Transcranial magnetic stimulation method and device based on digital twin data-mechanism driving

Abstract

The invention discloses a transcranial magnetic stimulation method and device based on digital twin data-mechanism driving. Synchronous evolution of the brain function model and real brain state information is realized by utilizing a domain self-adaption-based brain function model synchronous evolution method, so that lagging perception of brain state change information of a patient in a treatment process is reduced, coil parameters are adjusted according to a treatment effect evaluation result, mechanism-driven closed-loop control is realized, and the treatment effect of the patient is improved. The dynamic simulation result is used for virtual rehearsal of the stimulation scheme, so that closed-loop treatment strategy optimization of virtual-real bidirectional feedback is realized, and the technical problems that the open-loop targeted stimulation method has hysteresis quality on treatment effect evaluation and the stimulation effect cannot be optimized and adjusted in time are solved; besides, spatial position information registration of the virtual head model and the head of the patient is realized by using a real-time target point registration method based on visual servo, so that closed-loop interaction of virtual and real spatial position information of a target point is realized, and it is ensured that each pulse stimulation can accurately target a stimulation point.

Description

Technical Field

[0001] The present invention relates to the field of biomedical engineering technology, and in particular to a transcranial magnetic stimulation method and device driven by digital twin data-mechanism. Background Art

[0002] Neuromodulation technology is a method of using electrical stimulation or chemical reactions to exert neuromodulatory effects on the nervous system (central nervous system, peripheral nervous system and autonomic nervous system), either to excite or inhibit it, and to a certain extent can improve neurological symptoms. Current neuromodulation technologies mainly include non-invasive methods such as transcranial magnetic stimulation (TMS), transcranial direct current stimulation (tDCS) and focused ultrasound stimulation (FUS), as well as invasive methods such as electrode deep brain stimulation (DBS) and infrared nerve stimulation (INS). The emergence of the above neuromodulation technologies will bring new directions for rehabilitation for patients with nerve damage.

[0003] Transcranial magnetic stimulation (TMS) is a non-invasive neuromodulation technology. It generates a magnetic field perpendicular to the plane of the coil through instantaneous, high-voltage pulses generated by a magnetic coil placed on the scalp. It acts on brain tissue and generates induced currents to depolarize nerve cells and generate evoked potentials. This technology can be used to evaluate neurophysiological conduction pathways and has been used in neurorehabilitation treatment for diseases such as stroke, depression, and autism.

[0004] The traditional transcranial magnetic stimulation process uses an open-loop targeted stimulation method, which has a lag in the evaluation of treatment effects, ignores real-time brain status monitoring, and cannot make timely optimization adjustments to the stimulation effects. At the same time, since the patient's head will inevitably move during the treatment process, and the traditional transcranial magnetic therapy method uses a fixed device to prevent the patient's head from moving, it is impossible to keep the patient's head absolutely still during the stimulation process, resulting in the inability of pulse stimulation to accurately target the stimulation point, resulting in unsatisfactory transcranial magnetic stimulation treatment effects. Summary of the invention

[0005] The technical problem to be solved by the present invention is to provide a transcranial magnetic stimulation method and device driven by digital twin data-mechanism, which can effectively solve the technical problems that the open-loop targeted stimulation method has a lag in the evaluation of the treatment effect and cannot make timely optimization adjustments to the stimulation effect, and solve the technical problem that the open-loop targeted stimulation method cannot keep the patient's head absolutely still, resulting in the inability to accurately target the stimulation point with pulse stimulation.

[0006] In order to solve the above technical problems, the present invention adopts the following technical solutions:

[0007] A transcranial magnetic stimulation method based on digital twin data-mechanism drive, comprising the following steps: initializing spatial position information and coil parameters of a virtual head model; establishing a virtual coil in a virtual space and performing electromagnetic field simulation, and using a simulated grid search algorithm to find the optimal stimulation position of the coil for transcranial magnetic stimulation in the virtual space; controlling a therapeutic apparatus according to the coil parameters and the optimal stimulation position of the coil to perform transcranial magnetic stimulation treatment on a patient; obtaining real-time brain state data of the patient and obtaining a head image of the patient; based on the acquired real-time brain state data of the patient, using a domain-adaptive synchronous evolution method of a brain function model to calculate current virtual brain state information, and updating the current virtual brain state information to the brain function model; based on the acquired head image of the patient, using a real-time target registration method based on visual servoing to calculate the current spatial position information of the virtual head model; evaluating the treatment effect according to the real-time brain state data of the patient and the current virtual brain state information, adjusting the coil parameters according to the evaluation results, and updating the spatial position information of the virtual head model; repeating steps S20 to S70 until the transcranial magnetic stimulation treatment is completed;

[0008] The specific steps of calculating the current virtual brain state information by using the domain-adaptive brain function model synchronous evolution method include: based on the acquired real-time brain state data of the patient, introducing the incremental learning theory, combining the virtual-real state mapping mechanism constructed by domain adversarial training, updating the dynamic data by the ENTM method, iterating the brain function model and its mechanism model, and obtaining the current virtual brain state information;

[0009] The specific steps of calculating the current spatial position information of the virtual head model using the real-time target registration method based on visual servoing include: using a deep network framework with a variable number of trained facial feature points to extract a virtual facial feature point cloud and a patient's facial feature point cloud from the virtual head model and the patient's head image respectively; using a point cloud matching algorithm based on a Gaussian mixture model to minimize the gap between the virtual facial feature point cloud set and the patient's facial feature point cloud set, and combining the EM method to iteratively solve the mixed Gaussian problem to achieve spatial position information registration of the virtual head model and the patient's head, thereby obtaining the current spatial position information of the virtual head model.

[0010] A transcranial magnetic stimulation device driven by digital twin data-mechanism includes: an initialization module for initializing the spatial position information and coil parameters of a virtual head model; a virtual simulation module for establishing a virtual coil in a virtual space and performing electromagnetic field simulation, and using a simulated grid search algorithm to find the optimal stimulation position of the coil for transcranial magnetic stimulation in the virtual space; a treatment control module for controlling the treatment device according to the coil parameters and the optimal stimulation position of the coil to perform transcranial magnetic stimulation treatment on the patient; a data acquisition module for acquiring real-time brain state data of the patient and acquiring a head image of the patient; a first calculation module for using a domain-adaptive brain function algorithm based on the acquired real-time brain state data of the patient The method for synchronously evolving the energy model calculates the current virtual brain state information, and updates the current virtual brain state information to the brain function model; the second calculation module is used to calculate the current spatial position information of the virtual head model based on the acquired patient head image and the real-time target registration method based on visual servoing; the parameter adjustment module is used to evaluate the treatment according to the patient's real-time brain state data and the current virtual brain state information, and adjust the coil parameters according to the evaluation results, and update the spatial position information of the virtual head model; the repeated calling module is used to cyclically repeatedly call the virtual simulation module, the treatment control module, the data acquisition module, the first calculation module, the second calculation module, and the parameter adjustment module until the transcranial magnetic stimulation treatment is completed;

[0011] The specific steps of calculating the current virtual brain state information by using the domain-adaptive brain function model synchronous evolution method include: based on the acquired real-time brain state data of the patient, introducing the incremental learning theory, combining the virtual-real state mapping mechanism constructed by domain adversarial training, updating the dynamic data by the ENTM method, iterating the brain function model and its mechanism model, and obtaining the current virtual brain state information;

[0012] The specific steps of calculating the current spatial position information of the virtual head model using the real-time target registration method based on visual servoing include: using a deep network framework with a variable number of trained facial feature points to extract a virtual facial feature point cloud and a patient's facial feature point cloud from the virtual head model and the patient's head image respectively; using a point cloud matching algorithm based on a Gaussian mixture model to minimize the gap between the virtual facial feature point cloud set and the patient's facial feature point cloud set, and combining the EM method to iteratively solve the mixed Gaussian problem to achieve spatial position information registration of the virtual head model and the patient's head, thereby obtaining the current spatial position information of the virtual head model.

[0013] The beneficial technical effect of the present invention is that the digital twin data-mechanism driven transcranial magnetic stimulation method of the present invention is based on the patient's real-time brain state data, and uses the domain-adaptive brain function model synchronous evolution method to calculate the current virtual brain state information, so as to achieve the synchronous evolution of the brain function model and the real brain state information, thereby reducing the lagging perception of the patient's brain state change information during the treatment process, and then evaluating the treatment effect according to the patient's real-time brain state data and the current virtual brain state information, and adjusting the coil parameters according to the evaluation results to achieve mechanism-driven closed-loop control, and using the dynamic simulation results to perform a virtual preview of the stimulation scheme to achieve virtual-real two-way feedback The closed-loop treatment strategy is optimized, which solves the technical problem that the open-loop targeted stimulation method has a lag in the evaluation of the treatment effect and cannot make timely optimization adjustments to the stimulation effect. In addition, by obtaining the patient's head image and using the real-time target registration method based on visual servoing to calculate the current spatial position information of the virtual head model, the spatial position information of the virtual head model and the patient's head are registered, thereby realizing the closed-loop interaction of the virtual and real spatial position information of the target, ensuring that each pulse stimulation can accurately target the stimulation point, solving the technical problem that the open-loop targeted stimulation method cannot keep the patient's head absolutely still, resulting in the inability to accurately target the stimulation point. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 It is a schematic diagram of the process of the digital twin data-mechanism driven transcranial magnetic stimulation method of the present invention;

[0015] Figure 2 A schematic diagram of a process for calculating current virtual brain state information using a domain-adaptive brain function model synchronous evolution method of the present invention;

[0016] Figure 3 A schematic diagram of a flow chart for constructing a virtual-real state mapping mechanism of the present invention;

[0017] Figure 4 A schematic diagram of a flow chart of calculating the current spatial position information of a virtual head model using a real-time target registration method based on visual servoing according to the present invention;

[0018] Figure 5 A schematic diagram of the process of training a deep network framework with a variable number of facial feature points according to the present invention;

[0019] Figure 6 This is a structural block diagram of the transcranial magnetic stimulation device driven by digital twin data-mechanism of the present invention. DETAILED DESCRIPTION

[0020] In order to enable those skilled in the art to more clearly understand the objectives, technical solutions and advantages of the present invention, the present invention is further described below in conjunction with the accompanying drawings and embodiments.

[0021] like Figure 1 As shown, in one embodiment of the present invention, the transcranial magnetic stimulation method based on digital twin data-mechanism driven includes steps S10 to S80:

[0022] S10, initializing the spatial position information and coil parameters of the virtual head model.

[0023] S20. Establish a virtual coil in the virtual space and perform electromagnetic field simulation, and use a simulated grid search algorithm to find the optimal stimulation position of the coil for transcranial magnetic stimulation in the virtual space.

[0024] In the specific implementation process, by performing finite element simulation of the brain electromagnetic field in the virtual space, the influence of singular points in the finite element electromagnetic field simulation is eliminated, the simulation accuracy of the brain electromagnetic field is improved, and the simulation treatment is closer to the real transcranial magnetic brain stimulation. Among them, the position of the virtual head model in the virtual space is determined by the spatial position information of the virtual head model, and the position of the virtual head model in the virtual space corresponds to the real spatial position of the patient's head; the power, pulse, current change rate and other parameters of the virtual coil are determined by the coil parameters. Using the simulation grid search algorithm, the position of the virtual coil when the target brain area of ​​the virtual head model is maximally covered by the electromagnetic field is found in the virtual space, and this position is the optimal stimulation position of the coil for transcranial magnetic stimulation.

[0025] S30. Control the therapeutic device according to the coil parameters and the optimal stimulation position of the coil to perform transcranial magnetic stimulation treatment on the patient.

[0026] In one embodiment of the present invention, the step S30 further includes the following steps:

[0027] S31. Based on the current position of the coil and the optimal stimulation position of the coil, the coil trajectory planning path is calculated using the reinforcement learning algorithm of the joint spatiotemporal state vector to ensure the safety of the movement of the transcranial magnetic coil.

[0028] S32. Control the power, pulse waveform and current change rate of the coil of the therapeutic device according to the coil parameters to perform transcranial magnetic stimulation treatment on the patient; control the movement of the coil according to the coil trajectory planning path until the coil reaches the optimal stimulation position to ensure that each pulse stimulation can accurately target the stimulation point.

[0029] S40, obtaining the patient's real-time brain status data and the patient's head image.

[0030] When the therapeutic device is working to perform transcranial magnetic stimulation treatment on patients, the patient's real-time brain state data is collected through sensors as feedback data for transcranial magnetic stimulation. At the same time, the patient's head image is collected through a camera as reference data for subsequent adjustment of the coil position.

[0031] S50. Based on the acquired real-time brain state data of the patient, the current virtual brain state information is calculated using the domain-adaptive brain function model synchronous evolution method, and the current virtual brain state information is updated to the brain function model.

[0032] like Figure 2 As shown, the specific steps of calculating the current virtual brain state information using the domain-adaptive brain function model synchronous evolution method include: based on the acquired real-time brain state data of the patient, introducing incremental learning theory to monitor the time-varying characteristics of the brain state, combining the virtual-real state mapping mechanism constructed through domain adversarial training, updating dynamic data through the ENTM (Evolutionary neural turing machine) method, preventing the concept drift problem caused by the dynamic update of brain state data, avoiding the phenomenon of catastrophic forgetting of the brain function model and its mechanism model, iterating the brain function model and its mechanism model, realizing the synchronous evolution of the brain function model and the real brain state information, thereby obtaining the current virtual brain state information, and updating the current virtual brain state information to the brain function model.

[0033] like Figure 3 As shown, in one embodiment of the present invention, the specific steps of constructing a virtual-real state mapping mechanism include:

[0034] S51. Construct a feature mapping network based on the brain state information set, and design a feature extractor according to the brain state information, and map the feature data generated by the feature extractor to a feature space, wherein the feature space includes a real domain feature distribution and a virtual domain feature distribution.

[0035] S52, establishing a label classification network based on real-domain features and their distribution, and classifying real-domain data in the feature space in combination with an optimized residual network;

[0036] S53, using the virtual domain features and their distribution to set up a domain discrimination network, using MMD (Maximum Mean Discrepancy, MMD for short) as a metric to construct a discriminator, which is used to determine whether the feature data comes from the virtual domain or the real domain;

[0037] S54. Based on the feature extractor and the discriminator, domain adversarial training is carried out to achieve alignment of the feature space, derive the state mapping relationship between the virtual domain and the real domain, and construct a mapping mechanism between the brain functional connection network and the real brain, that is, to construct a virtual-real state mapping mechanism.

[0038] In this embodiment, a feature extractor is designed according to brain state information, so that the feature extractor can generate feature data that is conducive to maximizing the recognition of the label classification network for real-domain data and minimizing the resolution of the domain discrimination network for the brain state information set; by using an optimized residual network (ResNet) combined with a label classification network to classify the real-domain data in the feature space, the network degradation problem caused by too many hidden layers in the deep neural network during feature classification can be solved, and the real-domain data in the feature space can be correctly classified as much as possible; by constructing a virtual-real state mapping mechanism and establishing a clear consistency between virtual and reality, the gap between virtual and reality can be narrowed.

[0039] S60 : Based on the acquired patient head image, calculate the current spatial position information of the virtual head model using a real-time target registration method based on visual servoing.

[0040] like Figure 4 As shown, the specific steps of calculating the current spatial position information of the virtual head model using the real-time target registration method based on visual servoing include:

[0041] S61. Utilizing a trained deep network framework with a variable number of facial feature points, extract a virtual facial feature point cloud and a patient facial feature point cloud from the virtual head model and the patient head image, respectively.

[0042] S62. A point cloud matching algorithm based on a Gaussian mixture model (GMM) is used to minimize the gap between the virtual facial feature point cloud set and the patient's facial feature point cloud set. The EM (Expectation-Maximum) method is used to iteratively solve the mixed Gaussian problem to achieve spatial position information alignment between the virtual head model and the patient's head, and the current spatial position information of the virtual head model is obtained.

[0043] like Figure 5 As shown, in one embodiment of the present invention, the specific steps of training a deep network framework with a variable number of facial feature points include:

[0044] S601, extracting facial feature points of a face image through a face recognition algorithm based on deep learning, wherein the facial feature points include contour feature points and internal feature points. Wherein, the face image comes from a high-quality public data set, and by finely annotating the facial feature points of the face image, extracting contour feature points including jaw contour and forehead contour, and extracting internal feature points including nose, eyes, mouth, eyebrows, etc., so as to achieve accurate recognition of key facial features, thereby realizing non-calibrated visual positioning.

[0045] S602: Use a pose network (PoseNet) to estimate the face pose and obtain facial pose estimation information (quaternion of facial pose).

[0046] S603, constructing a dendritic structure using facial feature points, combining facial posture estimation information, and using a posture-conditioned dendritic convolutional neural network based on the Bayesian formula to separate the three-dimensional posture from the face image, which can reduce the positioning error when adjusting the posture. The specific steps of constructing the dendritic structure include: constructing a dendritic structure with the nose tip feature point as the root node, and placing the remaining facial feature points in order under the tree structure.

[0047] S604: Using the face image and facial feature points after stripping the three-dimensional posture as training data, a convolutional neural network is trained to obtain a deep network framework with a variable number of facial feature points.

[0048] S70, evaluating the treatment effect according to the patient's real-time brain state data and the current virtual brain state information, adjusting the coil parameters according to the evaluation results, and updating the spatial position information of the virtual head model.

[0049] In the specific implementation process, the current virtual brain state information calculated by step S50 is compared with the patient's real-time brain state data to evaluate the treatment effect, and the coil parameters are adjusted according to the evaluation results. For example, the power, pulse waveform, and current change rate of the coil of the therapeutic device are adjusted to optimize the stimulation intensity level of the targeted area in the target brain area; at the same time, the spatial position information of the virtual head model is updated to prepare for the next stage of simulation treatment.

[0050] S80, determine whether to end the transcranial magnetic stimulation treatment. If so, end the entire process of the transcranial magnetic stimulation method based on digital twin data-mechanism driven; if not, return to step S20 and execute steps S20 to S70 again. By repeatedly executing steps S20 to S70, the patient's real-time brain state data is continuously compared with the current virtual brain state information, thereby continuously optimizing the target stimulation strategy, thereby realizing the optimization of the closed-loop treatment strategy of virtual and real bidirectional feedback, and improving the effectiveness of the treatment.

[0051] The digital twin data-mechanism-driven transcranial magnetic stimulation method of the present invention is based on the patient's real-time brain state data, and uses the domain-adaptive brain function model synchronous evolution method to calculate the current virtual brain state information, so as to achieve the synchronous evolution of the brain function model and the real brain state information, thereby reducing the lagging perception of the patient's brain state change information during the treatment process, and then evaluates the treatment effect according to the patient's real-time brain state data and the current virtual brain state information, and adjusts the coil parameters according to the evaluation results to achieve mechanism-driven closed-loop control, and uses the dynamic simulation results to perform a virtual preview of the stimulation scheme, so as to achieve a closed-loop treatment strategy with virtual and real two-way feedback. The method is slightly optimized to solve the technical problem that the open-loop targeted stimulation method has a lag in the evaluation of the treatment effect and cannot make timely optimization adjustments to the stimulation effect; in addition, by obtaining the patient's head image and utilizing the real-time target registration method based on visual servoing to calculate the current spatial position information of the virtual head model, the spatial position information of the virtual head model and the patient's head are registered, thereby realizing a closed-loop interaction of the target's virtual and real spatial position information, ensuring that each pulse stimulation can accurately target the stimulation point, solving the technical problem that the open-loop targeted stimulation method cannot accurately target the stimulation point due to the inability to keep the patient's head absolutely still.

[0052] Based on the above-mentioned digital twin data-mechanism driven transcranial magnetic stimulation method, the present invention provides a digital twin data-mechanism driven transcranial magnetic stimulation device. The digital twin data-mechanism driven transcranial magnetic stimulation device can perform the following steps: Figure 1 The digital twin data-mechanism driven transcranial magnetic stimulation method in the illustrated embodiment has the functional modules and beneficial effects corresponding to the execution method.

[0053] like Figure 6 As shown, in one embodiment of the present invention, a transcranial magnetic stimulation device driven by digital twin data-mechanism includes an initialization module 10, a virtual simulation module 20, a treatment control module 30, a data acquisition module 40, a first calculation module 50, a second calculation module 60, a parameter adjustment module 70 and a repeated call module 80.

[0054] Initialization module 10 is used to initialize the spatial position information and coil parameters of the virtual head model, that is, to perform the following steps: Figure 1 Step S10 in the digital twin data-mechanism driven transcranial magnetic stimulation method in the illustrated embodiment.

[0055] The virtual simulation module 20 is used to establish a virtual coil in a virtual space and perform electromagnetic field simulation, and use a simulated grid search algorithm to find the optimal stimulation position of the coil for transcranial magnetic stimulation in the virtual space, that is, to perform the following steps: Figure 1 Step S20 in the digital twin data-mechanism driven transcranial magnetic stimulation method in the illustrated embodiment.

[0056] The treatment control module 30 is used to control the treatment device according to the coil parameters and the coil optimal stimulation posture to perform transcranial magnetic stimulation treatment on the patient, that is, to perform the following steps: Figure 1 Step S30 in the digital twin data-mechanism driven transcranial magnetic stimulation method in the illustrated embodiment.

[0057] The data acquisition module 40 is used to acquire the patient's real-time brain state data and the patient's head image, that is, to perform the following operations: Figure 1 Step S40 in the digital twin data-mechanism driven transcranial magnetic stimulation method in the illustrated embodiment.

[0058] The first calculation module 50 is used to calculate the current virtual brain state information based on the acquired real-time brain state data of the patient by using the synchronous evolution method of the brain function model based on domain adaptation, and update the current virtual brain state information to the brain function model, that is, to perform the following steps: Figure 1 Step S50 in the digital twin data-mechanism driven transcranial magnetic stimulation method in the illustrated embodiment. The specific steps of calculating the current virtual brain state information using the domain-adaptive brain function model synchronous evolution method include: based on the acquired real-time brain state data of the patient, introducing incremental learning theory, combining the virtual-real state mapping mechanism constructed by domain adversarial training, updating dynamic data through the ENTM method, iterating the brain function model and its mechanism model, and obtaining the current virtual brain state information.

[0059] The second calculation module 60 is used to calculate the current spatial position information of the virtual head model based on the acquired patient head image using a real-time target registration method based on visual servoing, that is, to perform the following steps: Figure 1 Step S60 in the digital twin data-mechanism driven transcranial magnetic stimulation method in the illustrated embodiment. The specific steps of calculating the current spatial position information of the virtual head model using the real-time target registration method based on visual servoing include: using a deep network framework with a variable number of trained facial feature points to extract virtual facial feature point clouds and patient facial feature point clouds from the virtual head model and the patient's head image, respectively; using a point cloud matching algorithm based on a Gaussian mixture model to minimize the gap between the virtual facial feature point cloud set and the patient's facial feature point cloud set, and combining the EM method to iteratively solve the mixed Gaussian problem to achieve spatial position information registration of the virtual head model and the patient's head, and obtain the current spatial position information of the virtual head model.

[0060] The parameter adjustment module 70 is used to evaluate the treatment according to the patient's real-time brain state data and the current virtual brain state information, and adjust the coil parameters according to the evaluation results, and update the spatial position information of the virtual head model, that is, to perform the following steps: Figure 1 Step S70 in the digital twin data-mechanism driven transcranial magnetic stimulation method in the illustrated embodiment.

[0061] The repeated calling module 80 is used to cyclically and repeatedly call the virtual simulation module, the treatment control module, the data acquisition module, the first calculation module, the second calculation module, and the parameter adjustment module until the transcranial magnetic stimulation treatment is completed.

[0062] The digital twin data-mechanism-driven transcranial magnetic stimulation device of the present invention calculates the current virtual brain state information based on the patient's real-time brain state data and the domain-adaptive brain function model synchronous evolution method, thereby realizing the synchronous evolution of the brain function model and the real brain state information, thereby reducing the lagging perception of the patient's brain state change information during the treatment process, and then evaluates the treatment effect based on the patient's real-time brain state data and the current virtual brain state information, and adjusts the coil parameters according to the evaluation results to realize mechanism-driven closed-loop control, and uses the dynamic simulation results to perform a virtual preview of the stimulation scheme, thereby realizing a closed-loop treatment strategy with virtual and real two-way feedback. The method is slightly optimized to solve the technical problem that the open-loop targeted stimulation method has a lag in the evaluation of the treatment effect and cannot make timely optimization adjustments to the stimulation effect; in addition, by obtaining the patient's head image and utilizing the real-time target registration method based on visual servoing to calculate the current spatial position information of the virtual head model, the spatial position information of the virtual head model and the patient's head are registered, thereby realizing a closed-loop interaction of the target's virtual and real spatial position information, ensuring that each pulse stimulation can accurately target the stimulation point, solving the technical problem that the open-loop targeted stimulation method cannot accurately target the stimulation point due to the inability to keep the patient's head absolutely still.

[0063] The above description is only a preferred embodiment of the present invention, and does not limit the present invention in any form. Those skilled in the art can make various equivalent changes and improvements based on the above embodiments, and all equivalent changes or modifications made within the scope of the claims should fall within the protection scope of the present invention.

Claims

1. A transcranial magnetic stimulation method based on digital twin data-mechanism driven, characterized in that: The steps include: S10, initializing the spatial position information and coil parameters of the virtual head model; S20, establishing a virtual coil in a virtual space and performing electromagnetic field simulation, and using a simulated grid search algorithm to find the optimal stimulation position of the coil for transcranial magnetic stimulation in the virtual space; S30, controlling the therapeutic apparatus according to the coil parameters and the optimal stimulation position of the coil to perform transcranial magnetic stimulation treatment on the patient; S40, obtaining real-time brain status data of the patient and obtaining a head image of the patient; S50, based on the acquired real-time brain state data of the patient, using a domain-adaptive brain function model synchronous evolution method to calculate current virtual brain state information, and updating the current virtual brain state information to the brain function model; S60, based on the acquired patient head image, using a real-time target registration method based on visual servoing to calculate the current spatial position information of the virtual head model; S70, evaluating the treatment effect according to the patient's real-time brain state data and the current virtual brain state information, adjusting the coil parameters according to the evaluation results, and updating the spatial position information of the virtual head model; Repeating steps S20 to S70 until the transcranial magnetic stimulation treatment is completed; The specific steps of calculating the current virtual brain state information by using the domain-adaptive brain function model synchronous evolution method include: based on the acquired real-time brain state data of the patient, introducing the incremental learning theory, combining the virtual-real state mapping mechanism constructed by domain adversarial training, updating the dynamic data by the ENTM method, iterating the brain function model and its mechanism model, and obtaining the current virtual brain state information; The specific steps of calculating the current spatial position information of the virtual head model using the real-time target registration method based on visual servoing include: Using the trained deep network framework with a variable number of facial feature points, the virtual facial feature point cloud and the patient's facial feature point cloud are extracted from the virtual head model and the patient's head image respectively; A point cloud matching algorithm based on a Gaussian mixture model is used to minimize the gap between the virtual facial feature point cloud set and the patient's facial feature point cloud set. The EM method is used to iteratively solve the mixed Gaussian problem to achieve spatial position information alignment between the virtual head model and the patient's head, and the current spatial position information of the virtual head model is obtained.

2. The digital twin data-mechanism driven transcranial magnetic stimulation method according to claim 1, characterized in that: The specific steps of building a virtual-real state mapping mechanism include: A feature mapping network is constructed based on the brain state information set, and a feature extractor is designed according to the brain state information. The feature data generated by the feature extractor is mapped to a feature space, which includes a real domain feature distribution and a virtual domain feature distribution. A label classification network is established based on real-domain features and their distribution, and the real-domain data in the feature space is classified using an optimized residual network. The domain discrimination network is set up using virtual domain features and their distribution, and the discriminator is constructed using MMD as the metric. The discriminator is used to determine whether the feature data comes from the virtual domain or the real domain. Domain adversarial training is carried out based on the feature extractor and the discriminator to achieve the alignment of the feature space, derive the state mapping relationship between the virtual domain and the real domain, and construct a virtual-real state mapping mechanism.

3. The digital twin data-mechanism driven transcranial magnetic stimulation method according to claim 1, characterized in that: The specific steps of training a deep network framework with a variable number of facial landmarks include: Extract facial feature points of a face image through a face recognition algorithm based on deep learning, where the facial feature points include contour feature points and internal feature points; The posture network is used to estimate the face posture and obtain the facial posture estimation information; The dendritic structure is constructed by using facial feature points, combined with the facial posture estimation information, and the posture-conditioned dendritic convolutional neural network based on the Bayesian formula is used to separate the three-dimensional posture from the face image; The face images and facial feature points after stripping the 3D posture are used as training data to train the convolutional neural network to obtain a deep network framework with a variable number of facial feature points.

4. The digital twin data-mechanism driven transcranial magnetic stimulation method according to claim 3, characterized in that: The specific steps of constructing the dendritic structure include: constructing the dendritic structure with the nose tip feature point as the root node, and placing the remaining facial feature points in order under the dendritic structure.

5. The digital twin data-mechanism driven transcranial magnetic stimulation method according to claim 1, characterized in that: The step S30 further comprises: S31, based on the current position of the coil and the optimal stimulation position of the coil, the coil trajectory planning path is calculated using a reinforcement learning algorithm of the joint spatiotemporal state vector; S32. Control the power, pulse waveform and current change rate of the coil of the therapeutic device according to the coil parameters, and control the movement of the coil according to the coil trajectory planning path until the coil reaches the optimal stimulation position.

6. A transcranial magnetic stimulation device driven by digital twin data-mechanism, characterized in that: Included are: An initialization module, used to initialize the spatial position information and coil parameters of the virtual head model; A virtual simulation module is used to establish a virtual coil in a virtual space and perform electromagnetic field simulation, and use a simulated grid search algorithm to find the optimal stimulation position of the coil for transcranial magnetic stimulation in the virtual space; A treatment control module, used to control the treatment device according to coil parameters and the optimal stimulation position of the coil, so as to perform transcranial magnetic stimulation treatment on the patient; A data acquisition module, used to acquire the patient's real-time brain status data and the patient's head image; A first calculation module is used to calculate the current virtual brain state information based on the acquired real-time brain state data of the patient by using the synchronous evolution method of the brain function model based on domain adaptation, and update the current virtual brain state information to the brain function model; A second calculation module is used to calculate the current spatial position information of the virtual head model based on the acquired patient head image by using a real-time target registration method based on visual servoing; A parameter adjustment module is used to evaluate the treatment based on the patient's real-time brain state data and the current virtual brain state information, adjust the coil parameters based on the evaluation results, and update the spatial position information of the virtual head model; A repeated calling module, used for repeatedly calling the virtual simulation module, the treatment control module, the data acquisition module, the first calculation module, the second calculation module, and the parameter adjustment module in a cyclic manner until the transcranial magnetic stimulation treatment is completed; The specific steps of calculating the current virtual brain state information by using the domain-adaptive brain function model synchronous evolution method include: based on the acquired real-time brain state data of the patient, introducing the incremental learning theory, combining the virtual-real state mapping mechanism constructed by domain adversarial training, updating the dynamic data by the ENTM method, iterating the brain function model and its mechanism model, and obtaining the current virtual brain state information; The specific steps of calculating the current spatial position information of the virtual head model using the real-time target registration method based on visual servoing include: Using the trained deep network framework with a variable number of facial feature points, the virtual facial feature point cloud and the patient's facial feature point cloud are extracted from the virtual head model and the patient's head image respectively; A point cloud matching algorithm based on a Gaussian mixture model is used to minimize the gap between the virtual facial feature point cloud set and the patient's facial feature point cloud set. The EM method is used to iteratively solve the mixed Gaussian problem to achieve spatial position information alignment between the virtual head model and the patient's head, and the current spatial position information of the virtual head model is obtained.

Images