AI intelligent regulation and control mental disease treatment system based on multi-mode microwaves

Abstract

The invention discloses an AI intelligent regulation and control mental disease treatment system based on multi-mode microwaves, and the system comprises a data collection module which collects multi-mode microwave data, and the multi-mode microwave data comprises basic microwave data, neural pathway activity state data and personalized physiological data; the optimization regulation and control module is used for extracting optimal regulation and control data in the multi-modal microwave data through an improved deep convolutional neural network; the dynamic strategy module is used for collecting historical microwave stimulation data, and learning and optimizing a nonlinear relationship between regulation and control data and microwave stimulation parameters through a hybrid transfer learning model to obtain optimal microwave stimulation parameters for specific brain region neural activities; and the treatment execution module is used for controlling microwave emission equipment based on the optimal microwave stimulation parameters, emitting microwaves to the corresponding brain region of the patient for treatment, generating a three-dimensional visual report of the treatment progress, dynamically adjusting model parameters, realizing real-time optimization of treatment parameters and solving the limitation of static parameters in a traditional method.

Description

Technical Field

[0001] The present invention relates to the technical field of microwave intelligent regulation, and particularly relates to an AI intelligent regulation mental disease treatment system based on multimodal microwave. Background Art

[0002] In the current application of mental disease treatment, microwave treatment, as a non-invasive neuromodulation technology, has been preliminarily applied to the intervention of diseases such as depression and anxiety. The occurrence of mental diseases is closely related to the abnormal activities of brain region neural networks (such as dysfunction of the prefrontal lobe and limbic system). However, there are significant differences in the neural pathway activity and brain region functional connections among different patients, and personalized stimulation parameters are required to restore neural balance.

[0003] In the existing microwave regulation mental disease treatment system, although microwave treatment has been applied, there are problems such as single stimulation parameters, lack of personalized and dynamic adjustment mechanisms, etc. Traditional microwave neuromodulation only relies on frequency and intensity, without considering brain region microstructure, neural pathway activity, patient physiological rhythm, and psychosocial factors, etc., making it difficult to accurately treat. Therefore, an AI intelligent regulation mental disease treatment system based on multimodal microwave is proposed here. Summary of the Invention

[0004] In order to overcome the above defects of the prior art and to achieve the above object, the present invention proposes the following technical solutions:

[0005] An AI intelligent regulation mental disease treatment system based on multimodal microwave, comprising:

[0006] A data acquisition module: collecting multimodal microwave data, where the multimodal microwave data includes basic microwave data, neural pathway activity state data, and individual physiological data;

[0007] An optimization regulation module: extracting the optimal regulation data in the multimodal microwave data through an improved deep convolutional neural network;

[0008] Among them, the improved deep convolutional neural network is implemented by adding a residual connection between two consecutive convolutional layers in the basic deep convolutional neural network and introducing a self-optimizing learning scalar during the residual connection process, and by using dilated convolution to replace ordinary convolution;

[0009] A dynamic strategy module: collecting historical microwave stimulation data, and obtaining the optimal microwave stimulation parameters for specific brain region neural activities by learning the non-linear relationship between the optimized regulation data and the microwave stimulation parameters through a hybrid transfer learning model;

[0010] Among them, the hybrid online transfer learning model is implemented by adding an improved online incremental learning process based on a two-channel neural network.

[0011] Treatment Execution Module: Based on the optimal microwave stimulation parameters, control the microwave emission device to emit microwaves to the corresponding brain regions of the patient for treatment and generate a three-dimensional visualization report of the treatment progress.

[0012] The implementation process of adding residual connections between two consecutive convolutional layers is as follows:

[0013] Take multi-modal microwave data as input data, and obtain the mapped outputs of two convolutional layers through operations of two consecutive convolutional layers;

[0014] Based on the mapped outputs of the two convolutional layers, add and fuse the multi-modal microwave data with the output after two convolutional layer mappings, and implement the residual connection process by combining individual data points and weight sets.

[0015] The specific implementation process of the improved deep convolutional neural network is as follows:

[0016] Preset a learning scalar and assign an initial learning scalar to it;

[0017] Introduce the initial learning scalar into the residual connection process to obtain the residual connection output with the initial learning scalar introduced;

[0018] Based on the residual connection output with the initial learning scalar introduced, preset a true output label, obtain the loss value between the residual connection output with the initial learning scalar introduced and the true output label through a regression loss function, and obtain the gradient of the loss value with respect to the self-optimizing learning scalar;

[0019] Based on the gradient of the loss value with respect to the self-optimizing learning scalar, use a Lookahead optimizer for updating, and obtain the self-optimizing learning scalar after updating to the preset maximum condition;

[0020] Introduce the self-optimizing learning scalar into the residual connection of the basic deep convolutional neural network, and use dilated convolution to replace ordinary convolution to obtain the improved deep convolutional neural network.

[0021] The historical microwave stimulation data includes microwave parameter frequencies, phases, treatment time points, and corresponding treatment effect evaluation data.

[0022] The training process of the hybrid transfer learning model is as follows:

[0023] Preset an initial hybrid transfer learning model, convert the optimal regulation data and historical microwave stimulation data into matrix forms and then perform dimensionality reduction, and input the dimensionality-reduced historical microwave stimulation data matrix and the current optimized regulation data into the preset initial hybrid transfer learning model respectively;

[0024] Based on the initial hybrid transfer learning model, add an improved online incremental learning process to obtain the final hybrid transfer learning model;

[0025] The final hybrid transfer learning model is trained using mini-batch gradient descent, and the loss value is calculated according to the loss function. The gradient of the loss function with respect to the model parameters is calculated through the backpropagation algorithm and the parameters are updated until the loss function converges to meet the preset conditions, obtaining a trained hybrid transfer learning model.

[0026] The initial hybrid transfer learning model is implemented based on a dual-channel neural network, and the dual-channel neural network includes a hybrid feature extractor, and the hybrid feature extractor includes an online channel and a transfer channel.

[0027] The implementation process of the improved online incremental learning process is as follows:

[0028] The improved online incremental learning is realized by introducing an optimized feedback gain coefficient based on a sliding window.

[0029] The process of obtaining the optimized feedback gain coefficient is as follows:

[0030] Set the initial feedback gain coefficient of the feedback gain coefficient based on prior knowledge;

[0031] Through a reinforcement learning idea, the adaptation process of the model to the optimal regulation data of the new input and the historical microwave stimulation data is regarded as a decision-making process, and the initial feedback gain coefficient is taken as the decision-making action. Rewards or punishments are given according to the processing effects of the model on the optimal regulation data of the new input and the historical microwave stimulation data under different initial feedback gain coefficient values until convergence, and finally an optimal optimized feedback gain coefficient is obtained.

[0032] The present invention has the following beneficial effects:

[0033] In the present invention, first, by introducing a residual connection between consecutive convolutional layers, the problem of gradient disappearance in deep networks is solved. At the same time, the Lookahead optimizer is used to dynamically adjust the learning scalar, adaptively balance the fusion weight of the input data and the convolutional features, and improve the complex feature extraction efficiency. Then, combined with dilated convolution to expand the receptive field, multi-scale spatial features are captured without increasing parameters, which is more suitable for heterogeneous feature processing of multi-modal data;

[0034] Secondly, by introducing an online channel (temporal recursive structure) to capture the dynamic changes of data, and a transfer channel (domain-invariant projection) to realize the knowledge transfer of historical data and current patient data, effectively utilize past treatment experience, reduce the training cost of new patients, and then based on the immediate feedback gain coefficient of the sliding window and reinforcement learning, respond to new data (such as changes in neural activity during treatment) in real time, dynamically adjust the model parameters, and realize the real-time optimization of treatment parameters, solving the limitations of static parameters in traditional methods;

[0035] Finally, through mini-batch gradient descent and backpropagation, the complex mapping between multi-modal data and microwave stimulation parameters (frequency, phase, treatment time) is accurately learned to support the generation of personalized parameters for specific brain regions. The treatment execution module combines electroencephalogram signals, hemoglobin concentration changes, and microwave parameters to generate a dynamic three-dimensional visualization report, intuitively presenting the association between treatment progress and physiological indicators, and assisting doctors in real-time evaluating the treatment effect and adjusting the treatment plan. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 FIG. is a system block diagram of an AI intelligent regulation mental disease treatment system based on multi-modal microwave proposed by the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0037] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0038] Embodiment 1

[0039] As Figure 1 shown, an AI intelligent regulation mental disease treatment system based on multi-modal microwave proposed by the present invention includes:

[0040] Data acquisition module: acquiring multi-modal microwave data, where the multi-modal microwave data includes basic microwave data, neural pathway activity state data, and individual physiological data;

[0041] Basic microwave data acquisition:

[0042] The basic microwave data is acquired through a microwave sensor, such as a superconducting nanowire single-photon detector, based on the principle that the superconducting nanowire generates an electrical signal after absorbing microwave photons and transitioning from the superconducting state to the normal state, and is represented as ;

[0043] Neural pathway activity state data acquisition:

[0044] The neural pathway activity state data is acquired through a functional near-infrared spectroscopy (fNIRS) device. Based on the absorption characteristics of biological tissues for near-infrared light, the neural activity is reflected by detecting the oxygenated and deoxygenated states of hemoglobin. Near-infrared light emitters and detectors are arranged on the scalp of the patient, and the change in hemoglobin concentration is calculated from the light intensity attenuation, and then the neural pathway activity state is obtained. The formula for calculating the change in hemoglobin concentration from the light intensity attenuation is expressed as:

[0045]

[0046] Among them, is the absorbance, is the incident light intensity, is the transmitted light intensity, is the molar extinction coefficient, is the optical path length, is the change in hemoglobin concentration;

[0047] The change in hemoglobin concentration is calculated from the light intensity attenuation, and then the data on the active state of the neural pathway is obtained ;

[0048] Individual physiological data collection:

[0049] For individual physiological data, an optoelectronic heart rate sensor is used. Based on photoplethysmography (PPG), when light irradiates the skin, the change in blood volume in blood vessels with the heartbeat causes changes in light absorption and scattering, and the detected pulse wave signal is converted into an electrical signal for acquisition, expressed as ;

[0050] Multimodal microwave data is expressed as ;

[0051] Multimodal microwave data has large differences in the scale of each feature dimension. For the collected basic microwave data , data on the active state of the neural pathway and individual physiological data are normalized to unify the data scale.

[0052] Optimization and regulation module: The optimal regulation data in multimodal microwave data is extracted through an improved deep convolutional neural network;

[0053] The process of obtaining the optimal regulation data is as follows:

[0054] The convolutional layer of the basic deep convolutional neural network is optimized and improved. Based on the basic deep neural network, it includes an input layer, two consecutive convolutional layers, a pooling layer, and an output layer;

[0055] A residual connection is added between the two consecutive convolutional layers in the basic deep convolutional neural network. The multimodal microwave data is received as input data through the input layer, and after being mapped by the first convolutional layer, it obtains , represents the basic operation of the first convolutional layer, is the output of the single-layer convolutional layer mapping, and then it is mapped by the second convolutional layer to obtain , is the basic operation of the second convolutional layer, is the output after being mapped by two convolutional layers;

[0056] Based on the output of the two-layer convolutional layer mapping, directly use the multimodal microwave data and directly add and fuse it with the output after convolutional layer mapping (in the multimodal microwave data ), the individual data points and weights. The output after residual connection is expressed as: , where represents the mapping relationship of the two-layer convolutional layer, is the individual data point in the multimodal microwave data , is the weight. Directly add the multimodal microwave data with the output after convolutional layer mapping , so that the network can learn the difference between the input and the output of the convolutional layer, making the network training process easier and extracting more complex spatial features in the multimodal microwave data more deeply;

[0057] Introduce a self-optimizing learning scalar during the residual connection process. The optimization process of the self-optimizing learning scalar is as follows:

[0058] Preset a learning scalar , and assign an initial learning scalar to the scalar (for example, 0.5), and optimize and update the initial learning scalar ;

[0059] Introduce the initial learning scalar into the residual connection process. For the input multimodal microwave data , obtain the convolutional layer output through the mapping relationship of the two-layer convolutional layer. At the same time, the input multimodal microwave data and the convolutional layer output participate in the residual connection calculation together. At this time, the residual connection output is:

[0060]

[0061] is the residual connection output with the initial learning scalar introduced;

[0062] This output value will continue to propagate forward in the network, participate in the calculation of subsequent layers, and the generation of the final output result;

[0063] Based on the residual connection output with the initial learning scalar introduced, preset a true output label, and obtain the loss value between the residual connection output with the initial learning scalar introduced and the true output label through a regression loss function. The loss value Backpropagate through the network using the backpropagation algorithm. During backpropagation, according to the chain rule of differentiation, calculate the loss value Regarding the self-optimizing learning scalar Calculate the gradient, denoted as This gradient value reflects The direction and degree of the impact of the change in on the loss value;

[0064] Specifically, the true output label is obtained by analyzing historical data. Collect a large amount of past historical data related to multimodal microwave therapy, including microwave stimulation parameters and feedback data from patients before and after treatment. Use statistical analysis methods to find the parameter combination with a strong correlation with the treatment effect as the true output label;

[0065] According to the calculated gradient Use a Lookahead optimizer for updating, and its update formula is:

[0066]

[0067] Where Is the learning rate of the Lookahead optimizer, used to update the value of the parameter The value of, Is the value of before update, and the value of after update is denoted as Value, Value is denoted as For the next forward propagation and subsequent calculations

[0068] As the training process progresses, repeat the above steps of forward propagation, loss calculation, backpropagation, and parameter update until the preset maximum condition is reached. For example:

[0069] Set a loss value threshold And the number of consecutive iterations When in consecutive Iterations, the decrease in the loss value L is less than At this time, the obtained Is the optimal learning scalar;

[0070] Specifically, different Affect the effect of the residual connection. When Approaches 0, in The contribution of to the output Becomes smaller, meaning that the network relies more on the self-mapping of the convolutional layer And focuses on the direct transformation and feature extraction of the input data. When Approaches 1, the residual connection is close to the traditional method, emphasizing the input multimodal microwave data And the output of the convolutional layer And the output of the convolutional layer Difference learning;

[0071] Meanwhile, in the convolutional layer part, dilated convolution is adopted to replace ordinary convolution. In dilated convolution, for the input data , the convolutional kernel , the dilation rate is , and the i-th output element is expressed as:

[0072]

[0073] where represents the j-th convolutional kernel, is the maximum number of convolutional kernels;

[0074] The preprocessed multi-modal microwave data is input into the optimized deep neural network. The improved convolutional layer adaptively optimizes spatial features of different scales, and then the pooling layer downsamples the output of the improved convolutional layer to retain the optimal features. The final formula is expressed as:

[0075]

[0076] A series of feature data finally output through the output layer is the optimal regulation data;

[0077] Specifically, after introducing the self-optimizing learning scalar , the deep neural network can automatically adjust the residual connection according to the characteristics of the input data and the requirements of the current task. For example, for some multi-modal microwave data with complex features that require in-depth transformation by the convolutional layer, the deep neural network can adjust to approach 0, so that the interference information entering the network through the residual connection in the complex features is also reduced, making the features output by the convolutional layer more pure. For some data similar to the historical data features, the deep neural network adjusts to approach 1 and further learns the difference information. When analyzing the change in the active state data of the neural pathway before and after microwave stimulation, by emphasizing this difference learning, the network can more sensitively capture the change in the active state of the neural pathway caused by microwave stimulation, so as to extract the features closely related to the treatment effect.

[0078] Dynamic strategy module: Collect historical microwave stimulation data, and learn and optimize the non-linear relationship between the regulation data and microwave stimulation parameters through a hybrid transfer learning model to obtain the optimal microwave stimulation parameters for the neural activity of a specific brain region;

[0079] Collect past microwave stimulation-related data, including the microwave parameter frequency during each microwave treatment , phase , treatment time point T, and corresponding treatment effect evaluation data , normalize and store these data to form historical microwave stimulation data ;

[0080] Preset an initial hybrid transfer learning model, and convert the optimal regulation data and historical microwave stimulation data into matrix form and then perform dimensionality reduction. Input the dimensionality-reduced historical microwave stimulation data matrix and the current optimized regulation data into the preset initial hybrid transfer learning model respectively;

[0081] The initial hybrid transfer learning model is implemented based on a dual-channel neural network, expressed as:

[0082]

[0083] where, is the input result of the input data passing through the initial hybrid transfer learning model, represents the bias term of the dual-channel neural network, represents the weight matrix of the dual-channel neural network, represents the hybrid feature extractor to perform feature extraction on the dimensionality-reduced historical microwave stimulation data matrix and the current optimized regulation data ;

[0084] Furthermore, includes an online channel and a transfer channel :

[0085] The online channel is a time-recursive mechanism that can continuously capture the dynamic changes of data over time. When recording the influence of microwave stimulation on neural activity over time, it obtains the fluctuations and trends of data at each time point;

[0086] The transfer channel introduces a domain-invariant projection mechanism that can achieve knowledge transfer between different domain data;

[0087] Add an improved online incremental learning process based on the initial hybrid transfer learning model to make the model more adaptable to new optimal regulation data and historical microwave stimulation data , and the implementation process of the improved online incremental learning process is:

[0088] The improved online incremental learning is implemented based on a sliding window by introducing an optimized feedback gain coefficient. The process of obtaining the optimized feedback gain coefficient is:

[0089] Set the initial feedback gain coefficient based on prior knowledge , for example, it is found based on prior knowledge that when is within a certain range, the model can respond well to new data, and a value within this range can be selected as the initial feedback gain coefficient;

[0090] Through a reinforcement learning approach, regard the adaptation process of the model to new data (the optimal regulation data newly input and historical microwave stimulation data ) as a decision-making process, set the value of as the decision-making action, and give rewards or punishments according to the processing effect of the model on new data (the improvement degree of prediction accuracy) under different values until convergence (the convergence condition is to let the model gradually find the most suitable value through continuous trial-and-error learning to achieve the best response to new data), and finally obtain an optimal feedback gain coefficient ;

[0091] Specifically, the basis of the reinforcement learning approach is that reinforcement learning is good at dealing with dynamically changing environments. Incorporate reinforcement learning innovatively into this specific scenario. As new data (optimal regulation data and historical microwave stimulation data ) is continuously input, the model needs to adapt to these new data dynamically. Reinforcement learning can continuously adjust the decision value according to the environmental changes brought by new data (the change in the model's processing effect on new data) so that the model can better handle new data;

[0092] Based on the optimal feedback gain coefficient When new data is input, the implementation process of improving the online incremental learning process is expressed as:

[0093]

[0094] Among them, represents the learning rate, represents the parameters of the model, represents the new parameters of the model, represents the loss function of the model with respect to the model parameters gradient, represents the newly input data (the optimal regulation data newly input and historical microwave stimulation data ), represents the model prediction result, represents the feature matrix of the newly input data;

[0095] Learning rate Adopt a learning rate decay strategy. Initially, set a relatively large learning rate (0.1). During the training process, after every specific number of rounds (10 rounds), reduce the learning rate by a certain proportion (multiply by 0.9), so that the model converges quickly in the early stage of training and adjusts the parameters more precisely in the later stage;

[0096] Specifically, by introducing a feedback gain coefficient in the sliding window to achieve improved online incremental learning, the model can quickly respond to the difference between new data and the predicted value. If the deviation between the predicted value and the actual situation of the new data is large, the feedback gain coefficient will make a large adjustment to the model parameters, enabling the model to quickly adapt to the characteristics of the new data;

[0097] Then, the final hybrid transfer learning model is expressed as:

[0098]

[0099] Wherein, is the output of the final hybrid transfer learning model combined with improved online incremental learning;

[0100] Finally, use the mini-batch gradient descent method to train the final hybrid transfer learning model. Each time, take a mini-batch of data from the training set, input it into the model to calculate the output , and then calculate the loss value according to the loss function L. Calculate the gradient of the loss function with respect to the model parameters of , and update the parameters according to the gradient update rule . Continuously repeat this process until the loss function converges when it meets the preset conditions (the decrease amplitude of the loss value in consecutive several rounds of training is less than a certain threshold). After the training is completed, the final hybrid transfer learning model outputs a series of data, that is, the optimal microwave stimulation parameters for specific brain region nerve activities obtained by learning the non-linear relationship between the optimized regulation data and the microwave stimulation parameters;

[0101] Specifically, the output of the final hybrid transfer learning model represents the output value of the microwave stimulation parameters given by the model after comprehensively considering the historical microwave stimulation data and the current optimized regulation data. This value is a quantitative representation of the optimal microwave stimulation parameters for specific brain region nerve activities, where:

[0102] The manifestation of the non-linear relationship: Through the hybrid feature extractor , which includes the online channel and the transfer channel , the data captured by the online channel changes dynamically over time, and the migration channel realizes the migration of data knowledge in different domains. The combination of the two can extract multiple features from complex data, thereby characterizing the complex non-linear relationship between the optimized regulation data and the microwave stimulation parameters, rather than a simple linear mapping;

[0103] Obtain the optimal parameters: Based on the input data, the model uses a dual-channel neural network structure for calculation, and then through a specific improved online incremental learning process, continuously adjusts the model parameters according to the difference between the new data and the predicted value (updated to ), through multiple iterative learning, the model output continuously approaches the optimal microwave stimulation parameters for the neural activity of a specific brain region, and finally outputs that is, the optimal microwave stimulation parameters.

[0104] Treatment execution module: Based on the optimal microwave stimulation parameters, control the microwave emission device to emit microwaves to the corresponding brain region of the patient for treatment and generate a three-dimensional visualization report of the treatment progress;

[0105] Input the relevant data (multi-modal microwave data) of the new patient into the trained model to obtain the optimal microwave stimulation parameters. The optimal microwave stimulation parameters include the frequency, phase frequency, and treatment time that are most suitable for the patient's data situation. For example:

[0106] Frequency (the value after anti-normalization), phase = 180°, treatment time minutes;

[0107] Treatment operation: The treatment execution module uses the Socket library of Python to establish a wired network connection with the microwave emission device (assuming the device supports the TCP / IP protocol), sends control instructions to the device, sets its output frequency to , the phase is 180°. During the treatment process, physiological indicators such as the patient's electroencephalogram signals are collected in real time through physiological monitoring devices such as electroencephalogram monitors, as well as microwave treatment device parameters (such as frequency, phase, power, etc.). Using the Plotly visualization library of Python, a three-dimensional coordinate system is constructed with the time axis as one dimension and the hemoglobin concentration change value and microwave treatment device parameters as the other two dimensions. The data at each time point is plotted as a data point in the coordinate system, and using the Plotly animation function, the dynamic change of the data point is displayed driven by time, generating a three-dimensional visualization report of the treatment progress, intuitively presenting the parameter change trend and the mutual relationship.

[0108] In the application, several formulas involved are calculated by taking their numerical values after dimensionless treatment. The establishment of the formulas is obtained by collecting a large amount of data for software simulation to get a formula closest to the actual situation. Some coefficients or weights in the formulas are set by those skilled in the art according to the actual situation, so no more details will be given here.

[0109] The above embodiments can be implemented in whole or in part by software, hardware, firmware or any other combination. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. Those of ordinary skill in the art can realize that the units and algorithm steps of each example described in combination with the embodiments disclosed herein can be implemented by electronic hardware or a combination of computer software and electronic hardware. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution.

[0110] Although the embodiments of the present invention have been shown and described, it will be understood by those of ordinary skill in the art that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. An AI intelligent regulation mental illness treatment system based on multimodal microwaves, characterized in that, Including: Data acquisition module: It acquires multi-modal microwave data, and the multi-modal microwave data includes basic microwave data, neural pathway activity state data, and individual physiological data; Optimization and regulation module: It extracts the optimal regulation data from the multi-modal microwave data through an improved deep convolutional neural network; Among them, the improved deep convolutional neural network is implemented by adding a residual connection between two consecutive convolutional layers in the basic deep convolutional neural network and introducing a self-optimizing learning scalar during the residual connection process, and by using dilated convolution to replace ordinary convolution; Dynamic strategy module: It acquires historical microwave stimulation data, and learns the non-linear relationship between the optimal regulation data and the microwave stimulation parameters through a hybrid transfer learning model to obtain the optimal microwave stimulation parameters for the neural activity of a specific brain region; Among them, the hybrid online transfer learning model is implemented based on a dual-channel neural network by adding an improved online incremental learning process; Treatment execution module: Based on the optimal microwave stimulation parameters, it controls the microwave emission device to emit microwaves to the corresponding brain region of the patient for treatment and generates a three-dimensional visualization report of the treatment progress.

2. The AI intelligent regulation mental illness treatment system based on multimodal microwave according to claim 1, characterized in that, The implementation process of adding a residual connection between the two consecutive convolutional layers is as follows: Taking the multi-modal microwave data as the input data, obtaining the mapping outputs of two convolutional layers through the operations of two consecutive convolutional layers; Based on the mapping outputs of the two convolutional layers, adding and fusing the multi-modal microwave data with the output after two convolutional layer mappings to realize the residual connection process by combining individual data points and weight sets.

3. The AI intelligent regulation mental illness treatment system based on multimodal microwave according to claim 2, characterized in that, The specific implementation process of the improved deep convolutional neural network is as follows: Presetting a learning scalar and assigning an initial learning scalar to the preset learning scalar; Introducing the initial learning scalar into the residual connection process to obtain the residual connection output with the initial learning scalar introduced; Based on the residual connection output with the initial learning scalar introduced, presetting a true output label, obtaining the loss value between the residual connection output with the initial learning scalar introduced and the true output label through a regression loss function, and obtaining the gradient of the loss value with respect to the self-optimizing learning scalar; Based on the gradient of the loss value with respect to the self-optimizing learning scalar, using a Lookahead optimizer for updating, and obtaining the self-optimizing learning scalar after updating to the preset maximum condition; Based on the self-optimizing learning scalar, introducing it into the residual connection of the basic deep convolutional neural network, and obtaining the improved deep convolutional neural network by using dilated convolution to replace ordinary convolution.

4. An AI intelligent regulation mental illness treatment system based on multimodal microwaves according to claim 1, characterized in that, The historical microwave stimulation data includes microwave parameter frequency, phase, treatment time point, and corresponding treatment effect evaluation data.

5. An AI intelligent regulation mental illness treatment system based on multimodal microwave according to claim 1, characterized in that, The training process of the hybrid transfer learning model is as follows: Presetting an initial hybrid transfer learning model, converting the optimal regulation data and the historical microwave stimulation data into matrix forms and then performing dimensionality reduction, and inputting the dimensionality-reduced historical microwave stimulation data matrix and the current optimized regulation data into the preset initial hybrid transfer learning model respectively; Based on the initial hybrid transfer learning model, adding an improved online incremental learning process to obtain the final hybrid transfer learning model; The final hybrid transfer learning model is trained using mini-batch gradient descent, and the loss value is calculated according to the loss function. The gradient of the loss function with respect to the model parameters is calculated through the backpropagation algorithm and the parameters are updated until the loss function converges to meet the preset conditions, obtaining a trained hybrid transfer learning model.

6. An AI intelligent regulation mental illness treatment system based on multimodal microwave according to claim 5, characterized in that, The initial hybrid transfer learning model is implemented based on a dual-channel neural network, and the dual-channel neural network includes a hybrid feature extractor, and the hybrid feature extractor includes an online channel and a transfer channel.

7. An AI intelligent regulation mental illness treatment system based on multimodal microwave according to claim 5, characterized in that, The implementation process of the improved online incremental learning process is as follows: The improved online incremental learning is achieved by introducing an optimized feedback gain coefficient based on a sliding window.

8. An AI intelligent regulation mental illness treatment system based on multimodal microwave according to claim 7, characterized in that, The process of obtaining the optimized feedback gain coefficient is as follows: Set the initial feedback gain coefficient of the feedback gain coefficient based on prior knowledge; Through a reinforcement learning idea, regard the adaptation process of the model to the optimal regulation data of the new input and the historical microwave stimulation data as a decision-making process, take the initial feedback gain coefficient value as the decision-making action, and give rewards or punishments according to the processing effects of the model on the optimal regulation data of the new input and the historical microwave stimulation data under different initial feedback gain coefficient values until convergence, and finally obtain an optimal optimized feedback gain coefficient.

Images