A method and device for optimizing multi-channel transcranial stimulation parameters based on generative models.

Abstract

This invention provides a method and apparatus for optimizing multi-channel transcranial stimulation parameters based on a generative model, relating to the field of smart healthcare technology. The method includes: determining target field feature parameters; inputting the target field feature parameters into a stimulation parameter determination model to obtain transcranial stimulation parameters output by the model, wherein the transcranial stimulation parameters are used to realize the target field corresponding to the target field feature parameters; the stimulation parameter determination model is trained based on sample field feature parameters and multi-channel sample transcranial stimulation parameters, and can provide high-precision multi-channel transcranial stimulation parameters.

Description

Technical Field

[0001] This invention relates to the field of smart medical technology, and in particular to a method and apparatus for optimizing multi-channel transcranial stimulation parameters based on a generative model. Background Technology

[0002] Transcranial neuromodulation techniques, such as transcranial magnetic stimulation (TMS) and transcranial electrical stimulation (tES), play an important role in neuroscience research and clinical treatment as non-invasive neuromodulation methods. These techniques are widely used in the intervention and treatment of neurological diseases such as depression, Parkinson's disease, and stroke rehabilitation.

[0003] Currently, traditional transcranial stimulation techniques have certain limitations in terms of stimulation precision. Taking transcranial electrical stimulation as an example, it typically uses a pair of large-area sheet electrodes for stimulation. This single-channel stimulation method can easily lead to the electric field generated by stimulation being too dispersed in the cortex, making it difficult to achieve sub-centimeter-level precise targeted stimulation, and may also cause unnecessary interference to non-target brain regions, affecting the therapeutic effect and safety.

[0004] To improve the localization accuracy of transcranial stimulation (TCS), multi-channel array electrode stimulation (MAES) technology has gradually developed and attracted attention in recent years. This technology aims to achieve more precise brain region modulation through the coordinated operation of multiple electrode channels. When optimizing the current parameters of each channel for a specific target brain region to achieve the desired stimulation effect, current mainstream methods are mostly based on linear methods such as convex optimization algorithms. These methods perform relatively well in solving simple linear optimization problems, but their performance is often limited when facing more complex nonlinear optimization problems, such as parameter optimization for novel brain modulation methods like time-coherent electrical stimulation and ultrasound stimulation, making it difficult to achieve the desired stimulation effect. Summary of the Invention

[0005] To address the problems existing in the prior art, this invention provides a method and apparatus for optimizing multi-channel transcranial stimulation parameters based on a generative model.

[0006] This invention provides a method for optimizing multi-channel transcranial stimulation parameters based on a generative model, comprising:

[0007] Determine the target field characteristic parameters;

[0008] The target field feature parameters are input into the stimulation parameter determination model to obtain the multi-channel transcranial stimulation parameters output by the stimulation parameter determination model. The transcranial stimulation parameters are used to realize the target field corresponding to the target field feature parameters. The stimulation parameter determination model is trained based on the sample field feature parameters and the multi-channel sample transcranial stimulation parameters.

[0009] According to the present invention, a multi-channel transcranial stimulation parameter optimization method based on a generative model, prior to training based on sample transcranial stimulation parameters and multi-channel sample transcranial stimulation parameters, the method further includes:

[0010] Determine the skull structure model;

[0011] Based on the aforementioned skull structure model, multiple sets of multi-channel transcranial stimulation parameters are determined; each set of multi-channel transcranial stimulation parameters is used to achieve a specific transcranial stimulation.

[0012] Simulation calculations are performed based on the transcranial stimulation parameters of the samples and the skull structure model to obtain the corresponding sample field characteristic parameters in the cerebral cortex. The field characteristic parameters include the sample field intensity distribution and / or sample field distribution characteristic parameters.

[0013] According to the generative model-based multichannel transcranial stimulation parameter optimization method provided by the present invention, after obtaining the corresponding sample field feature parameters in the cerebral cortex, the method further includes:

[0014] The transcranial stimulation parameters of the sample are input into the forward module of the original stimulation parameter determination model to obtain Gaussian noise data output by the forward module; the Gaussian noise data is obtained by forward diffusion of gradually adding noise to the transcranial stimulation parameters.

[0015] The Gaussian noise data is input into the inverse module of the original stimulation parameter determination model to obtain the predicted transcranial stimulation parameters recovered by the inverse module; the predicted transcranial stimulation parameters are obtained by inverse processing of the Gaussian noise data;

[0016] Determine the loss function between the predicted transcranial stimulation parameters and the sample transcranial stimulation parameters, and determine the parameters of the model by adjusting the original stimulation parameters by minimizing the loss function;

[0017] Repeat the step of inputting the transcranial stimulation parameters of the sample into the forward module of the original stimulation parameter determination model until the original stimulation parameter determination model converges to obtain the stimulation parameter determination model.

[0018] The sample field feature parameters are the control conditions for reverse processing.

[0019] According to the generative model-based multichannel transcranial stimulation parameter optimization method provided by the present invention, after obtaining the corresponding sample field feature parameters in the cerebral cortex, the method further includes:

[0020] The sample field feature parameters and the sample transcranial stimulation parameters are input into the pre-training matching module of the original stimulation parameter determination model to obtain the sample transcranial stimulation parameter encoding vector output by the pre-training matching module.

[0021] The encoded vector is input into the forward module of the original stimulus parameter determination model to obtain Gaussian noise data output by the forward module; the Gaussian noise data is obtained by forward diffusion of the encoded vector with noise added stepwise.

[0022] The Gaussian noise data is input into the inverse module of the original stimulation parameter determination model to obtain the predicted transcranial stimulation parameter encoding vector recovered by the inverse module; the predicted transcranial stimulation parameter encoding vector is obtained by inverse processing of the Gaussian noise data;

[0023] Determine the loss function for the predicted transcranial stimulation parameter encoding vector and the sample transcranial stimulation parameter encoding vector, and determine the model parameters by adjusting the original stimulation parameters by minimizing the loss function;

[0024] Repeat the steps of inputting the sample field feature parameters and the sample transcranial stimulation parameters into the pre-training matching module of the original stimulation parameter determination model until the original stimulation parameter determination model converges to obtain the stimulation parameter determination model.

[0025] The sample field feature parameters are the control conditions for the reverse processing; the stimulation parameter determination model includes a decoding module, which is used to decode the predicted transcranial stimulation parameter encoding vector output by the reverse module to obtain the predicted transcranial stimulation parameters.

[0026] According to the present invention, a method for optimizing multi-channel transcranial stimulation parameters based on a generative model is provided, wherein determining the cranial structure model includes:

[0027] A standard skull structure model is determined, which is constructed based on the average brain structure of a population; or

[0028] Identify the target structural image data, and construct the corresponding individualized skull structure model based on the target structural image data.

[0029] The present invention also provides a multi-channel transcranial stimulation parameter optimization device based on a generative model, comprising:

[0030] The feature parameter determination module is used to determine the feature parameters of the target field;

[0031] The stimulation parameter determination module is used to input the target field feature parameters into the stimulation parameter determination model to obtain the multi-channel transcranial stimulation parameters output by the stimulation parameter determination model. The transcranial stimulation parameters are used to realize the target field corresponding to the target field feature parameters. The stimulation parameter determination model is trained based on the sample field feature parameters and the multi-channel sample transcranial stimulation parameters.

[0032] A multi-channel transcranial stimulation parameter optimization device based on a generative model according to the present invention further includes:

[0033] The skull structure model determination module is used to determine the skull structure model;

[0034] The sample stimulation parameter determination module is used to determine multiple sets of multi-channel transcranial stimulation parameters based on the skull structure model; each set of multi-channel transcranial stimulation parameters is used to achieve a transcranial stimulation.

[0035] The sample feature parameter determination module is used to perform simulation calculations based on the transcranial stimulation parameters of the sample and the skull structure model to obtain the corresponding sample field feature parameters in the cerebral cortex. The field feature parameters include the sample field strength distribution and / or sample field distribution feature parameters.

[0036] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the multi-channel transcranial stimulation parameter optimization method based on the generative model described above. The electronic device includes a transcranial stimulation device.

[0037] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the multi-channel transcranial stimulation parameter optimization method based on a generative model as described above.

[0038] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the multi-channel transcranial stimulation parameter optimization method based on a generative model as described above.

[0039] The present invention provides a method and apparatus for optimizing multi-channel transcranial stimulation parameters based on a generative model. This method determines the target field feature parameters and clarifies the field characteristics expected to be achieved in the target cerebral cortex. It facilitates the generation of high-precision transcranial stimulation parameters by the stimulation parameter determination model through the linear or nonlinear relationship between the learned target field feature parameters and transcranial stimulation parameters. Furthermore, the stimulation parameter determination model is trained based on multi-channel sample transcranial stimulation parameters, thus facilitating the generation of multi-channel transcranial stimulation parameters and enhancing the stimulation effect when optimizing parameters for novel brain modulation methods such as time-coherent electrical stimulation and ultrasound stimulation. Attached Figure Description

[0040] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0041] Figure 1 This is a flowchart illustrating the multi-channel transcranial stimulation parameter optimization method based on a generative model provided by the present invention.

[0042] Figure 2 This is a schematic diagram of the structure of the multi-channel transcranial stimulation parameter optimization device based on a generative model provided by the present invention.

[0043] Figure 3 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0044] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0045] The following is combined Figures 1-3 The present invention describes a method and apparatus for optimizing multichannel transcranial stimulation parameters based on a generative model, which can be used in the fields of brain-computer interfaces and neuromodulation.

[0046] Figure 1 This is a flowchart illustrating the multi-channel transcranial stimulation parameter optimization method based on a generative model provided by the present invention, as shown below. Figure 1 As shown, the method includes the following steps:

[0047] Step 101: Determine the target field characteristic parameters.

[0048] Target field characteristic parameters refer to parameters that describe the expected field distribution characteristics of transcranial stimulation parameters in the target cerebral cortex, and are used to characterize the specific physical characteristics of the stimulation effect produced by transcranial stimulation parameters.

[0049] For example, brain function regulation targets can be set first, such as activating the left primary motor cortex through transcranial electrical stimulation to treat motor dysfunction; then the target field strength and distribution range to be applied to the left primary motor cortex can be determined, as long as the left primary motor cortex can be effectively stimulated; then the final target field characteristic parameters can be determined based on the target field strength and distribution range.

[0050] Step 102: Input the target field feature parameters into the stimulation parameter determination model to obtain the multi-channel transcranial stimulation parameters output by the stimulation parameter determination model. The transcranial stimulation parameters are used to realize the target field corresponding to the target field feature parameters. The stimulation parameter determination model is trained based on the sample field feature parameters and the multi-channel sample transcranial stimulation parameters.

[0051] Transcranial stimulation parameters refer to the specific configuration parameters of transcranial stimulation required to achieve the corresponding field strength distribution for brain function regulation. It can be understood that, based on the stimulation parameter determination model trained with multi-channel sample transcranial stimulation parameters, the transcranial stimulation parameters obtained by inputting the target field feature parameters into the stimulation parameter determination model are also multi-channel transcranial stimulation parameters.

[0052] Multichannel refers to the simultaneous or coordinated application of transcranial stimulation using multiple independent stimuli or electrodes (channels) to achieve precise modulation of specific regions of the cerebral cortex. For example, each channel can independently control its stimulation parameters such as current intensity, phase, frequency, or the position of the positioning electrode.

[0053] The target field refers to the field generated in the target cerebral cortex by transcranial stimulation parameters. Depending on the specific type of transcranial stimulation parameters, the target field can be an electric field, a magnetic field, or a sound field, etc.

[0054] The multi-channel transcranial stimulation parameter optimization method based on a generative model provided in this invention determines the target field feature parameters and clarifies the field characteristics expected to be achieved in the target cerebral cortex. This facilitates the generation of high-precision transcranial stimulation parameters by the stimulation parameter determination model through the linear or nonlinear relationship between the learned target field feature parameters and transcranial stimulation parameters. Furthermore, the stimulation parameter determination model is trained based on multi-channel sample transcranial stimulation parameters, thus facilitating the generation of multi-channel transcranial stimulation parameters and enhancing the stimulation effect when optimizing parameters for novel brain modulation methods such as time-coherent electrical stimulation and ultrasound stimulation.

[0055] It is understood that the multi-channel transcranial stimulation parameter optimization method based on generative models provided in this embodiment of the invention generates high-precision transcranial stimulation parameters by learning the linear or nonlinear relationship between the target field feature parameters and the transcranial stimulation parameters. It is not limited to linear optimization scenarios and can be applied to various types of transcranial brain modulation scenarios such as multi-channel transcranial magnetic stimulation, transcranial ultrasound stimulation, and time-coherent electrical stimulation.

[0056] Based on the above embodiments, before training based on sample transcranial stimulation parameters and multi-channel sample transcranial stimulation parameters, the method further includes:

[0057] Determine the skull structure model;

[0058] Based on the aforementioned skull structure model, multiple sets of multi-channel transcranial stimulation parameters are determined; each set of multi-channel transcranial stimulation parameters is used to achieve a specific transcranial stimulation.

[0059] Simulation calculations are performed based on the transcranial stimulation parameters of the samples and the skull structure model to obtain the corresponding sample field characteristic parameters in the cerebral cortex. The field characteristic parameters include the sample field intensity distribution and / or sample field distribution characteristic parameters.

[0060] Specifically, the sample field strength distribution can be an electric field distribution, a magnetic field distribution, or a sound field distribution, etc., and the sample field distribution characteristic parameters can be distribution characteristics obtained based on the field strength distribution, such as focus and extrema.

[0061] Taking transcranial electrical stimulation (tDCS) as an example of a non-invasive neuromodulation method to achieve brain function regulation, different tDCS stimulation parameters can be determined for each individualized cranial structure model. The types of tDCS stimulation parameters can include bipolar tDCS stimulation parameters, high-resolution tDCS stimulation parameters, and multi-channel tDCS stimulation parameters under different optimization methods.

[0062] For each type of tDCS stimulation parameters, a specific set of multi-channel sample transcranial stimulation parameters can be obtained by adjusting parameters such as the position of the positioning electrodes and the stimulation intensity on the corresponding individualized skull structure model.

[0063] After obtaining multiple sets of specific multi-channel transcranial stimulation parameters, for example, a skull structure model and a corresponding set of transcranial stimulation parameters can be input into a simulation calculation system to obtain the sample field characteristic parameters within the individualized skull structure model calculated using the finite element analysis method, as output by the simulation calculation system. The skull structure model and multiple sets of transcranial stimulation parameters can be repeatedly input into the simulation calculation system to obtain the corresponding sample field characteristic parameters.

[0064] Based on any of the above embodiments, after obtaining the corresponding sample field feature parameters in the cerebral cortex, the method further includes:

[0065] The transcranial stimulation parameters of the sample are input into the forward module of the original stimulation parameter determination model to obtain Gaussian noise data output by the forward module; the Gaussian noise data is obtained by forward diffusion of gradually adding noise to the transcranial stimulation parameters.

[0066] The Gaussian noise data is input into the inverse module of the original stimulation parameter determination model to obtain the predicted transcranial stimulation parameters recovered by the inverse module; the predicted transcranial stimulation parameters are obtained by inverse processing of the Gaussian noise data;

[0067] Determine the loss function between the predicted transcranial stimulation parameters and the sample transcranial stimulation parameters, and determine the parameters of the model by adjusting the original stimulation parameters by minimizing the loss function;

[0068] Repeat the step of inputting the transcranial stimulation parameters of the sample into the forward module of the original stimulation parameter determination model until the original stimulation parameter determination model converges to obtain the stimulation parameter determination model.

[0069] The sample field feature parameters are the control conditions for reverse processing.

[0070] Specifically, the forward module, also known as the forward module or the precursor module, is used to implement the forward process of the stimulus parameter determination model. For example, the input transcranial stimulation parameters can be progressively transformed into Gaussian noise data based on parameterized Markov chain rules. This process can be represented by the following formula:

[0071]

[0072] in, Forward process, The input sample transcranial stimulation parameters, The data contains Gaussian noise. For the first t Noise data of the step, It follows a Gaussian distribution. The noise scheduling parameters added at each step of the forward diffusion process determine the amount of noise added at each step. I It is the identity matrix, and T is the total number of steps in the forward process.

[0073] The variance is used to control the intensity of the noise.

[0074] In this way, the transcranial stimulation parameters of the sample gradually accumulate noise during the forward diffusion steps, gradually losing their original distinct characteristics. Finally, when T approaches infinity, It is equivalent to an isotropic Gaussian noise distribution.

[0075] In one embodiment, different time steps It is predefined. It can be linear decay, exponential decay, etc., satisfying... .

[0076] The inverse module, also known as the backward module or the backward module, is used to implement the inverse process of the stimulus parameter determination model. For example, based on parameterized Markov chain rules and control conditions for the inverse process, Gaussian noise data can be gradually reduced to approximate the transcranial stimulation parameters of the sample. Specifically, a neural network can be used. Fit the reverse process.

[0077]

[0078] Where c represents the control condition, i.e., the sample field feature parameters, used to guide the generation of specific features. The number of steps in the reverse diffusion. The true inverse mean depends on and , The variance coefficients of the actual reverse process, and the noise scheduling parameters. Related, The mean is fitted to the neural network. These are noise scheduling parameters during the diffusion process. for The cumulative product, This is the noise term predicted by the neural network.

[0079] For example, This controls the proportion of original information retained in the current step's data, and its value is between 0 and 1; , indicating from the initial data To the The overall attenuation of the original signal during the step is used to adjust the dependence on noise when generating data; It is the number of steps in control condition c and reverse diffusion. Below is the noise data at the current step number. The estimate is used to approximate the feature parameters added to the original sample field during the forward process, or the forward process. The actual noise is gradually removed during the reverse process to generate target data that meets control condition c.

[0080] Compared to traditional transcranial stimulation parameters using a few electrodes, the present invention determines transcranial stimulation parameters for multiple channels, providing richer data on channel locations and parameter configurations. This allows the inverse processing of the stimulation parameter determination model to operate in a high-dimensional, diverse parameter space, or in other words, in a larger selection space. This increases the number of transcranial stimulation parameter combinations that the stimulation parameter determination model can explore, thereby improving the optimization accuracy and robustness of the output transcranial stimulation parameters.

[0081] For example, a mean squared error formula can be designed to define the loss function between the predicted transcranial stimulation parameters and the sample transcranial stimulation parameters. This allows the model parameters to be determined by adjusting the original stimulation parameters using gradient descent by minimizing the mean squared error formula. When the loss function converges to a preset threshold, such as a value less than 0.001, or when the maximum number of iterations is reached, such as 1000 iterations, the original stimulation parameter determination model is considered converged, resulting in the stimulation parameter determination model.

[0082] In one embodiment, after training the stimulation parameter determination model, the corresponding target field feature parameters can be determined first based on the actual brain function regulation effect or stimulation target requirements. Then, the target field feature parameters are input into the stimulation parameter determination model. Through the reverse module of the stimulation parameter determination model, the reverse process of the stimulation parameter determination model can be realized, thereby generating and outputting optimized high-precision multi-channel transcranial stimulation parameters.

[0083] In this embodiment, by using the field feature parameters corresponding to the transcranial stimulation parameters of the sample as control conditions, the inverse module is guided to perform reverse processing to remove noise. This can accelerate the convergence speed of the inverse module in mapping data with different noise levels to predicted transcranial stimulation parameters that are close to the sample transcranial stimulation parameters during the training process, and improve the accuracy of the transcranial stimulation parameters output by the stimulation parameter determination model.

[0084] Based on any of the above embodiments, after obtaining the corresponding sample field feature parameters in the cerebral cortex, the method further includes:

[0085] The sample field feature parameters and the sample transcranial stimulation parameters are input into the pre-training matching module of the original stimulation parameter determination model to obtain the sample transcranial stimulation parameter encoding vector output by the pre-training matching module.

[0086] The encoded vector is input into the forward module of the original stimulus parameter determination model to obtain Gaussian noise data output by the forward module; the Gaussian noise data is obtained by forward diffusion of the encoded vector with noise added stepwise.

[0087] The Gaussian noise data is input into the inverse module of the original stimulation parameter determination model to obtain the predicted transcranial stimulation parameter encoding vector recovered by the inverse module; the predicted transcranial stimulation parameter encoding vector is obtained by inverse processing of the Gaussian noise data;

[0088] Determine the loss function for the predicted transcranial stimulation parameter encoding vector and the sample transcranial stimulation parameter encoding vector, and determine the model parameters by adjusting the original stimulation parameters by minimizing the loss function;

[0089] Repeat the steps of inputting the sample field feature parameters and the sample transcranial stimulation parameters into the pre-training matching module of the original stimulation parameter determination model until the original stimulation parameter determination model converges to obtain the stimulation parameter determination model.

[0090] The sample field feature parameters are the control conditions for the reverse processing; the stimulation parameter determination model includes a decoding module, which is used to decode the predicted transcranial stimulation parameter encoding vector output by the reverse module to obtain the predicted transcranial stimulation parameters.

[0091] The pre-trained matching model can use contrastive learning to encode the sample field feature parameters and sample transcranial stimulation parameters to obtain the encoded vectors of the sample field feature parameters and the encoded vectors of the sample transcranial stimulation parameters, so as to learn their mapping relationship in the common embedding space and obtain key correlation features.

[0092] For example, the pre-trained matching model can be a CLIP (Contrastive Language-Image Pre-Training) contrastive learning model, and the decoding module can be a contrastive learning decoder.

[0093] The working principle and technical effect of the training stimulus parameter determination model in this embodiment are basically the same as those in the previous embodiments, and will not be repeated here. The difference is that in this embodiment, before inputting the positive module, the sample field feature parameters and sample transcranial stimulation parameters are first input into the pre-trained matching model of the original stimulus parameter determination model to obtain the encoded vector of sample transcranial stimulation parameters generated by encoding the sample field feature parameters and sample transcranial stimulation parameters using contrastive learning, which can better capture key features.

[0094] Compared to directly using sample field feature parameters and sample transcranial stimulation parameters to train the original stimulus parameter determination model, using encoding vectors can provide more consistent and normalized inputs, while learning the correlation features between the two in the encoding space, thus increasing the convergence speed of the original stimulus parameter determination model in the early training stage.

[0095] Furthermore, the pre-trained matching model can transform the sample field feature parameters and sample transcranial stimulation parameters into low-dimensional encoding vectors. This dimensionality reduction can reduce the computational overhead required for training the stimulation parameter determination model, thereby accelerating the training process of the stimulation parameter determination model.

[0096] Based on any of the above embodiments, determining the skull structure model includes:

[0097] A standard skull structure model is determined, which is constructed based on the average brain structure of a population;

[0098] Alternatively, target structural image data can be determined, and a corresponding individualized skull structure model can be constructed based on the target structural image data.

[0099] Specifically, the standard skull structure model, also known as the average head model or template skull structure model, refers to a general skull structure model constructed based on the average brain structure of a population, and is used as a skull structure template.

[0100] For example, MRI (Magnetic Resonance Imaging) data such as T1-weighted imaging (T1WI) collected from previous experiments with the consent of the subjects can be collected, or open-source structural image datasets can be used to align the brain images of each individual to a common reference space, and map the anatomical structure of each brain onto a standard template through nonlinear transformation; average all aligned individual brain images or their segmentation results on each voxel to construct a standard skull structure model.

[0101] The target structural image data can also be structural image data collected in previous experiments with the consent of the subjects, and / or open-source structural image datasets, etc. For example, the structural image data can be preprocessed with denoising, image segmentation and edge detection to extract the corresponding key geometric features; and a personalized skull structure model corresponding to the target structural image data can be constructed using 3D reconstruction algorithms such as stereo rendering or surface rendering.

[0102] In this embodiment, by designing both a standard cranial structure model and an individualized cranial structure model in parallel, a suitable cranial structure model can be selected according to the specific transcranial stimulation strategy, thereby enhancing the flexibility and specificity of the transcranial stimulation parameter determination method.

[0103] Specifically, when the transcranial stimulation parameter determination method is used to determine uniform transcranial stimulation parameters for a population, a standard cranial structure model is selected; when the transcranial stimulation parameter determination method is used to determine individualized stimulation strategies, an individualized cranial structure model is selected. This can balance the determination of universal transcranial stimulation parameters and the determination of individualized transcranial stimulation parameters, meeting the needs of different levels such as scientific research, clinical treatment, or product development.

[0104] To illustrate the functionality of the generative model-based multichannel transcranial stimulation parameter optimization method provided in this implementation, a specific example is given below.

[0105] A standard cranial structure model is determined, which is constructed based on the average brain structure of a population; or target structural image data is determined, and a corresponding individualized cranial structure model is constructed based on the target structural image data; multiple sets of multi-channel transcranial stimulation parameters are determined based on the cranial structure model; each set of multi-channel transcranial stimulation parameters is used to implement a transcranial stimulation; simulation calculations are performed based on the transcranial stimulation parameters and the cranial structure model to obtain the corresponding sample field feature parameters in the cerebral cortex, the sample field feature parameters including sample field strength distribution and / or sample field distribution feature parameters;

[0106] The sample field feature parameters and the sample transcranial stimulation parameters are input into the pre-training matching module of the original stimulation parameter determination model to obtain the sample transcranial stimulation parameter encoding vector output by the pre-training matching module; the encoding vector is input into the forward module of the original stimulation parameter determination model to obtain Gaussian noise data output by the forward module; the Gaussian noise data is obtained by progressively adding noise to the encoding vector and performing forward diffusion; the Gaussian noise data is input into the reverse module of the original stimulation parameter determination model to obtain the predicted transcranial stimulation parameter encoding vector recovered by the reverse module; the predicted transcranial stimulation parameter encoding vector is obtained by inversely processing the Gaussian noise data. The process involves: determining the loss function for the predicted transcranial stimulation parameter encoding vector and the sample transcranial stimulation parameter encoding vector; adjusting the parameters of the original stimulation parameter determination model by minimizing the loss function; repeating the steps of inputting the sample field feature parameters and the sample transcranial stimulation parameters into the pre-training matching module of the original stimulation parameter determination model until the original stimulation parameter determination model converges, thus obtaining the stimulation parameter determination model; wherein, the sample field feature parameters are the control conditions for the inverse processing; the stimulation parameter determination model includes a decoding module, which is used to decode the predicted transcranial stimulation parameter encoding vector output by the inverse module to obtain the predicted transcranial stimulation parameters;

[0107] Determine the target field characteristic parameters; input the target field characteristic parameters into the stimulation parameter determination model to obtain the optimized high-precision multi-channel transcranial stimulation parameters generated and output by the stimulation parameter determination model through the reverse process implemented by the reverse module.

[0108] The following describes the multi-channel transcranial stimulation parameter optimization device based on a generative model provided by the present invention. The multi-channel transcranial stimulation parameter optimization device based on a generative model described below can be referred to in correspondence with the multi-channel transcranial stimulation parameter optimization method based on a generative model described above.

[0109] Figure 2 This is a schematic diagram of the multi-channel transcranial stimulation parameter optimization device based on a generative model provided by the present invention, as shown below. Figure 2 As shown, the device includes:

[0110] The feature parameter determination module 210 is used to determine the target field feature parameters;

[0111] The stimulation parameter determination module 220 is used to input the target field feature parameters into the stimulation parameter determination model to obtain the multi-channel transcranial stimulation parameters output by the stimulation parameter determination model. The transcranial stimulation parameters are used to realize the target field corresponding to the target field feature parameters. The stimulation parameter determination model is trained based on the sample field feature parameters and the multi-channel sample transcranial stimulation parameters.

[0112] Based on any of the above embodiments, the multi-channel transcranial stimulation parameter optimization device based on a generative model further includes:

[0113] The skull structure model determination module is used to determine the skull structure model;

[0114] The sample stimulation parameter determination module is used to determine multiple sets of multi-channel transcranial stimulation parameters based on the skull structure model; each set of multi-channel transcranial stimulation parameters is used to achieve a transcranial stimulation.

[0115] The sample feature parameter determination module is used to perform simulation calculations based on the transcranial stimulation parameters of the sample and the individualized skull structure model to obtain the corresponding sample field feature parameters in the cerebral cortex. The field feature parameters include the sample field strength distribution and / or the sample field distribution feature parameters.

[0116] Based on any of the above embodiments, the multi-channel transcranial stimulation parameter optimization device based on a generative model further includes a model training module, used for:

[0117] The transcranial stimulation parameters of the sample are input into the forward module of the original stimulation parameter determination model to obtain Gaussian noise data output by the forward module; the Gaussian noise data is obtained by forward diffusion of gradually adding noise to the transcranial stimulation parameters.

[0118] The Gaussian noise data is input into the inverse module of the original stimulation parameter determination model to obtain the predicted transcranial stimulation parameters recovered by the inverse module; the predicted transcranial stimulation parameters are obtained by inverse processing of the Gaussian noise data;

[0119] Determine the loss function between the predicted transcranial stimulation parameters and the sample transcranial stimulation parameters, and determine the parameters of the model by adjusting the original stimulation parameters by minimizing the loss function;

[0120] Repeat the step of inputting the transcranial stimulation parameters of the sample into the forward module of the original stimulation parameter determination model until the original stimulation parameter determination model converges to obtain the stimulation parameter determination model.

[0121] The sample field feature parameters are the control conditions for reverse processing.

[0122] Based on any of the above embodiments, the model training module is further configured to:

[0123] The sample field feature parameters and the sample transcranial stimulation parameters are input into the pre-training matching module of the original stimulation parameter determination model to obtain the sample transcranial stimulation parameter encoding vector output by the pre-training matching module.

[0124] The encoded vector is input into the forward module of the original stimulus parameter determination model to obtain Gaussian noise data output by the forward module; the Gaussian noise data is obtained by forward diffusion of the encoded vector with noise added stepwise.

[0125] The Gaussian noise data is input into the inverse module of the original stimulation parameter determination model to obtain the predicted transcranial stimulation parameter encoding vector recovered by the inverse module; the predicted transcranial stimulation parameter encoding vector is obtained by inverse processing of the Gaussian noise data;

[0126] Determine the loss function for the predicted transcranial stimulation parameter encoding vector and the sample transcranial stimulation parameter encoding vector, and determine the model parameters by adjusting the original stimulation parameters by minimizing the loss function;

[0127] Repeat the steps of inputting the sample field feature parameters and the sample transcranial stimulation parameters into the pre-training matching module of the original stimulation parameter determination model until the original stimulation parameter determination model converges to obtain the stimulation parameter determination model.

[0128] The sample field feature parameters are the control conditions for the reverse processing; the stimulation parameter determination model includes a decoding module, which is used to decode the predicted transcranial stimulation parameter encoding vector output by the reverse module to obtain the predicted transcranial stimulation parameters.

[0129] Based on any of the above embodiments, the skull structure model determination module is specifically used for:

[0130] A standard skull structure model is determined, which is constructed based on the average brain structure of a population;

[0131] Alternatively, target structural image data can be determined, and a corresponding individualized skull structure model can be constructed based on the target structural image data.

[0132] Figure 3An example is a schematic diagram of the structure of an electronic device, such as... Figure 3 As shown, the electronic device may include a processor 310, a communications interface 320, a memory 330, and a communication bus 340, wherein the processor 310, communications interface 320, and memory 330 communicate with each other via the communication bus 340. The processor 310 can call logical instructions in the memory 330 to execute a multi-channel transcranial stimulation parameter optimization method based on a generative model. This method includes: determining target field feature parameters; inputting the target field feature parameters into a stimulation parameter determination model to obtain transcranial stimulation parameters output by the stimulation parameter determination model, wherein the transcranial stimulation parameters are used to realize the target field corresponding to the target field feature parameters; the stimulation parameter determination model is trained based on sample field feature parameters and multi-channel sample transcranial stimulation parameters.

[0133] Furthermore, the logical instructions in the aforementioned memory 330 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, essentially, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. 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.

[0134] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the multi-channel transcranial stimulation parameter optimization method based on the generative model provided by the above methods. The method includes: determining target field feature parameters; inputting the target field feature parameters into a stimulation parameter determination model to obtain transcranial stimulation parameters output by the stimulation parameter determination model, wherein the transcranial stimulation parameters are used to realize the target field corresponding to the target field feature parameters; the stimulation parameter determination model is trained based on sample field feature parameters and multi-channel sample transcranial stimulation parameters.

[0135] In another aspect, the present invention also provides a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements a method for optimizing multi-channel transcranial stimulation parameters based on a generative model provided by the methods described above. This method includes: determining target field feature parameters; inputting the target field feature parameters into a stimulation parameter determination model to obtain transcranial stimulation parameters output by the stimulation parameter determination model, wherein the transcranial stimulation parameters are used to realize the target field corresponding to the target field feature parameters; and the stimulation parameter determination model is trained based on sample field feature parameters and multi-channel sample transcranial stimulation parameters.

[0136] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0137] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0138] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for optimizing multi-channel transcranial stimulation parameters based on a generative model, characterized in that, include: Determine the target field characteristic parameters; wherein, the target field characteristic parameters are used to characterize the stimulation effect produced by the transcranial stimulation parameters; The target field feature parameters are input into the stimulation parameter determination model to obtain the multi-channel transcranial stimulation parameters output by the stimulation parameter determination model. The transcranial stimulation parameters are used to realize the target field corresponding to the target field feature parameters. The stimulation parameter determination model is trained based on the sample field feature parameters and the multi-channel sample transcranial stimulation parameters. Training based on sample field feature parameters and multi-channel sample transcranial stimulation parameters includes: The transcranial stimulation parameters of the sample are input into the forward module of the original stimulation parameter determination model to obtain Gaussian noise data output by the forward module; the Gaussian noise data is obtained by forward diffusion of gradually adding noise to the transcranial stimulation parameters. The Gaussian noise data is input into the inverse module of the original stimulation parameter determination model to obtain the predicted transcranial stimulation parameters recovered by the inverse module; the predicted transcranial stimulation parameters are obtained by inverse processing of the Gaussian noise data; Determine the loss function between the predicted transcranial stimulation parameters and the sample transcranial stimulation parameters, and determine the parameters of the model by adjusting the original stimulation parameters by minimizing the loss function; Repeat the step of inputting the transcranial stimulation parameters of the sample into the forward module of the original stimulation parameter determination model until the original stimulation parameter determination model converges to obtain the stimulation parameter determination model. The sample field feature parameters are the control conditions for reverse processing.

2. The method for optimizing multi-channel transcranial stimulation parameters based on a generative model according to claim 1, characterized in that, Prior to training based on sample transcranial stimulation parameters and multi-channel sample transcranial stimulation parameters, the method further includes: Determine the skull structure model; Based on the aforementioned skull structure model, multiple sets of multi-channel transcranial stimulation parameters are determined; each set of multi-channel transcranial stimulation parameters is used to achieve a specific transcranial stimulation. Simulation calculations are performed based on the transcranial stimulation parameters of the samples and the skull structure model to obtain the corresponding sample field characteristic parameters in the cerebral cortex. The sample field characteristic parameters include the sample field intensity distribution and / or sample field distribution characteristic parameters.

3. The method for optimizing multi-channel transcranial stimulation parameters based on a generative model according to claim 2, characterized in that, After obtaining the corresponding sample field feature parameters in the cerebral cortex, the method further includes: The sample field feature parameters and the sample transcranial stimulation parameters are input into the pre-training matching module of the original stimulation parameter determination model to obtain the sample transcranial stimulation parameter encoding vector output by the pre-training matching module. The encoded vector is input into the original stimulus parameters to determine the forward module of the model, and Gaussian noise data output by the forward module is obtained; the Gaussian noise data is obtained by forward diffusion of the encoded vector with noise added stepwise. The Gaussian noise data is input into the inverse module of the original stimulation parameter determination model to obtain the predicted transcranial stimulation parameter encoding vector recovered by the inverse module; the predicted transcranial stimulation parameter encoding vector is obtained by inverse processing of the Gaussian noise data; Determine the loss function for the predicted transcranial stimulation parameter encoding vector and the sample transcranial stimulation parameter encoding vector, and determine the model parameters by adjusting the original stimulation parameters by minimizing the loss function; Repeat the steps of inputting the sample field feature parameters and the sample transcranial stimulation parameters into the pre-training matching module of the original stimulation parameter determination model until the original stimulation parameter determination model converges to obtain the stimulation parameter determination model. The sample field feature parameters are the control conditions for the reverse processing; the stimulation parameter determination model includes a decoding module, which is used to decode the predicted transcranial stimulation parameter encoding vector output by the reverse module to obtain the predicted transcranial stimulation parameters.

4. The method for optimizing multi-channel transcranial stimulation parameters based on a generative model according to claim 2, characterized in that, The determination of the skull structure model includes: A standard skull structure model is determined, which is constructed based on the average brain structure of a population; or Identify the target structural image data, and construct the corresponding individualized skull structure model based on the target structural image data.

5. A multi-channel transcranial stimulation parameter optimization device based on a generative model, characterized in that, include: The feature parameter determination module is used to determine the target field feature parameters; wherein, the target field feature parameters are used to characterize the stimulation effect produced by the transcranial stimulation parameters; The stimulation parameter determination module is used to input the target field feature parameters into the stimulation parameter determination model to obtain multi-channel transcranial stimulation parameters output by the stimulation parameter determination model. The transcranial stimulation parameters are used to realize the target field corresponding to the target field feature parameters. The stimulation parameter determination model is trained based on the sample field feature parameters and the multi-channel sample transcranial stimulation parameters. The model training module is used for: The transcranial stimulation parameters of the sample are input into the forward module of the original stimulation parameter determination model to obtain Gaussian noise data output by the forward module; the Gaussian noise data is obtained by forward diffusion of gradually adding noise to the transcranial stimulation parameters. The Gaussian noise data is input into the inverse module of the original stimulation parameter determination model to obtain the predicted transcranial stimulation parameters recovered by the inverse module; the predicted transcranial stimulation parameters are obtained by inverse processing of the Gaussian noise data; Determine the loss function between the predicted transcranial stimulation parameters and the sample transcranial stimulation parameters, and determine the parameters of the model by adjusting the original stimulation parameters by minimizing the loss function; Repeat the step of inputting the transcranial stimulation parameters of the sample into the forward module of the original stimulation parameter determination model until the original stimulation parameter determination model converges to obtain the stimulation parameter determination model. The sample field feature parameters are the control conditions for reverse processing.

6. The multi-channel transcranial stimulation parameter optimization device based on a generative model according to claim 5, characterized in that, Also includes: The skull structure model determination module is used to determine the skull structure model; The sample stimulation parameter determination module is used to determine multiple sets of multi-channel transcranial stimulation parameters based on the skull structure model; each set of multi-channel transcranial stimulation parameters is used to achieve a transcranial stimulation. The sample feature parameter determination module is used to perform simulation calculations based on the transcranial stimulation parameters of the sample and the skull structure model to obtain the corresponding sample field feature parameters in the cerebral cortex. The field feature parameters include the sample field strength distribution and / or sample field distribution feature parameters.

7. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the multi-channel transcranial stimulation parameter optimization method based on a generative model as described in any one of claims 1 to 4; the electronic device includes a transcranial stimulation device.

8. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the multi-channel transcranial stimulation parameter optimization method based on a generative model as described in any one of claims 1 to 4.

9. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the multi-channel transcranial stimulation parameter optimization method based on a generative model as described in any one of claims 1 to 4.

Images