A system and method for optimizing spinal nerve stimulation parameters based on a surrogate model

By constructing a patient-specific three-dimensional finite element model and machine learning model of the spinal cord, and combining it with a multi-objective optimization algorithm, the spinal cord electrical stimulation parameters are automatically searched, which solves the problems of low efficiency and lack of personalization in existing technologies, and achieves efficient and safe personalized treatment.

CN122531622BActive Publication Date: 2026-11-17NANCHANG UNIV
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202610488953.X
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2026-04-14
Publication Date
2026-11-17
Estimated Expiration
2046-04-14

AI Technical Summary

Technical Problem

Existing methods for optimizing spinal cord stimulator parameters rely on human experience, which is inefficient, makes it difficult to achieve personalized treatment, and lacks a systematic approach to balancing multiple treatment goals, thus failing to effectively utilize individual patient anatomical data.

Method used

The surrogate model-based spinal cord stimulation parameter optimization system constructs a specific three-dimensional finite element model of the spinal cord of the target patient, uses a machine learning model to establish a rapid prediction mapping, and combines a multi-objective optimization algorithm to iteratively search for the optimal combination of stimulation parameters.

Benefits of technology

It enables personalized, automated, and efficient determination of optimal parameters for spinal cord nerve electrical stimulation, improving the accuracy and safety of treatment effects, shortening parameter tuning time, avoiding the trap of local optima, and providing a multi-objective balance scheme.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122531622B_ABST
    Figure CN122531622B_ABST
Patent Text Reader

Abstract

This invention discloses a system and method for optimizing spinal cord stimulation parameters based on a surrogate model. The system first constructs a specific three-dimensional finite element model of the spinal cord for the target patient based on spinal medical data; then, the stimulation parameters are encoded as input vectors, and biophysical simulation is performed using the specific spinal cord three-dimensional finite element model to obtain accurate individual effect evaluation; based on this, a surrogate model is trained as an efficient substitute for the simulator; finally, a multi-objective optimization algorithm is driven to iteratively search in the individual parameter space, ultimately outputting the optimal combination of stimulation parameters for the patient's anatomical characteristics. This invention combines computational biology modeling with artificial intelligence optimization, overcoming the shortcomings of traditional SCS parameter tuning that rely on experience and ignore individual differences, achieving a leap from a "general solution" to a "one-person-one-policy" precision medicine, significantly improving the effect, safety, and tuning efficiency of spinal cord stimulation.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of medical devices and artificial intelligence, and more specifically, to a system and method for optimizing spinal cord nerve stimulation parameters based on a surrogate model. Background Technology

[0002] Spinal cord stimulators, as active implantable medical devices, are important medical equipment used to treat neurological diseases such as traumatic brain injury, chronic pain, and limb spasticity. They work by delivering electrical stimulation through an array of electrodes implanted in the epidural space of the spinal cord to modulate the activity of spinal nerve circuits. The effectiveness of spinal cord stimulators is highly dependent on the settings of stimulation parameters, including but not limited to the voltage / current amplitude, pulse width, frequency, or stimulation location and their combinations of multiple electrodes.

[0003] Currently, the optimization of stimulation parameters in clinical practice mainly relies on manual testing and adjustment based on the experience of doctors or technicians, i.e., the "trial and error method." This method has the following significant drawbacks:

[0004] 1. The parameter space is huge, and manual traversal or search is time-consuming and inefficient, which prolongs the treatment cycle for patients.

[0005] 2. Relying on human experience can easily lead to local optima, making it impossible to maximize the treatment effect.

[0006] 3. It is highly subjective and lacks consistency. Different operators have different adjustment strategies, making it difficult to standardize treatment plans.

[0007] 4. The treatment effect may involve multiple aspects such as the degree of pain suppression, side effects (such as abnormal sensations), and tissue damage. It is difficult to systematically balance multiple goals through manual adjustment.

[0008] 5. Lack of individual adaptation: Existing methods usually use a general or averaged spinal cord model, which fails to take into account the specific differences of individual patients in terms of spinal cord geometry, electrode implantation location, tissue electrical properties (such as cerebrospinal fluid thickness and dural sac shape).

[0009] Existing research has also attempted some automated optimization methods, such as search based on predefined rules or simple single-objective optimization. However, these methods often fail to fully utilize patients' individualized anatomical data to construct accurate biophysical models, or are unable to efficiently handle high-dimensional, nonlinear parameter spaces, and are difficult to flexibly integrate complex, personalized simulation-based evaluation metrics. Therefore, they still fall short in terms of the accuracy and efficiency of personalized treatment.

[0010] Therefore, there is an urgent need for a system and method that can achieve personalized, automated, and efficient searching of optimal spinal cord stimulation parameters. Summary of the Invention

[0011] The purpose of this invention is to provide a system and method for optimizing spinal cord nerve stimulation parameters based on a surrogate model, which can automatically and efficiently determine the optimal combination of spinal cord nerve electrical stimulation treatment parameters that fit the individual anatomical characteristics of the target patient, so as to shorten the parameter adjustment time and optimize the effect of spinal cord electrical stimulation.

[0012] To achieve the above objectives, in a first aspect, the present invention proposes a spinal cord nerve stimulation parameter optimization system based on a surrogate model, comprising:

[0013] a. Data acquisition and personalized modeling module, used to acquire spinal medical data of the target patient and construct a specific three-dimensional finite element model of the spinal cord of the target patient based on the spinal medical data;

[0014] b. Stimulus parameter quantization module, used to encode a combination of stimulation parameters containing multiple stimulation channels into an input vector;

[0015] c. A stimulation effect evaluation module, connected to the stimulation parameter quantification module, is used to receive the input vector and generate a multi-objective evaluation vector characterizing the stimulation effect based on the specific spinal cord three-dimensional finite element model;

[0016] d. A proxy model unit, connected to the stimulus effect evaluation module, is used to acquire the input vector and the multi-objective evaluation vector to establish a sample database, and to train a machine learning model based on the sample database to establish a fast prediction mapping from the input vector to the multi-objective evaluation vector;

[0017] e. A multi-objective optimization unit, connected to the stimulus parameter quantization module and the surrogate model unit respectively, is used to iteratively generate candidate input vectors in the parameter space, call the surrogate model unit to make predictions, and update the candidate input vectors based on the multi-objective optimization strategy to obtain the optimal solution set;

[0018] f. An optimal solution output unit, connected to the multi-objective optimization unit, is used to determine and output recommended combinations of stimulus parameters from the optimal solution set obtained by the multi-objective optimization unit.

[0019] Optionally, the step of constructing a specific three-dimensional finite element model of the spinal cord for the target patient based on the spinal medical data includes: acquiring MRI or CT image data of the target patient, segmenting the image data using a simulation tool to generate a three-dimensional geometric model of the spinal cord, and assigning different tissue electrical property parameters to generate the specific three-dimensional finite element model of the spinal cord for electric field simulation.

[0020] Optionally, the expression for the input vector is as follows:

[0021] ;

[0022] Where n is the number of controlled stimulation channels or the total number of parameter dimensions, and the input vector represents a combination of stimulation parameters to be evaluated.

[0023] Optionally, the generation of a multi-objective evaluation vector representing the stimulus effect based on the specific three-dimensional finite element model of the spinal cord includes:

[0024] The input vector is imported into the specific spinal cord three-dimensional finite element model for simulation, and the electric field data of the stimulation area is output. Based on the electric field data, the multi-target evaluation vector corresponding to the input vector is calculated through a predefined mathematical function of stimulation effect and target side effect.

[0025] Optionally, calculating the multi-objective evaluation vector corresponding to the input vector includes:

[0026] Calculate the neural activation function of the corresponding voxel in the spinal cord white matter: , yes The magnitude of the electric potential at that point, It is the voxel corresponding to the spinal cord white matter; and the average value of the neural activation function is calculated as evaluation index one: ; It is the number of voxels corresponding to the white matter of the spinal cord;

[0027] Calculate the SAR value using the formula: ; yes It means conductivity at that point It means Electric field strength at that location, yes Tissue density at the location; and take the maximum value as evaluation index two: = ;

[0028] Determine the multi-objective evaluation vector .

[0029] Optionally, the machine learning model used by the proxy model unit is a Gaussian process regression model.

[0030] Optionally, the multi-objective optimization strategy is optimized using a multi-objective evolutionary algorithm based on Pareto dominance, including the NSGA-II algorithm.

[0031] Optionally, determining and outputting the recommended combination of stimulus parameters includes: visualizing the optimal solution set in the form of a Pareto front, receiving the user's selection instruction, and outputting the combination of stimulus parameters corresponding to the selected solution as a recommended scheme.

[0032] Optionally, the multi-objective optimization unit is also connected to the stimulus effect evaluation module, and is used to select a portion of the candidate input vectors and return them to the stimulus effect evaluation module to generate the corresponding multi-objective evaluation vectors, and to expand the candidate input vectors and the corresponding multi-objective evaluation vectors to the sample database for updating the surrogate model unit.

[0033] Secondly, the present invention provides a method for optimizing spinal cord nerve stimulation parameters based on a surrogate model, comprising the following steps:

[0034] S1: Obtain spinal medical data of the target patient and construct a specific three-dimensional finite element model of the spinal cord of the target patient based on the spinal medical data;

[0035] S2: Encode the combination of stimulation parameters containing multiple stimulation channels into an input vector;

[0036] S3: Based on the specific three-dimensional finite element model of the spinal cord, the input vector is simulated to generate a multi-objective evaluation vector for individualized stimulation effects;

[0037] S4: Generate a sample database of the surrogate model using the input vector and the multi-objective evaluation vector, train the surrogate model based on the sample database, and establish a fast prediction mapping from the input vector to the multi-objective evaluation vector;

[0038] S5: Run the multi-objective optimization algorithm and iteratively execute the following sub-steps: S51: Generate a set of candidate input vectors; S52: Predict the multi-objective evaluation vector corresponding to the candidate input vectors based on the surrogate model; S53: Perform multi-objective optimization based on the predicted multi-objective evaluation vectors to select the next generation of candidate input vectors;

[0039] S54: Select several high-value samples from the next-generation candidate input vectors, input them into step S3 for real simulation, generate the multi-objective evaluation vector of individualized stimulus effects, form new sample data, and update the surrogate model based on the new sample data;

[0040] S6: Based on the iteration results, output the optimal combination of stimulation parameters suitable for the target patient.

[0041] Optionally, the surrogate model adopts a Gaussian regression model, and the multi-objective optimization algorithm includes the NSGA-II algorithm.

[0042] In summary, the present invention has at least one of the following beneficial effects:

[0043] 1. This invention provides a system and method for optimizing spinal cord stimulation parameters based on a surrogate model. The system and method first construct a specific three-dimensional finite element model of the spinal cord for the target patient based on spinal medical data; then, the stimulation parameters are encoded as input vectors, and biophysical simulation is performed using the specific spinal cord three-dimensional finite element model to obtain accurate individual effect evaluation; based on this, a surrogate model is trained as an efficient substitute for the simulator; finally, a multi-objective optimization algorithm is driven to iteratively search in the individual parameter space, ultimately outputting the optimal combination of stimulation parameters for the patient's anatomical characteristics. This invention combines computational biology modeling with artificial intelligence optimization, overcoming the shortcomings of traditional SCS parameter tuning that rely on experience and ignore individual differences. It achieves a leap from a "general solution" to a "one-person-one-policy" precision medicine, significantly improving the effect, safety, and tuning efficiency of spinal cord stimulation.

[0044] 2. The present invention provides a system and method for optimizing spinal nerve stimulation parameters based on a surrogate model. By using a surrogate model to replace time-consuming simulations or clinical tests for internal iteration, the optimization search process is greatly accelerated, and fully automatic parameter optimization is achieved. It has the characteristics of high efficiency and automation.

[0045] 3. The present invention provides a system and method for optimizing spinal cord nerve stimulation parameters based on a surrogate model. It adopts an advanced multi-objective optimization algorithm, which can effectively explore in a high-dimensional parameter space, find the global or near-global Pareto optimal solution set, avoid the local optimum trap of manual adjustment, and has global optimum.

[0046] 4. The present invention provides a system and method for optimizing spinal nerve stimulation parameters based on a surrogate model. The system can integrate the medical imaging data of an individual patient to construct a simulation model, or use the patient's historical treatment data to train a surrogate model, thereby achieving truly personalized parameter optimization and realizing personalized and precision medicine.

[0047] 5. The present invention provides a system and method for optimizing spinal nerve stimulation parameters based on a surrogate model, which can clearly handle multiple competing treatment objectives (such as maximizing stimulation effect and minimizing side effects), provide a series of trade-off schemes for clinical decision-making, make the spinal nerve stimulation effect better, and achieve a comprehensive multi-objective balance.

[0048] 6. The present invention provides a system and method for optimizing spinal nerve stimulation parameters based on a surrogate model. The optimization process can be performed in advance in a simulation environment, and safety can be assessed through a biophysical model. This reduces the risk of conducting extensive parameter tests directly on target patients and is therefore safe. Attached Figure Description

[0049] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0050] Figure 1 This is a schematic diagram of the structure of a spinal cord nerve stimulation parameter optimization system based on a surrogate model provided in an embodiment of the present invention;

[0051] Figure 2 This is a schematic diagram of a finite element mesh model of the spinal cord provided in an embodiment of the present invention;

[0052] Figure 3 This is a schematic diagram of a quantitative input stimulation parameter provided in an embodiment of the present invention, wherein the left side is the implant, the numbers on the electrodes represent voltage values ​​in V, and the right side is the input vector formed after the parameters are quantized;

[0053] Figure 4 The Pareto front plot output by the system after iterative optimization, as provided in an embodiment of the present invention;

[0054] Figure 5 This is a flowchart of a method for optimizing spinal cord nerve stimulation parameters based on a surrogate model, provided as an embodiment of the present invention. Detailed Implementation

[0055] This invention provides a system and method for optimizing spinal cord nerve stimulation parameters based on a surrogate model. To enable those skilled in the art to better understand the technical solutions of this invention, the technical solutions in the embodiments of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this invention, and not all of them. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of this invention.

[0056] Please refer to Figure 1 , Figure 1 A schematic diagram of a spinal cord stimulation parameter optimization system based on a surrogate model, provided in an embodiment of the present invention, includes:

[0057] a. A data acquisition and personalized modeling module is used to acquire spinal medical data of the target patient and construct a specific three-dimensional finite element model of the spinal cord based on the spinal medical data. Specifically, the acquired spinal medical data includes, but is not limited to, spinal CT or MRI image data, and key structures such as the spinal cord, cerebrospinal fluid, vertebrae, and electrode arrays (if implanted) are manually or automatically segmented using image segmentation tools. In a specific embodiment of the present invention, MRI image data is imported into the Sim4life simulation tool for image segmentation. After segmentation, a three-dimensional geometric model of the spinal cord is automatically generated. Sim4life assigns corresponding conductivity parameters to different tissues, sets boundary conditions, uses Sim4Life's multi-channel function for electrode voltage values, and sets the current density at the spatial boundary of the entire simulation model to zero, thereby generating a specific three-dimensional finite element model of the spinal cord for the target patient used for electric field simulation, such as... Figure 2 As shown, the model can accurately reflect the spinal cord geometry of the target patient, the actual location and orientation of electrode implantation, and the distribution and conductivity properties of various tissues (such as cerebrospinal fluid, white matter, gray matter, and dura mater).

[0058] b. Stimulation parameter quantization module, used to encode stimulation parameters from multiple stimulation channels into an input vector. Specifically, for a multi-electrode stimulation system, the stimulation intensity (e.g., voltage value, current value) of each electrode is sequentially combined into an input vector, the expression of which is as follows:

[0059] ;

[0060] Where n is the number of controlled stimulation channels or the total number of parameter dimensions, and this input vector represents a combination of stimulation parameters to be evaluated.

[0061] In one specific embodiment of the present invention, taking a voltage configuration optimization system for a spinal cord stimulator with 16 electrode contacts as an example, the voltage of each electrode contact can be set to be... The stimulation parameters are adjusted and encoded into a 16-dimensional input vector:

[0062] .

[0063] c. A stimulus effect evaluation module, connected to the stimulus parameter quantification module, is used to receive the input vector and generate a multi-objective evaluation vector characterizing the stimulus effect based on at least one evaluation unit. The output of the stimulus effect evaluation module is one or more evaluation indicators, and multiple evaluation indicators can constitute a multi-objective evaluation vector.

[0064] ;

[0065] m represents the number of evaluation indicators.

[0066] Optionally, the stimulus effect evaluation module includes at least one of a functional evaluation unit, a data-driven evaluation unit, and a simulation-driven evaluation unit. The functional evaluation unit is configured to directly calculate the input vector using predefined mathematical functions of stimulus effect (e.g., neural activation function AF) and target side effects (e.g., specific absorptivity SAR). The corresponding evaluation metrics. The data-driven evaluation unit is configured to: access a database storing historical stimulus parameters and their corresponding clinical effect markers, and predict the input vector through table lookup or regression models. The corresponding clinical outcome score. The simulation-driven evaluation unit is configured to: construct a patient-specific spinal cord-electric field finite element simulation model, and input the vector... Import the simulation model for calculation, output the electric field distribution in the stimulation area, and calculate the activation volume or proportion of the target neural tissue based on the preset electric field activation threshold.

[0067] In a specific embodiment of the present invention, a combination of simulation-driven and function-based approaches is used to define two optimization objectives. The input vector is imported into the specific spinal cord three-dimensional finite element model for simulation, and the electric field data of the stimulation area is output. Based on the electric field data, a multi-objective evaluation vector corresponding to the input vector is calculated using predefined mathematical functions of stimulation effect (neural activation function AF) and target side effects (specific absorption rate SAR).

[0068] Objective 1 (Maximize): Spinal cord ascending nerve activation function value ( ).

[0069] Based on the constructed specific three-dimensional finite element model of the spinal cord of the target patient, multi-channel basis vectors are obtained through multi-channel simulation using Sim4life, and an electric field data vector space is constructed based on the multi-channel basis vectors. In a specific embodiment of the present invention, 16 electrode contacts correspond to 16 channels. Sim4life performs 16 simulations of single-channel independent excitation to obtain 16 basis vectors. A 16-dimensional electric field vector space is constructed based on these 16 basis vectors. By importing any 16-dimensional input vector, the product of the matrix formed by the 16 basis vectors and the 16-dimensional input vector is the electric field data of the specific three-dimensional finite element model of the spinal cord corresponding to the 16-dimensional input vector.

[0070] Using the obtained electric field data and masking techniques, the neural activation function of the corresponding voxel in the spinal cord white matter was calculated based on a formula:

[0071] ;

[0072] Where x is the voxel corresponding to the spinal white matter, and V(x) is the potential at x. The masking technique uses a binary mask, with the voxel corresponding to the spinal white matter marked as 1 and the voxel corresponding to the non-spinal white matter marked as 0. Only the voxel corresponding to the spinal white matter marked as 1 is used for calculation.

[0073] Calculate the average value of the neural activation function:

[0074] ;

[0075] Where n is the number of voxels corresponding to the white matter of the spinal cord.

[0076] Objective 2 (Minimization): Specific Absorption SAR ( ), calculate the SAR value using the formula:

[0077] ;

[0078] Where x is a voxel. This refers to the conductivity at point x. This refers to the electric field strength at point x. It represents the tissue density at point x.

[0079] Take the maximum value as the evaluation metric:

[0080] = ;

[0081] .

[0082] Therefore, multi-objective evaluation vector This transforms the minimization problem into maximizing the negative value.

[0083] d. A surrogate model unit, connected to the stimulus effect evaluation module, is used to establish a sample database based on the input vector and the multi-objective evaluation vector obtained from a specific three-dimensional finite element model of the target patient's spinal cord, and to train a machine learning model based on the sample database to establish an individualized rapid prediction mapping from the input vector to the multi-objective evaluation vector. Specifically, the surrogate model unit uses approximate mathematical models such as Gaussian process regression, neural networks, support vector regression, or radial quantity functions as approximate substitutes for the stimulus effect evaluation module. This surrogate model uses a certain number of initial samples... Training is conducted to establish a rapid mapping relationship from the stimulus parameter space to the effect evaluation space. Its characteristic is that the prediction speed is much faster than direct simulation or clinical assessment.

[0084] In one specific embodiment of the present invention, a Latin hypercube sampling method is used to generate 200 initial input vectors in a 16-dimensional parameter space. Using the Sim4life finite element simulation software and the aforementioned evaluation function, the multi-objective evaluation vectors corresponding to these 200 samples are calculated. Gaussian process regression was chosen as the surrogate model, based on 200 initial samples. A Gaussian process regression surrogate model was trained to establish a mapping relationship between the input vector and the multi-objective evaluation vector. Gaussian process regression was chosen as the surrogate model because it can provide an estimate of the uncertainty of the predicted values, which is beneficial for guiding the exploration.

[0085] e. A multi-objective optimization unit, connected to both the stimulus parameter quantization module and the surrogate model unit, iteratively generates candidate input vectors in the parameter space, calls the surrogate model unit for prediction, and updates the candidate input vectors based on a multi-objective optimization strategy to obtain the optimal solution set. Specifically, the multi-objective optimization unit integrates multi-objective optimization algorithms such as NSGA-II, MOEA / D, and Bayesian optimization, and works collaboratively with the surrogate model unit: such as... Figure 1 As shown, the multi-objective optimization unit proposes a batch of candidate input vectors through optimization algorithms. The surrogate model quickly predicts its corresponding multi-objective evaluation vector. ; The optimization algorithm is based on prediction Incorporating uncertainties, a new generation of better candidate input vectors is generated using a multi-objective optimization strategy (such as Pareto front search). This process iterates until stopping conditions (such as the number of iterations or a convergence criterion) are met. Optionally, the multi-objective optimization unit is configured to use a Pareto-dominant multi-objective evolutionary algorithm, including the NSGA-II algorithm, for optimization. Furthermore, the multi-objective optimization unit is also connected to the stimulus effect evaluation module, used to select some candidate solutions during the iteration process and return them to the stimulus effect evaluation module for real-world simulation to obtain the multi-objective evaluation vector. The candidate input vectors and the multi-objective evaluation vectors are then added to the sample database to update the surrogate model unit.

[0086] In one specific embodiment of the present invention, the NSGA-II algorithm is used as a multi-objective optimizer.

[0087] Iterative process: a. Generate 100 candidate input vectors using the NSGA-II algorithm. .

[0088] b. Based on the previously trained Gaussian process regression surrogate model, quickly predict the multi-objective evaluation vectors corresponding to the 100 candidate input vectors. And forecasting uncertainty.

[0089] c. Based on the above predictions, NSGA-II calculates the fitness of each candidate input vector, performs non-dominated sorting and crowding distance calculation, and selects the next generation population.

[0090] d. Based on the expected improvement EI criterion, select several high-value samples from the new generation population, i.e., samples with the highest prediction uncertainty and the greatest optimization potential. Use Sim4life finite element simulation software to import the samples into a specific spinal cord three-dimensional finite element model for simulation. Based on the aforementioned evaluation function, obtain accurate... This data was then expanded into the training sample library to update the Gaussian process regression surrogate model.

[0091] e. Repeat step ad for 100 iterations. The surrogate model eventually reaches optimal accuracy, and NSGA-II outputs the final Pareto optimal solution set.

[0092] f. An optimal solution output unit, connected to the multi-objective optimization unit, is used to determine and output recommended combinations of stimulus parameters from the optimal solution set obtained by the multi-objective optimization unit. Specifically, this includes visualizing the optimal solution set in the form of a Pareto front, receiving user selection instructions, and outputting the stimulus parameter combination corresponding to the selected solution as a recommended solution.

[0093] In a specific embodiment of the present invention, after the iteration is complete, the algorithm outputs a Pareto front solution set. The system visualizes the distribution of these solutions in the two-dimensional space of "activation function value (AF)" and "specific absorptivity (SAR)" in graphical form, such as... Figure 4 As shown. Clinicians can choose an equilibrium point on this frontier as needed, for example, choosing the solution with the lowest energy that satisfies the minimum coverage constraint (i.e., Figure 4 (The point marked with an asterisk) The system outputs the corresponding combination of electrode voltage parameters and receives the user's selection instructions. For example, if the user selects the scheme of "maximum activation function value under the condition of SAR<500W / Kg", the system will output the corresponding 16 electrode voltage values ​​as the optimal treatment parameters for the patient.

[0094] In other embodiments of the present invention, the constructed patient-specific three-dimensional finite element model of the spinal cord can not only be simulated using multi-channel numerical simulation software like Sim4life, but the electric field distribution map output from the electric field data constructed based on the multi-channel basis vectors can also be directly used as a visual aid to help doctors understand the mechanism of action of the optimization plan. For patients who have not yet had electrodes implanted, different electrode implantation sites can be simulated in the patient-specific three-dimensional finite element model to perform preoperative planning optimization and select the optimal implantation target. At the same time, the surrogate model can be updated periodically using new feedback data (such as efficacy scores) during follow-up of the target patient to achieve dynamic adaptive optimization.

[0095] like Figure 5 As shown, an embodiment of the present invention also provides a method for optimizing spinal cord nerve stimulation parameters based on a patient-specific model, comprising the following steps:

[0096] S1: Obtain spinal medical data of the target patient and construct a specific three-dimensional finite element model of the spinal cord of the target patient based on the spinal medical data;

[0097] S2: Encode the combination of stimulation parameters containing multiple stimulation channels into an input vector;

[0098] S3: Based on the specific three-dimensional finite element model of the spinal cord, the input vector is simulated to generate a multi-objective evaluation vector for individualized stimulation effects;

[0099] S4: Generate a sample database of the surrogate model using the input vector and the multi-objective evaluation vector, train the surrogate model based on the sample database, and establish a fast prediction mapping from the input vector to the multi-objective evaluation vector;

[0100] S5: Run the multi-objective optimization algorithm and iteratively execute the following sub-steps: S51: Generate a set of candidate input vectors; S52: Predict the multi-objective evaluation vector based on the surrogate model;

[0101] S53: Perform multi-objective optimization based on the predicted multi-objective evaluation vector to select the next generation of candidate input vectors;

[0102] S54: Select several high-value samples from the next-generation candidate input vectors, input them into step S3 for real simulation, generate the multi-objective evaluation vector of individualized stimulus effects, form new sample data, and update the surrogate model based on the new sample data;

[0103] S6: Based on the iteration results, output the optimal combination of stimulation parameters suitable for the target patient.

[0104] The modules, units, or steps in the embodiments of the present invention can be implemented in whole or in part by software, hardware, or a combination thereof. The modules described above can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the operations or steps corresponding to the modules described above.

[0105] The above description is merely a preferred embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A spinal cord nerve stimulation parameter optimization system based on a surrogate model, characterized in that, include: a. Data acquisition and personalized modeling module, used to acquire spinal medical data of the target patient and construct a specific three-dimensional finite element model of the spinal cord of the target patient based on the spinal medical data; b. A stimulus parameter quantization module, used to encode a combination of stimulus parameters containing multiple stimulus channels into an input vector; the expression of the input vector is as follows: ; Where n is the number of controlled stimulation channels or the total number of parameter dimensions, and the input vector represents a combination of stimulation parameters to be evaluated; c. A stimulus effect evaluation module, connected to the stimulus parameter quantification module, is used to receive the input vector and generate a multi-objective evaluation vector characterizing the stimulus effect based on the specific spinal cord three-dimensional finite element model, including: The input vector is imported into the specific spinal cord three-dimensional finite element model for simulation, and the electric field data of the stimulation area is output. Based on the electric field data, the multi-objective evaluation vector corresponding to the input vector is calculated through a predefined mathematical function of stimulation effect and target side effects, including: Calculate the neural activation function of the corresponding voxel in the spinal cord white matter: , yes The magnitude of the electric potential at that point, It is the voxel corresponding to the spinal cord white matter; and the average value of the neural activation function is calculated as evaluation index one: ; It is the number of voxels corresponding to the white matter of the spinal cord; Calculate the SAR value using the formula: ; yes It means conductivity at that point It means Electric field strength at that location, yes Tissue density at the location; and take the maximum value as evaluation index two: = ; Determine the multi-objective evaluation vector ; d. A proxy model unit, connected to the stimulus effect evaluation module, is used to acquire the input vector and the multi-objective evaluation vector to establish a sample database, and to train a machine learning model based on the sample database to establish a fast prediction mapping from the input vector to the multi-objective evaluation vector; e. A multi-objective optimization unit, connected to the stimulus parameter quantization module and the surrogate model unit respectively, is used to iteratively generate candidate input vectors in the parameter space, call the surrogate model unit to make predictions, and update the candidate input vectors based on the multi-objective optimization strategy to obtain the optimal solution set; f. An optimal solution output unit, connected to the multi-objective optimization unit, is used to determine and output recommended combinations of stimulus parameters from the optimal solution set obtained by the multi-objective optimization unit.

2. The spinal cord nerve stimulation parameter optimization system based on a surrogate model according to claim 1, characterized in that, The construction of a specific three-dimensional finite element model of the spinal cord for the target patient based on the spinal medical data includes: acquiring MRI or CT image data of the target patient, segmenting the image data using a simulation tool to generate a three-dimensional geometric model of the spinal cord, assigning different tissue electrical property parameters, and generating the specific three-dimensional finite element model of the spinal cord for electric field simulation.

3. The spinal cord nerve stimulation parameter optimization system based on a surrogate model according to claim 1 or 2, characterized in that, The machine learning model used by the proxy model unit is a Gaussian process regression model.

4. The spinal cord stimulation parameter optimization system based on a surrogate model according to claim 1 or 2, characterized in that, The multi-objective optimization strategy is optimized using a multi-objective evolutionary algorithm based on Pareto dominance, including the NSGA-II algorithm.

5. The spinal cord stimulation parameter optimization system based on a surrogate model according to claim 1 or 2, characterized in that, The process of determining and outputting the recommended combination of stimulus parameters includes: visualizing the optimal solution set in the form of a Pareto front, receiving the user's selection instruction, and outputting the combination of stimulus parameters corresponding to the selected solution as a recommended scheme.

6. The spinal cord stimulation parameter optimization system based on a surrogate model according to claim 1 or 2, characterized in that, The multi-objective optimization unit is also connected to the stimulus effect evaluation module, and is used to select a portion of the candidate input vectors and return them to the stimulus effect evaluation module to generate the corresponding multi-objective evaluation vectors, and to expand the candidate input vectors and the corresponding multi-objective evaluation vectors to the sample database for updating the surrogate model unit.

7. A method for optimizing spinal cord nerve stimulation parameters based on a surrogate model, employing the spinal cord nerve stimulation parameter optimization system based on a surrogate model as described in any one of claims 1-6, characterized in that, Includes the following steps: S1: Obtain spinal medical data of the target patient and construct a specific three-dimensional finite element model of the spinal cord of the target patient based on the spinal medical data; S2: Encode the combination of stimulation parameters containing multiple stimulation channels into an input vector; S3: Based on the specific three-dimensional finite element model of the spinal cord, the input vector is simulated to generate a multi-objective evaluation vector for individualized stimulation effects; S4: Generate a sample database of the surrogate model using the input vector and the multi-objective evaluation vector, train the surrogate model based on the sample database, and establish a fast prediction mapping from the input vector to the multi-objective evaluation vector; S5: Run the multi-objective optimization algorithm and iteratively execute the following sub-steps: S51: Generate a set of candidate input vectors; S52: Predict the multi-objective evaluation vector corresponding to the candidate input vectors based on the surrogate model; S53: Perform multi-objective optimization based on the predicted multi-objective evaluation vectors to select the next generation of candidate input vectors; S54: Select several high-value samples from the next-generation candidate input vectors, input them into step S3 for real simulation, generate the multi-objective evaluation vector of individualized stimulus effects, form new sample data, and update the surrogate model based on the new sample data; S6: Based on the iteration results, output the optimal combination of stimulation parameters suitable for the target patient.

Citation Information

Patent Citations

  • Rapid multi-target space reflector optimization design method based on agent model

    CN113065268A

  • Spinal cord electrical stimulation parameter determination method, electronic equipment and storage medium

    CN118846377A