massive training data, a simulation parameter design method based on Latin Hypercube Sampling (LHS) can be used to generate an offline training data set. Compared with traditional grid sampling, LHS can achieve more uniform exploration of the parameter space with fewer sample points, thus significantly improving the efficiency and final performance of model training.

[0068] In some embodiments, after the construction and training of the above high-fidelity probabilistic electric field model is completed, the pre-trained probabilistic ablation effect prediction model is used to predict the initial electrode array configuration using a first objective function to solve a theoretically optimal initial electrode array configuration , the expression of the first objective function is as follows:

[0069]

[0070] In the formula, represents the initial electrode array configuration, is used to maximize the expected ablation coverage volume of the target region, is used to minimize the overall uncertainty of the prediction results, represents the average probability of any point in the target region being ablated by each electrode under the array configuration X, represents the variance of the prediction results of the point, r represents the non-negative weight hyperparameter, which is pre-set and used to balance the expected return and uncertainty risk. The mean and variance r in the first objective function, λ ​​​All of them are from the aforementioned GPR model which deeply integrates physical priori and professional knowledge, thus ensuring the high quality and reliability of the initial planning scheme.

[0071] In view of the complexity of the optimization problem of the initial electrode array configuration, the initial planning based on the GPR model is realized preoperatively, and the calculation process can be completed within a few minutes, which fully meets the preparation requirements of the clinic for the high-quality initial deployment scheme.

[0072] In some embodiments, the model adaptation module 102 is further configured to conditionally process the probabilistic ablation effect prediction model based on the actual pose information of the current electrode, so as to update the probabilistic ablation effect prediction model and the predicted electrode array configuration. The actual pose information and the conditional processing are the bridge connecting the physical operation and the digital planning, and are the core link for realizing the closed-loop feedback control. The function is not to change the model itself, but to update the real physical operation result as the established fact for subsequent planning.

[0073] In some examples, the real-time pose measurement is implemented as follows:

[0074] When the operator places the first k root electrode into the target area under the guidance of medical images and preliminarily fixes it, the surgical navigation system (for example, an optical tracking system or an electromagnetic tracking system) integrated with the system will measure and output the accurate pose of the electrode in the unified coordinate system of the surgical space in real time. The pose information usually includes:

[0075] Position vector : describes the accurate coordinates of the electrode tip or a specific marker point in three-dimensional space.

[0076] Direction vector : describes the accurate orientation of the electrode shaft (for example, the direction of the tip pointing to the marker point).

[0077] The measured pose information is immediately transmitted to the system.

[0078] Further, the model is updated after the condition processing: after receiving the actual pose of the first k root electrode, the planning state is immediately updated, and this process is the conditional processing. The specific processing method is as follows: in the probabilistic ablation effect prediction model, the electrode array configuration X is originally a variable set to be solved which contains multiple uncertain members, and the core of the conditional processing is to use the accurately measured actual pose k of the first , The variable θ is replaced, transforming it from an uncertain variable into a fixed constant. After this update, the probabilistic ablation effect prediction model itself (i.e., its hyperparameter set θ) remains unchanged, but the scenario it evaluates has changed. All subsequent planning and calculations will be based on this new configuration where some members are now fixed. In this way, the system seamlessly integrates the actual operational results of the physical world into the digital planning loop, providing a definite and more realistic decision-making starting point for dynamic compensation planning, effectively suppressing the cumulative effect caused by previous operational errors, thereby reducing the uncertainty of the overall planning scheme.

[0079] In some embodiments, the effective position domain that satisfies the preset geometric constraints is the position domain determined by the constraint manifold calculated based on parallel constraints and distance constraints. The constraint manifold is the Cartesian product of the reference direction and the preset effective interval. The reference direction is the direction calculated based on the actual pose information of all electrodes placed in the target area.

[0080] In practical implementation, after receiving an updated posterior model and electrode array configuration that more closely reflects intraoperative reality, the next electrode (the...) is... k For each electrode to be implanted (+1), an optimal compensation pose is calculated efficiently in real-time (within seconds). This optimal pose aims to maximize a weighted objective function composed of the expected ablation volume gain and the reduction in model uncertainty, thereby actively suppressing the cumulative effect caused by previous operational errors while pursuing therapeutic efficacy. This embodiment abandons the inefficient "generate-verify" mode of traditional optimization algorithms in a general space. Instead, it combines dimensionality reduction through parallel constraints with geometric constraint embedding through distance constraints. A low-dimensional search space (i.e., a constraint manifold) with embedded geometric constraints is constructed using mathematical methods. Subsequent optimization will only be performed on this constraint manifold, thereby fundamentally reducing the computational complexity of the problem while ensuring that the constraints are met.

[0081] In some examples, the specific implementation for determining the valid location field is as follows:

[0082] (1) Dimensionality reduction of parallel constraints is implemented as follows:

[0083] According to confirmed k The set of actual direction vectors of the root electrode { A unified reference direction is calculated. ,For example, The average value of all direction vectors can be taken and normalized. Parallel-constrained manifolds Strictly defined as the direction vector of all members of the electrode array u Equal identity equals The pose set. Through this definition, the direction vector in the pose information of the next electrode... It has been forcibly constrained as The next electrode to be solved The pose search dimension is reduced from the general six degrees of freedom (position x, y, z + pose α, β, γ) to a position space with only three degrees of freedom (x, y, z), which greatly simplifies the problem.

[0084] (2) Geometric embedding of distance constraints, implemented as follows:

[0085] Building upon the aforementioned dimensionality reduction, the distance constraint is further geometrically refined. For the position vector of the next electrode to be optimized... It must satisfy the condition that it is compatible with the set of electrodes already in place. At least one adjacent electrode Euclidean distance Within the preset effective range This constraint, in three-dimensional position space, defines a region consisting of multiple elements. Centered on, with inner and outer radii respectively and The effective search domain is formed by the intersection or merging of spherical shells. Any point outside this domain need not be considered.

[0086] (3) Definition of final optimization space: The final optimization search space is precisely defined as a constrained manifold M. This manifold is the effective location domain that satisfies the distance constraints. (That is, the distance to the adjacent electrode is within the preset effective range) (the spatial region within) and a single fixed reference direction that satisfies parallel constraints. Cartesian product:

[0087]

[0088] Subsequent optimizations will be limited to searching on this manifold M. By determining the effective location domain, the geometric constraints such as the parallelism and spacing of the electrode array are transformed into a mathematical expression that facilitates efficient computation and optimization, thereby improving computational efficiency.

[0089] Furthermore, on the constructed low-dimensional constrained manifold M that naturally satisfies the constraints, an optimization algorithm is used to efficiently solve the problem in order to find the optimal position of the next electrode (the position vector in the target pose). The second objective function of the optimization algorithm The core optimization criterion remains consistent with the initial plan, namely balancing expected returns with uncertainty risk. However, its specific form has been adjusted to evaluate the incremental contribution of a single operation, namely: maximizing the expected ablation volume gain while minimizing the reduction in the uncertainty of the predicted result. Furthermore, the model upon which this objective function relies is an updated posterior model to ensure that decisions are based on the latest intraoperative realities. Ultimately, the objective function of the optimization algorithm is defined as the weighted sum of the two core metrics: expected ablation volume gain and the overall reduction in model uncertainty.

[0090] In some examples, based on the current electrode array configuration and the actual pose information of the current electrode placed in the target area, a second objective function is used to determine the target pose information of the next electrode within the effective position domain that satisfies preset geometric constraints. This ensures that the expected ablation volume gain is maximized while minimizing the reduction in uncertainty of the prediction result. The expression for the second objective function is as follows:

[0091]

[0092] In the formula, This represents the second objective function. This indicates the expected ablation volume gain. Indicates non-negative weight coefficients. This represents the reduction in uncertainty of the forecast result. Indicates the valid location field. This represents any candidate position in the valid position field.

[0093] Expected ablation volume gain This can be obtained by comparing the model prediction results before and after the new electrode is placed. Specifically, the values ​​are: assuming at the candidate location... p The new expected total ablation volume achievable after implanting the new electrode, minus the current expected total ablation volume achievable based on the currently implanted electrode. This represents the overall reduction in uncertainty of the model prediction results. Similarly, this is obtained by comparing the model states before and after the addition of the new electrode. The specific value is: the overall uncertainty of the current model prediction based on the current electrode array, minus the assumption that the new electrode is at the candidate position. p The overall uncertainty of the new model after adding the new electrode. A preset or user-adjustable non-negative weighting coefficient used to balance the two optimization objectives.

[0094] In some examples, in order to be in the effective location domain Find the internal efficiency The solution that maximizes the value can be obtained using optimization algorithms, including but not limited to the following forms:

[0095] (1) Algorithm based on projected gradient ascent: From Starting from an initial point within the objective function, along... The gradient direction is used for iterative search. After each iteration, if the candidate point exceeds the range... The boundary is then pulled back to the nearest point in the domain by a projection operator, ensuring that the entire search trajectory always lies on the effective manifold.

[0096] (2) Algorithm based on adaptive sampling: In The system employs intelligent random sampling. Initially, it performs global uniform sampling to explore the entire space. Subsequently, based on the objective function values ​​of the sampled points, it performs denser local sampling in the "desired region" where the performance is better, thereby quickly converging to the vicinity of the optimal solution.

[0097] In some embodiments, the model adaptation module 102 is further configured to: after placing at least one electrode into the target area based on the current electrode array configuration, acquire the patient's physical data during the ablation process, and perform hyperparameter correction of the probabilistic ablation effect prediction model based on the difference between the patient's physical data and the physical data predicted by the hyperparameters of the current probabilistic ablation effect prediction model. Through a closed-loop feedback mechanism, the model is adaptively corrected online using physical data that can be acquired in real time during the procedure, so that it can match the biophysical characteristics of a specific patient.

[0098] By collecting real-time patient-specific physical data during surgery, the underlying parameters of the established model are fine-tuned online, thereby addressing individual differences among patients and enabling the planning to evolve from general to personalized, further improving planning accuracy. The underlying parameters here include the basic length scale. The intensity of the repulsion effect η and the distance metric matrix Relevant parameters Hyperparameters, physical data including impedance.

[0099] In some cases, the specific method for obtaining the patient's physical data during ablation is as follows: after one or more new electrodes are placed, a low-energy test pulse is applied to the electrode by controlling the pulse generator of the IRE ablation device, and data reflecting the current tissue biophysical properties, such as tissue impedance, are measured in real time by built-in sensors. This data is directly related to tissue conductivity.

[0100] In some examples, according to the difference between the physical data of the patient and the physical data predicted based on the hyperparameters of the current probabilistic ablation effect prediction model, the hyperparameters of the probabilistic ablation effect prediction model are corrected, and the model parameter adaptive correction is realized. The specific means is: comparing the measured data with the predicted value of the model based on the current hyperparameters θ (such as the basic length scale of the kernel function). The difference between the two will be used as an error signal. By standard Bayesian updating method, the hyperparameters θ are fine-tuned, and a posteriori hyperparameters more suitable for the real tissue condition of the current patient can be obtained .

[0101] The hyperparameters θ (l, η, , ) learned by the probabilistic ablation effect prediction model are not only mathematical fitting parameters, but also quantitative descriptions of the specific tissue biophysical properties (such as conductivity, dielectric constant, etc.) of the current patient. Since the tissue impedance is the direct electrical manifestation of these biophysical properties under a specific electrode configuration, in the case of obtaining physical data as tissue impedance, the intelligent planning system of the electrode array can also include an impedance prediction module to realize the prediction of tissue impedance by constructing a personalized conductivity field, solving the current density and calculating the total impedance. Among them, constructing a personalized conductivity field: using the learned hyperparameters θ, a personalized conductivity distribution map reflecting tissue heterogeneity and anisotropy is constructed in the entire target area; solving the current density: based on the conductivity field described by the conductivity distribution map, the current electrode (geometry) configuration and the voltage setting, the corresponding physical equation (such as Laplace equation) is solved to calculate the theoretical three-dimensional current density distribution; calculating the total impedance of the tissue: the total current I is obtained by integrating the current on the electrode surface, and the predicted total impedance of the tissue is finally calculated according to the macro form of Ohm's law (where V is the applied voltage).

[0102] Based on the system of the embodiment, the entire electrode array planning process is a closed loop of iteration: through the operator to place the electrode into the target area, accurately perceive the actual pose, realize the execution and perception of electrode placement, based on the updated information, calculate the optimal target pose of the next electrode, realize the dynamic re-planning based on the constraint manifold; real-time physical data can be selectively collected to further correct the model, realize the learning and self-adaptation of the model. This cycle will continue until the system finally updates the model prediction result, and the entire electrode array expected cumulative ablation volume has met the preoperative clinical target (for example, the ablation coverage rate of the target target area reaches the pre-set standard), at this time the entire planning process is completed, and the system outputs the final determined and complete electrode array deployment scheme.

[0103] The system of the embodiment defines the construction process of the electrode array as an iterative optimization cycle of "prediction-execution-perception-adaptation". The planning process can be divided into two core stages, solving the logical cycle of needing to predict the electrode array configuration to evaluate the effect and needing to determine the array configuration according to the expected effect. In the initial planning stage, the electrode array configuration X is an optimization variable to be solved, and the probabilistic ablation effect prediction model is used as an evaluation tool at this time to iteratively search for the theoretically optimal initial configuration X0 in a virtual environment. In the dynamic re-planning stage, the actual pose of the electrode that has been placed is converted into a known input condition, and only the target pose of the next electrode to be placed is solved on this basis, so that a complex global optimization problem is converted into a dynamic, step-by-step sequential decision problem.

[0104] Compared with the prior art, the present application has the following beneficial effects:

[0105] (1) Better real-time performance: The constraint manifold enables the originally complex path calculation to be completed within 1 second, so that there is no need to wait during the operation, and adjustment suggestions can be obtained immediately, and the ablation process of irreversible perforation is more smooth.

[0106] (2) More accurate and reliable: The model can predict the ablation effect, and the dynamic re-planning ensures that the planning scheme meets the safety requirements 100%. The combination of the two can make the ablation more thorough and reduce the risk of tumor recurrence.

[0107] (3) More intelligent and personalized: The system can adjust the planning scheme in real time according to the actual situation during the operation and the physical characteristics of the patient. Since the electrode array planning scheme is an adaptively adjusted scheme for each patient, the safety and effectiveness are higher.

[0108] In one example, the system architecture of the embodiment is as shown in Figure 2 The system architecture of the embodiment is as shown in

[0109] Further, in addition to the above-mentioned probabilistic ablation effect prediction module 101, model adaptation module 102 and constraint-aware dynamic optimization module 103, an intelligent decision support module can also be included, which outputs the electrode array planning and deployment scheme determined by the aforementioned modules, provides the decision-making suggestion of the electrode array configuration, and provides the results of risk assessment such as whether the target area is completely ablated according to the electrode array planning and deployment scheme, etc. Thus, the decision support level of the system is further improved, and the complex and experience-dependent electrode array planning is changed into a scientific decision-making process with data support, and the system directly provides optimal decision-making suggestions and risk assessment results, reducing the difficulty of irreversible perforation ablation.

[0110] It is worth mentioning that each module involved in the embodiment is a logical module. In actual application, a logical unit can be a physical unit, a part of a physical unit, or a combination of multiple physical units. In addition, in order to highlight the innovative part of the present application, units not closely related to solving the technical problems proposed in the present application are not introduced in the embodiment, but this does not mean that there are no other units in the embodiment.

[0111] Embodiment two:

[0112] The intelligent planning method of the electrode array of the embodiment can be applied to an electronic device with communication, calculation and data storage capabilities,

[0113] The specific process can be as shown in Figure 3 , including:

[0114] Step S1, probabilistic ablation field modeling and initial planning.

[0115] This step constructs a probabilistic ablation effect prediction model (probabilistic joint electric field model) as the core of decision-making and solves a theoretically optimal initial electrode array configuration. The model accurately predicts the ablation effect of any electrode array based on preoperative images of the patient and strictly quantifies its uncertainty. After the construction of the above-mentioned high-fidelity probabilistic joint electric field model, the optimization of the initial deployment scheme is performed. Due to the complexity of the optimization problem, this step is designed as a preoperative link, and its calculation process can be completed within a few minutes, which fully meets the preparation requirements of the clinic for high-quality initial deployment schemes.

[0116] Step S2, real-time pose capture and model conditioning.

[0117] This step is the bridge connecting physical operation and digital planning, and is the core link of closed-loop feedback control. Its function is not to change the model itself, but to update the real physical operation result as the established fact for subsequent planning. After this update, the probabilistic ablation effect prediction model itself (i.e. its hyperparameters θ) does not change, but the scenario it evaluates has changed. All subsequent planning and calculation will be based on this new configuration in which part of the members have been fixed. In this way, the system seamlessly integrates the actual operation results of the physical world into the cycle of digital planning, providing a determined and more realistic decision starting point for dynamic re-planning in step S3, effectively suppressing the cumulative effect caused by previous operation errors, thereby reducing the uncertainty of the overall planning scheme.

[0118] Step S3, dynamic re-planning based on constraint manifold.

[0119] This step is to calculate an optimal compensation pose for the next electrode to be placed in real time (within seconds) and efficiently after receiving the updated posterior probability model in step S2, which is closer to the intraoperative reality. This optimal pose aims to maximize the weighted objective function composed of the expected ablation volume gain and the reduction of model uncertainty, thereby pursuing ablation effect while actively suppressing the cumulative effect caused by previous operation errors. Unlike traditional optimization algorithms that use the inefficient "generate-verify" mode in general space, this step constructs and parameterizes the constraint manifold through mathematical methods, and then optimizes and solves it efficiently on the constraint manifold.

[0120] Step S4, online model adaptation and iterative closed loop.

[0121] This step is an advanced adaptive link for further improving planning accuracy. Its goal is to fine-tune the underlying parameters (at least part of the hyperparameters) of the probabilistic model established in step S1 by collecting real-time patient-specific physical data (including impedance) online, thereby solving the individual difference problem between patients and enabling the model to evolve from generalization to personalization.

[0122] The specific implementation of each step of the present embodiment can be referred to the foregoing embodiments, which will not be repeated here.

[0123] Embodiment Three:

[0124] Another embodiment of the present application relates to an electronic device, comprising: at least one processor; and a memory connected with the at least one processor in communication; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform at least: predicting an initial electrode array configuration by using a pre-trained probabilistic ablation effect prediction model, aiming to maximize an expected ablation coverage volume of a target region and minimize uncertainty of a prediction result, the probabilistic ablation effect prediction model comprising a Gaussian process regression model describing a probability of any point in the target region being ablated; obtaining actual pose information of a current electrode during electrode implantation into the target region based on a current electrode array configuration; determining target pose information of a next electrode in an effective position domain satisfying a preset geometric constraint based on the current electrode array configuration and the actual pose information of the current electrode implanted into the target region, to implant the next electrode into the target region based on the target pose information; and the preset geometric constraint comprises a parallel constraint and a distance constraint, the parallel constraint comprising that all electrodes in the electrode array are consistent with a reference reference direction, and the distance constraint comprising that a Euclidean distance value between the next electrode and at least one implanted electrode is in a preset effective interval.

[0125] The memory and the processor are connected in a bus manner, the bus can include any number of interconnected buses and bridges, and the bus connects various circuits of the one or more processors and the memory together. The bus can also connect various other circuits such as peripheral devices, voltage stabilizers, and power management circuits together, which are well known in the art, and thus, further description thereof will not be given herein. The bus interface provides an interface between the bus and the transceiver. The transceiver can be one element or multiple elements such as multiple receivers and transmitters, which provide units for communicating with various other devices on the transmission medium. The data processed by the processor is transmitted on the wireless medium through the antenna, and further, the antenna also receives data and transmits the data to the processor.

[0126] The processor is responsible for managing the bus and general processing, and can also provide various functions including timing, peripheral interface, voltage regulation, power management, and other control functions. And the memory can be used to store data used by the processor in performing operations.

[0127] Embodiment four:

[0128] Another embodiment of this application relates to a computer-readable storage medium storing a computer program. When executed by a processor, the computer program performs the following: using a pre-trained probabilistic ablation effect prediction model, with the goal of maximizing the expected ablation coverage volume of the target area and minimizing the uncertainty of the prediction result, it predicts an initial electrode array configuration. The probabilistic ablation effect prediction model includes a Gaussian process regression model describing the probability of ablation at any point within the target area. During the process of placing electrodes into the target area based on the current electrode array configuration, it acquires the actual pose information of the current electrodes. Based on the current electrode array configuration and the actual pose information of the current electrodes placed in the target area, it determines the target pose information of the next electrode within an effective position domain that satisfies preset geometric constraints, and places the next electrode into the target area based on the target pose information. The preset geometric constraints include parallel constraints and distance constraints. The parallel constraints include that all electrodes in the electrode array are aligned with a reference direction, and the distance constraints include that the Euclidean distance between the next electrode and at least one already placed electrode is within a preset effective range.

[0129] That is, those skilled in the art will understand that all or part of the steps in the methods of the above embodiments can be implemented by a program instructing related hardware. This program is stored in a storage medium and includes several instructions to cause a device (which may be a microcontroller, chip, etc.) or processor to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0130] Those skilled in the art will understand that the above embodiments are specific embodiments for implementing this application, and in practical applications, various changes can be made to them in form and detail without departing from the spirit and scope of this application.

Claims

1. An intelligent planning system for an electrode array, characterized in that, include: The probabilistic ablation effect prediction module is used to predict the initial electrode array configuration by using a pre-trained probabilistic ablation effect prediction model with the goal of maximizing the expected ablation coverage volume of the target area and minimizing the uncertainty of the prediction result. The probabilistic ablation effect prediction model includes a Gaussian process regression model that describes the probability of ablation at any point in the target area. The model adaptation module is used to obtain the actual pose information of the current electrode during the process of placing the electrode into the target region based on the current electrode array configuration; The constraint-aware dynamic optimization module is used to determine the target pose information of the next electrode within an effective position domain that satisfies preset geometric constraints, based on the current electrode array configuration and the actual pose information of the current electrode placed in the target region. The next electrode is then placed in the target region based on the target pose information. The preset geometric constraints include parallel constraints and distance constraints. The parallel constraints include that all electrodes in the electrode array are aligned with the reference reference direction. The distance constraints include that the Euclidean distance between the next electrode and at least one already placed electrode is within a preset effective range.

2. The intelligent planning system for electrode arrays according to claim 1, characterized in that, The mean function of the Gaussian process regression model is a physical prior-based mean function constructed from the cumulative probability model of the electric field effect at any point within the target area. r The total probability of ablation arises from the sum of the contributions of each electrode; the contribution of a single electrode is proportional to its voltage and varies with any point. r When the distance increases and the value decreases, the expression for the mean function based on physical priors is as follows: In the formula, and They represent the first i The voltage and position of the root electrode. Indicates any point within the target area r The total probability of not being ablated, where N represents the number of electrodes in the electrode array. Represents any point within the target region under array configuration X. r The average probability of being ablated by each electrode.

3. The intelligent planning system for electrode arrays according to claim 1, characterized in that, The kernel function of the Gaussian process regression model is a kernel function that includes an anisotropic component and an inter-field suppression component. The anisotropic component is used to describe the physical properties of the electric field extending further along the electrode axis, and the inter-field suppression component is used to describe the electric field extending further along the electrode axis based on any point within the target region. r The field strength is modulated by the distance to each electrode.

4. The intelligent planning system for the electrode array according to claim 3, characterized in that, The kernel function of the Gaussian process regression model is expressed as follows: In the formula, Represents the kernel function. r Indicates any point within the target area. Indicates the target area r Any point other than that, l Indicates the basic length scale. This represents the anisotropic distance metric matrix under array configuration X. Indicates the strength of the repulsion effect. express r The variance of the Euclidean distance to each electrode in array configuration X. express The variance of the Euclidean distance to each electrode in array configuration X.

5. The intelligent planning system for electrode arrays according to claim 1, characterized in that, Using a pre-trained probabilistic ablation effect prediction model, the initial electrode array configuration is predicted using a first objective function, the expression of which is as follows: In the formula, This indicates the initial electrode array configuration. Used to maximize the target area The expected ablation coverage volume, Used to minimize the overall uncertainty of the prediction results, Represents any point within the target region under array configuration X. r The average probability of being ablated by each electrode. express r The variance of the point prediction results λ This represents a non-negative weight hyperparameter.

6. The intelligent planning system for electrode arrays according to claim 1, characterized in that, The model adaptation module is further configured to: conditionally process the probabilistic ablation effect prediction model based on the actual pose information of the current electrode, so as to update the probabilistic ablation effect prediction model and its predicted electrode array configuration.

7. The intelligent planning system for electrode arrays according to claim 1, characterized in that, The effective position domain that satisfies the preset geometric constraints is the position domain determined by the constraint manifold calculated based on parallel constraints and distance constraints. The constraint manifold is the Cartesian product of the reference reference direction and the preset effective interval. The reference reference direction is the direction calculated based on the actual pose information of all electrodes that have been placed in the target area.

8. The intelligent planning system for the electrode array according to claim 1, characterized in that, Based on the current electrode array configuration and the actual pose information of the current electrode placed in the target region, the target pose information of the next electrode is determined within the effective position domain that satisfies the preset geometric constraints using a second objective function. This ensures that the expected ablation volume gain is maximized while minimizing the reduction in uncertainty of the prediction result. The expression for the second objective function is as follows: In the formula, This represents the second objective function. This indicates the expected ablation volume gain. Indicates non-negative weight coefficients. This represents the reduction in uncertainty of the forecast result. Indicates the valid location field. This represents any candidate position in the valid position field.

9. The intelligent planning system for electrode arrays according to claim 1, characterized in that, The model adaptation module is further configured to: after placing at least one electrode into the target area based on the current electrode array configuration, acquire the patient's physical data during the ablation process, and perform hyperparameter correction of the probabilistic ablation effect prediction model based on the difference between the patient's physical data and the physical data obtained by hyperparameter prediction based on the current probabilistic ablation effect prediction model.

10. An electronic device, characterized in that, include: At least one processor; as well as, A memory communicatively connected to the at least one processor; wherein, The memory stores instructions executable by the at least one processor, which are executed by the at least one processor to enable the at least one processor to perform at least the following: using a pre-trained probabilistic ablation effect prediction model, with the objective of maximizing the expected ablation coverage volume of the target area and minimizing the uncertainty of the prediction result, predicting the initial electrode array configuration, wherein the probabilistic ablation effect prediction model includes a Gaussian process regression model describing the probability of ablation at any point within the target area; and acquiring the actual pose information of the current electrode during the process of placing electrodes into the target area based on the current electrode array configuration. Based on the current electrode array configuration and the actual pose information of the current electrode placed in the target area, the target pose information of the next electrode is determined within the effective position domain that satisfies the preset geometric constraints, so as to place the next electrode into the target area based on the target pose information; the preset geometric constraints include parallel constraints and distance constraints, the parallel constraints include that all electrodes in the electrode array are aligned with the reference reference direction, and the distance constraints include that the Euclidean distance between the next electrode and at least one already placed electrode is within a preset effective range.

11. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the following: using a pre-trained probabilistic ablation effect prediction model, with the goal of maximizing the expected ablation coverage volume of the target area and minimizing the uncertainty of the prediction result, to predict the initial electrode array configuration. The probabilistic ablation effect prediction model includes a Gaussian process regression model describing the probability of ablation at any point in the target area; and during the process of placing electrodes into the target area based on the current electrode array configuration, it acquires the actual pose information of the current electrodes. Based on the current electrode array configuration and the actual pose information of the current electrode placed in the target area, the target pose information of the next electrode is determined within the effective position domain that satisfies the preset geometric constraints, so as to place the next electrode into the target area based on the target pose information; the preset geometric constraints include parallel constraints and distance constraints, the parallel constraints include that all electrodes in the electrode array are aligned with the reference reference direction, and the distance constraints include that the Euclidean distance between the next electrode and at least one already placed electrode is within a preset effective range.

Images