A method for constructing a neural network stack acupuncture prediction model
By constructing a neural network stack acupuncture prediction model and combining it with continuous learning methods, the problems of low accuracy and catastrophic forgetting in existing acupuncture prediction models are solved, achieving high accuracy and applicability prediction of acupuncture efficacy.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- TIANJIN UNIV
- Filing Date
- 2023-11-16
- Publication Date
- 2026-06-12
Smart Images
Figure CN117497134B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of acupuncture prediction technology, and in particular to a method for constructing a neural network stack acupuncture prediction model. Background Technology
[0002] Acupuncture is a fundamental therapy in Traditional Chinese Medicine (TCM), and its efficacy in treating various diseases has been proven through long-term clinical practice worldwide. The essence of acupuncture lies in combining different acupoints and applying various techniques to these points. This transforms the mechanical motion of needle insertion into electrochemical signals, thereby stimulating a holistic regulatory response in the body and regulating its functional networks. Acupuncture inherently involves complex interventions and is inextricably linked to the practitioner's experience; applying appropriate acupuncture methods can enhance its effectiveness. The main purpose of acupoint combinations and technique quantification is to predict the body's functional network response to different acupoint combinations and acupuncture techniques, reveal the coding mechanism of acupuncture, and identify diseases where acupuncture has advantages or potential advantages. However, due to the complexity of the physical operation, it is difficult to monitor acupuncture methods during treatment; currently, the efficacy of acupuncture mainly relies on the physician's experience.
[0003] Acupuncture stimulation is complex, and this complexity makes it difficult to establish rigorous biological markers for evaluating acupuncture treatment methods and effects. With advancements in modern science and technology, recording methods have become increasingly sophisticated, resulting in massive amounts of data from acupuncture experiments. The application of acupuncture depth prediction simulation methods can analyze and fuse relevant data to identify one or more features that describe the effects of acupuncture, thereby optimizing the most effective acupuncture regimen. By modeling the response phenomena of acupuncture pathways, stimulation signals, and electrophysiological signals, acupuncture depth prediction simulation methods can simulate the stimulus-network dynamics of acupuncture methods, thus predicting treatment effects. However, due to current limitations in computing power and the high data requirements of acupuncture depth prediction simulation, these methods can only calculate the relationship between specific data and specific functional networks, and cannot predict changes in the overall functional network of the body based on different acupuncture data. Furthermore, due to the uncertainty of acupuncture experimental conditions and the influence of patient specificity, the prediction results obtained from acupuncture depth prediction simulation do not completely match the actual results of clinical treatment observation. Moreover, as the patient sample size increases, the specificity of acupuncture treatment effects also increases, which can lead to catastrophic forgetting in neural network models, making it impossible to make consistently accurate predictions of acupuncture efficacy.
[0004] While acupuncture depth prediction simulation methods can predict specific acupuncture input data and corresponding functional networks, their accuracy is low, making it difficult for the prediction model to meet global prediction requirements. Therefore, this model needs supplementation and correction. Multiple neural networks were used to build prediction models for neural, endocrine, and immune system pathways, forming a neural network stack acupuncture prediction model. The acupuncture input data was then standardized. Based on this, continuous learning methods were used for retraining to improve the model's small-sample learning ability and address the problem of catastrophic forgetting in the network. Summary of the Invention
[0005] The technical problem to be solved by the present invention is to overcome the shortcomings of the prior art and provide a method for constructing a neural network stack acupuncture prediction model.
[0006] This invention is achieved through the following technical solution:
[0007] A method for constructing a neural network stacked acupuncture prediction model, the method comprising the following steps:
[0008] S1, Based on the nodes and connections in the nervous system, a spiking neural network prediction model for the acupuncture-nervous system pathway is established;
[0009] S2. Based on the temporal and cyber-spatial characteristics of physiological data of the endocrine system, a deep neural network prediction model of the acupuncture-endocrine system pathway is established.
[0010] S3. Based on the characteristics of local information in immune system data, a convolutional neural network prediction model for the acupuncture-immune system pathway is established.
[0011] S4, pre-train the spiking neural network prediction model, the deep neural network prediction model, and the convolutional neural network prediction model;
[0012] S5, the spiking neural network prediction model, the deep neural network prediction model, and the convolutional neural network prediction model are stacked in parallel to form a neural network stack acupuncture prediction model;
[0013] S6, The neural network stack acupuncture prediction model is trained using a continuous learning method.
[0014] Preferably, S1 includes the following steps:
[0015] S101, Set the input parameters for the acupuncture-nervous system pathway spiking neural network prediction model;
[0016] S102, Set the membrane potential update method of neurons in the spiking neural network;
[0017] S103, Set the input summation method for each neuron in the spiking neural network;
[0018] S104 defines the mechanism for neuronal impulse firing;
[0019] S105 performs finer-grained channel normalization.
[0020] Preferably, S2 includes the following steps:
[0021] S201, Set the input parameters for the acupuncture-endocrine system pathway deep neural network prediction model;
[0022] S202, initializes the weights and biases of the deep neural network with normally distributed random values to break its symmetry;
[0023] S203, based on the ReLU activation function, selects the activation function through cross-validation to solve the gradient explosion problem and obtains the probability calculation function formula for the output layer;
[0024] S204, set the loss function of the deep neural network to the cross-entropy function.
[0025] Preferably, S3 includes the following steps:
[0026] S301, Set the input parameters for the acupuncture-immune system pathway convolutional neural network prediction model;
[0027] S302, perform a convolution operation between the processed input feature map and the convolution kernel, and calculate the output value at each position;
[0028] S303 extracts feature maps from acupuncture image data by performing multiple convolution and pooling operations in a convolutional neural network.
[0029] S304 performs pooling on the acupuncture feature map obtained after each convolution to reduce the spatial size of the feature map and extract key features;
[0030] S305: After the convolutional and pooling layers, a fully connected layer of the convolutional neural network is constructed. The acupuncture feature map data is used as the input of the fully connected layer, and the activation function and error function are selected.
[0031] Preferably, S4 includes the following steps:
[0032] S401, collects acupuncture data;
[0033] S402 deletes, fills in, and corrects missing and outlier values in the numerical features of acupuncture data, and uses Min-Max scaling to scale the data to a specified range.
[0034] S403, label encoding and unique heat encoding are performed on the classification features in the acupuncture data, and feature scaling is also performed;
[0035] S404 preprocesses the time series data in acupuncture data, extracts the frequency domain features of the time series data through Fourier transform, calculates the mean, variance, and skewness of the time series data to obtain the time domain features, and extracts the autocorrelation features of the time series data through an autoregressive moving average model.
[0036] S405, perform image denoising, image enhancement and cropping operations on the image data in the acupuncture data, extract image features to construct a basic acupuncture image dataset;
[0037] S406, the dataset is divided into a training set and a test set. The spiking neural network prediction model, the deep neural network prediction model, and the convolutional neural network prediction model are pre-trained using the training set, and the spiking neural network prediction model, the deep neural network prediction model, and the convolutional neural network prediction model are tested using the test set.
[0038] Preferably, S5 includes the following steps:
[0039] S501, the same data is input into the spiking neural network prediction model, the deep neural network prediction model and the convolutional neural network prediction model to obtain three sets of prediction results;
[0040] S502, combine the three sets of prediction results to obtain a combined prediction result;
[0041] S503 inputs the combined prediction results into the linear regression model to obtain the final prediction result.
[0042] Preferably, S6 includes the following steps:
[0043] S601, randomly remove data from the training dataset in S4 and add new data to construct a new dataset;
[0044] S602, obtains the classification labels of the new dataset from the original neural network stack acupuncture prediction model to provide supervision signals;
[0045] S603, trains a neural network stack acupuncture prediction model using a new dataset.
[0046] The beneficial effects of this invention are as follows: This invention proposes a method for constructing an acupuncture efficacy prediction model based on a neural network stack for continuous learning. First, the current massive acupuncture data is standardized and processed to construct a neural network stack that integrates neural, endocrine, and immune system pathways. Then, combined with the idea of continuous learning, limited acupuncture data is used to adapt to new patients and diseases, so that it can be applied to various diseases and acupuncture treatment plans with high accuracy and applicability. Attached Figure Description
[0047] Figure 1 This is a flowchart illustrating the overall design of the acupuncture efficacy prediction method based on a continuous learning neural network stack according to the present invention.
[0048] Figure 2 This is a schematic diagram of the method principle of the present invention.
[0049] Figure 3 This is a structural diagram of the acupuncture efficacy prediction model based on the neural network stack of the present invention.
[0050] Figure 4 This is a diagram showing the experimental results of the present invention. Detailed Implementation
[0051] To enable those skilled in the art to better understand the technical solutions of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings and preferred embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.
[0052] In the description of the invention, it should be noted that the terms "center", "upper", "lower", "left", "right", "vertical", "horizontal", "inner", "outer", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the invention.
[0053] Furthermore, it should be noted that, in the description of this invention, unless otherwise explicitly specified and limited, the terms "installation," "setting," and "connection" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a direct connection, an indirect connection through an intermediate medium, or a connection within two components. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.
[0054] Reference Figures 1 to 3This invention provides a method for constructing a neural network stack acupuncture prediction model, which includes the following steps:
[0055] S1 establishes a spiking neural network (SNN) prediction model for the acupuncture-nervous system pathway based on nodes and connections in the nervous system. S1 specifically includes S101 to S105.
[0056] S101 sets the input parameters for the acupuncture-nervous system pathway spiking neural network prediction model. Input parameters include the patient's age, gender, medical history, body mass index, pain location, pain intensity, duration, acupoint combination, needle depth, and EEG signal. The spiking neural network prediction model includes an input layer, hidden layers, and an output layer.
[0057] S102, Set the membrane potential update method for neurons in the spiking neural network:
[0058]
[0059] Where the subscript m represents the membrane potential of neurons in the neural pathway, j∈{0,n} l The superscript 'l' represents a specific layer within the pathway. Acupuncture stimulation input representing the neural pathway, V th The critical voltage representing a neuron; when the membrane potential exceeds a certain threshold under acupuncture stimulation, the current neuron will emit a signal. pulse.
[0060] S103, Define the input summation method for each neuron in the spiking neural network:
[0061]
[0062] Where w and b are the weights and biases in the nervous system.
[0063] S104, defines the mechanism for neuronal impulse firing:
[0064]
[0065] Here, U(x) represents the unit step function.
[0066] S105 performs finer-grained channel normalization to achieve rapid and effective information transmission in the acupuncture-neural pathway:
[0067]
[0068] Where w, λ, and b are the weights, the maximum activation calculated from the acupuncture training dataset, and the bias in layer l, respectively. i and j are the channel indices, and the weights w in layer l are calculated using the maximum activation in each channel. To normalize.
[0069] S2 establishes a deep neural network (DNN) prediction model for the acupuncture-endocrine system pathway based on the temporal and spatial characteristics of physiological data from the endocrine system. S2 specifically includes S201 to S204.
[0070] S201 sets the input parameters for the acupuncture-endocrine system pathway deep neural network prediction model. Input parameters include the patient's age, gender, body mass index, acupoint combinations, acupuncture depth, hormone levels, metabolic data, and biomarkers. The deep neural network prediction model includes an input layer, an output layer, and multiple hidden layers.
[0071] S202 initializes the weights and biases of the deep neural network with normally distributed random values to break its symmetry.
[0072] S203, based on the ReLU activation function, uses cross-validation to select the activation function to solve the gradient explosion problem and obtains the probability calculation function formula for the output layer. The probability calculation function formula for the output layer is:
[0073]
[0074] Where Softmax represents the normalization function, z i This represents the score value corresponding to the i-th symptom, and n represents the total number of all symptoms within the defined symptom range.
[0075] S204, The loss function for the deep neural network is set to the cross-entropy function. The relevant calculation formula is:
[0076]
[0077] Where L represents the cross-entropy loss value, K represents the number of acupuncture training sets, y is the output symptom type, and P(y i Q(y) represents the probability of the current disease type. i ) represents the probability of the i-th disease in the actual training set. For the correct disease category y, Q(y) i The probability is 1 for the first category and 0 for other categories.
[0078] S3, based on the characteristics of local information in immune system data, establishes a convolutional neural network (CNN) prediction model for the acupuncture-immune system pathway. S3 specifically includes S301 to S305.
[0079] S301 sets the input parameters for the acupuncture-immune system pathway convolutional neural network prediction model. Input parameters include patient data such as heart rate, body temperature, blood pressure, acupuncture point combinations, needling depth and technique, MRI, and optical microscopy data. The convolutional neural network prediction model includes multiple convolutional layers, pooling layers, and fully connected layers.
[0080] S302, performs a convolution operation between the processed input feature map and the convolution kernel, calculating the output value at each position. The formula for the convolution operation is as follows:
[0081]
[0082] Where H(i,j) is an element after acupuncture feature extraction, m and n represent the length and width of the acupuncture image dataset, respectively, and F is a convolution kernel that is multiplied element-wise with the feature map G and then added.
[0083] S303 extracts feature maps from acupuncture image data through multiple convolution and pooling operations within a convolutional neural network. The formula for calculating the size is as follows:
[0084]
[0085] Among them, O l W represents the size of the feature map of the acupuncture image data extracted after the l-th convolutional layer. in K represents the size of the acupuncture image data input to this layer. l P represents the kernel size of the current layer for the acupuncture data. l S is a padding operation performed on acupuncture image data. l This refers to the stride of the current layer.
[0086] S304 performs pooling on the acupuncture feature maps obtained after each convolution to reduce the spatial size of the feature maps and extract key features. The pooling formula is as follows:
[0087]
[0088] Where f(x,y) is an element in the acupuncture output feature map, K l W l These represent the height and width of the acupuncture data pooling window in the l-th pooling layer, respectively.
[0089] S305: After the convolutional and pooling layers, a fully connected layer of the convolutional neural network is constructed. The acupuncture feature map data is used as the input of the fully connected layer, and the activation function and error function are selected.
[0090] S4 involves pre-training the spiking neural network prediction model, the deep neural network prediction model, and the convolutional neural network prediction model. S4 specifically includes S401 to S406.
[0091] S401, collect acupuncture data. Acupuncture data includes the patient's age, gender, medical history, body mass index, pain location, pain intensity, duration, acupoint combination, acupuncture depth, acupuncture technique, EEG signal, hormone levels, metabolic data, biomarkers such as heart rate, body temperature, blood pressure, MRI, and optical microscopy data.
[0092] S402 removes, fills in, and corrects missing and outlier values in the acupuncture data, and uses Min-Max scaling to scale the data to a specified range. Numerical features include, but are not limited to, patient age, body mass index, duration, and heart rate. The specific formula for Min-Max scaling is:
[0093]
[0094] Among them, X scaled X represents the scaled value of the acupuncture numerical features, ranging from [0,1]. max ,X min These are the maximum and minimum values in the original acupuncture data, respectively.
[0095] S403 involves label encoding and unique heat encoding of the categorical features in the acupuncture data, followed by feature scaling. Categorical features include, but are not limited to, patient gender, pain location, and acupoint combinations.
[0096] S404 preprocesses the time-series data in acupuncture data. This time-series data includes, but is not limited to, EEG signals and metabolic data. Because EEG signals possess high-dimensional characteristics, Fourier transform is used to extract the frequency domain features of the time-series data, and the mean, variance, and skewness of the EEG signals are calculated to obtain their time-domain features. Then, an autoregressive moving average model is used to extract the autocorrelation features of the EEG signals. The specific formula is as follows:
[0097] (1-φ1L-φ2L 2 -…-φ p L p (1-L) d X t = (1+θ1L+θ2L) 2 +…+θ q L q ) ε t
[0098] Among them, X tφ1, φ2, ..., φ are the observed values of the EEG signal at time t, and d represents the number of differences in the EEG signal. p These are autoregressive parameters, representing the influence of EEG observations from the past p time steps on the current signal, θ1, θ2, ..., θ q It is the moving average parameter, representing the impact of the EEG error over the past q time steps on the current signal, ε. t This represents the white noise error term in the EEG signal. Finally, downsampling and interpolation operations are performed on the metabolic data.
[0099] S405 performs image denoising, image enhancement, and cropping operations on the image data in the acupuncture data to extract image features and construct a basic acupuncture image dataset. The image data includes, but is not limited to, MRI and optical microscopy data.
[0100] S406, the dataset generated in S401 to S402 is divided into a training set and a test set. The spiking neural network prediction model, the deep neural network prediction model, and the convolutional neural network prediction model are pre-trained using the training set, and the spiking neural network prediction model, the deep neural network prediction model, and the convolutional neural network prediction model are tested using the test set.
[0101] S5, the spiking neural network prediction model, the deep neural network prediction model, and the convolutional neural network prediction model are stacked in parallel to form a neural network stack acupuncture prediction model. S5 includes S501 to S503.
[0102] S501, the same data is input into the spiking neural network prediction model, the deep neural network prediction model, and the convolutional neural network prediction model to obtain three sets of prediction results.
[0103] S502 combines the three sets of prediction results to obtain a combined prediction result. The specific combination formula is as follows:
[0104] Y f =β0 + β1X1 + β2X2 + ... + β p X p +∈ a
[0105] Among them, Y f X1, X2, ..., X represent the final predicted acupuncture efficacy. p The values β0, β1, ..., β represent the acupuncture efficacy prediction results output from the pre-trained model. p This represents the coefficients corresponding to the output of the pre-trained model, ∈ a This represents the final random error in the model.
[0106] S503 inputs the combined prediction results into the linear regression model to obtain the final prediction result.
[0107] In summary, the neural network prediction model, composed of the spiking neural network prediction model, the deep neural network prediction model, the convolutional neural network prediction model, and the linear regression model, forms the neural network stack acupuncture prediction model.
[0108] S6 involves training the neural network stack acupuncture prediction model using a continuous learning method. Specifically, a new sample set is constructed based on actual acupuncture data from treated patients before and after acupuncture. This sample set is then fed into the neural network stack acupuncture prediction model for retraining using a continuous learning method. After fine-tuning, the final prediction model is obtained. S6 specifically includes S601 to S603.
[0109] In S601, as patient diversity increases, new tasks need to be continuously added to refine the existing neural network stack. Therefore, data from the training dataset in S4 is randomly removed, and new data is added to construct a new dataset.
[0110] S602 obtains the classification labels of the new dataset from the existing neural network stack acupuncture prediction model to provide supervision signals.
[0111] S603 trains the neural network stack acupuncture prediction model using a new dataset. To prevent the neural network stack from excessively altering important parameters, a parameter regularization method is added, with the relevant formula as follows:
[0112]
[0113] in, This is the current loss relative to the context, θ * This is the parameter vector penalized relative to its changes, and ∑ is the old model's estimate of important parameters for the context. The L2 norm is typically used as follows:
[0114]
[0115] This allows for fine-tuning of the network's structure and weights.
[0116] Further, S601 to S603 are repeated to obtain a new neural network stack model.
[0117] Subsequently, the predicted evaluation indicators are used in the model evaluation judgment, and continuous monitoring and performance evaluation are carried out to output a well-evaluated prediction model. The current acupoint combination, acupuncture technique, specific disease condition, low-dimensional characteristics of the neuro-endocrine-immune system, response rate of each system, and quantitative results of acupuncture effect are obtained in real time and normalized. Then, based on the well-evaluated prediction model, the acupuncture efficacy is predicted.
[0118] To verify the effectiveness of the method of this invention, specific experiments were conducted on a constructed real acupuncture dataset. The experimental environment was: Windows operating system, Intel(R) Core(TM) i5-8400 CPU, 32GB memory, and PyTorch deep learning framework. The dataset included information in seven aspects: patient gender, age, body mass index, acupuncture point combination, acupuncture method, EEG signal, and MRI image. The constructed dataset contained acupuncture information from 100 patients. A sliding window was used to expand the data to 10,000 sets, with 80% of the data used for training and 20% for testing.
[0119] Figure 4 Experimental results show that the predicted low-dimensional manifold of EEG obtained by this invention ( Figure 4 (above) and actual data ( Figure 4 The fact that the following (below) exhibits the same variation characteristics indicates that accurate prediction can be achieved in predicting the efficacy of acupuncture, proving the feasibility of the method of the present invention.
[0120] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.
Claims
1. A method for constructing a neural network stack acupuncture prediction model, characterized in that, Includes the following steps: S1, Based on the nodes and connections in the nervous system, a spiking neural network prediction model for the acupuncture-nervous system pathway is established; S2. Based on the temporal and cyber-spatial characteristics of physiological data of the endocrine system, a deep neural network prediction model of the acupuncture-endocrine system pathway is established. S3. Based on the characteristics of local information in immune system data, a convolutional neural network prediction model for the acupuncture-immune system pathway is established. S4, pre-train the spiking neural network prediction model, the deep neural network prediction model, and the convolutional neural network prediction model; S5, the spiking neural network prediction model, the deep neural network prediction model, and the convolutional neural network prediction model are stacked in parallel to form a neural network stack acupuncture prediction model; S6, The neural network stack acupuncture prediction model is trained using a continuous learning method; Step S1 includes: S101, set the input parameters of the acupuncture-nervous system pathway spiking neural network prediction model. The input parameters include the patient's age, gender, medical history, body mass index, pain location, pain intensity, duration, acupoint combination, acupuncture depth, and EEG signal. The spiking neural network prediction model includes an input layer, a hidden layer, and an output layer. S102, Set the membrane potential update method for neurons in the spiking neural network: ； Where the subscript m represents the membrane potential of neurons in the neural pathway, j∈{0,n} l }, where the superscript 'l' represents a specific layer within the pathway. Acupuncture stimulation input represents the neural pathways. The critical voltage representing a neuron; when the membrane potential exceeds a certain threshold under acupuncture stimulation, the current neuron will emit a signal. pulse; Step S2 includes: S201, set the input parameters of the acupuncture-endocrine system pathway deep neural network prediction model. The input parameters include the patient's age, gender, body mass index, acupoint combination, acupuncture depth, hormone level, metabolic data, and biomarkers. The deep neural network prediction model includes an input layer, an output layer, and multiple hidden layers. Step S3 includes: S301 sets the input parameters for the acupuncture-immune system pathway convolutional neural network prediction model. The input parameters include the patient's heart rate, body temperature, blood pressure, acupoint combination, acupuncture depth and technique, MRI, and optical microscope. The convolutional neural network prediction model includes multiple convolutional layers, pooling layers, and fully connected layers.
2. The method for constructing a neural network stacked acupuncture prediction model according to claim 1, characterized in that, S1 includes the following steps: S101, Set the input parameters for the acupuncture-nervous system pathway spiking neural network prediction model; S102, Set the membrane potential update method of neurons in the spiking neural network; S103, Set the input summation method for each neuron in the spiking neural network; S104 defines the mechanism for neuronal impulse firing; S105 performs finer-grained channel normalization.
3. The method for constructing a neural network stacked acupuncture prediction model according to claim 1, characterized in that, S2 includes the following steps: S201, Set the input parameters for the acupuncture-endocrine system pathway deep neural network prediction model; S202, initializes the weights and biases of the deep neural network with normally distributed random values to break its symmetry; S203, based on the ReLU activation function, selects the activation function through cross-validation to solve the gradient explosion problem and obtains the probability calculation function formula for the output layer; S204, set the loss function of the deep neural network to the cross-entropy function.
4. The method for constructing a neural network stacked acupuncture prediction model according to claim 1, characterized in that, S3 includes the following steps: S301, Set the input parameters for the acupuncture-immune system pathway convolutional neural network prediction model; S302, perform a convolution operation between the processed input feature map and the convolution kernel, and calculate the output value at each position; S303 extracts feature maps from acupuncture image data by performing multiple convolution and pooling operations in a convolutional neural network. S304 performs pooling on the acupuncture feature map obtained after each convolution to reduce the spatial size of the feature map and extract key features; S305: After the convolutional and pooling layers, a fully connected layer of the convolutional neural network is constructed. The acupuncture feature map data is used as the input of the fully connected layer, and the activation function and error function are selected.
5. The method for constructing a neural network stacked acupuncture prediction model according to claim 1, characterized in that, S4 includes the following steps: S401, collects acupuncture data; S402 deletes, fills in, and corrects missing and outlier values in the numerical features of acupuncture data, and uses Min-Max scaling to scale the data to a specified range. S403, label encoding and unique heat encoding are performed on the classification features in the acupuncture data, and feature scaling is also performed; S404 preprocesses the time series data in acupuncture data, extracts the frequency domain features of the time series data through Fourier transform, calculates the mean, variance, and skewness of the time series data to obtain the time domain features, and extracts the autocorrelation features of the time series data through an autoregressive moving average model. S405, perform image denoising, image enhancement and cropping operations on the image data in the acupuncture data, extract image features to construct a basic acupuncture image dataset; S406, the dataset is divided into a training set and a test set. The spiking neural network prediction model, the deep neural network prediction model, and the convolutional neural network prediction model are pre-trained using the training set, and the spiking neural network prediction model, the deep neural network prediction model, and the convolutional neural network prediction model are tested using the test set. The acupuncture data includes the patient's age, gender, medical history, body mass index, pain location, pain intensity, duration, acupoint combination, acupuncture depth, acupuncture technique, EEG signal, hormone levels, metabolic data, biomarkers heart rate, body temperature, blood pressure, MRI, and optical microscopy.
6. The method for constructing a neural network stacked acupuncture prediction model according to claim 1, characterized in that, S5 includes the following steps: S501, the same data is input into the spiking neural network prediction model, the deep neural network prediction model and the convolutional neural network prediction model to obtain three sets of prediction results; S502, combine the three sets of prediction results to obtain a combined prediction result; S503 inputs the combined prediction results into the linear regression model to obtain the final prediction result.
7. The method for constructing a neural network stacked acupuncture prediction model according to claim 1, characterized in that, S6 includes the following steps: S601, randomly remove data from the training dataset in S4 and add new data to construct a new dataset; S602, obtains the classification labels of the new dataset from the original neural network stack acupuncture prediction model to provide supervision signals; S603, trains a neural network stack acupuncture prediction model using a new dataset.