Clinical feature-based central nervous disease acupuncture response prediction method

Abstract

The invention relates to the technical field of acupuncture response prediction, and discloses a central nervous disease acupuncture response prediction method based on clinical characteristics. Wherein the frequency spectrum embedding network module captures frequency changes of different levels through a multi-level feature embedding mode, further introduces dynamic frequency spectrum disturbance to each feature, evaluates contribution of the feature in a frequency domain, dynamically adjusts learning weight of a model to the feature, and finally analyzes and extracts key frequency components in combination with the frequency spectrum disturbance; the adaptive decoding module of the frequency spectrum characteristics constructs a weighting matrix through disturbance response energy, weighs frequency spectrum components channel by channel, inputs the weighted frequency spectrum components to an adaptive frequency spectrum decoding mechanism to realize frequency band level decoupling transformation and aggregation, and fuses low-frequency and high-frequency sub-bands according to a disturbance response dominated energy proportion through a frequency spectrum partition weighting fusion mechanism to realize frequency band level decoupling transformation and aggregation; therefore, efficient analysis and integration of multi-frequency information are realized, the discrimination ability of the model to acupuncture response is enhanced, and the problem of acupuncture response prediction of central nervous diseases is solved.

Description

Technical Field

[0001] The present invention relates to the technical field of acupuncture response prediction, and in particular to a method for predicting acupuncture response to central nervous system diseases based on clinical characteristics. Background Art

[0002] Central nervous system diseases are a type of neurological dysfunction that seriously affects patients' cognition, movement and quality of life, such as stroke, Parkinson's disease, and Alzheimer's disease. In recent years, acupuncture, as an important treatment method in traditional Chinese medicine, has received widespread attention for its auxiliary efficacy in the rehabilitation of central nervous system diseases. Clinical studies have shown that there are significant differences in the response of different individuals to acupuncture stimulation, and its efficacy is affected by multiple factors such as the patient's clinical characteristics, physiological state, and disease type. Therefore, establishing a method for predicting acupuncture response based on clinical characteristics will not only help to formulate individualized treatment plans, but also improve the accuracy and safety of acupuncture intervention, which is of great significance in the modernization of traditional Chinese medicine and the development of precision medicine.

[0003] At present, researchers at home and abroad have begun to try to introduce machine learning methods to predict the efficacy of acupuncture. Common technical means include support vector machines, random forests, artificial neural networks and ensemble learning methods. These methods model clinical data such as patients' basic physical signs, medical history, laboratory indicators, etc., in order to identify groups that are sensitive or hyporesponsive to acupuncture. Some studies also integrate neural activity data such as EEG signals and functional magnetic resonance imaging to construct multimodal prediction models, thereby improving the explanatory power of acupuncture mechanisms and the accuracy of response assessment.

[0004] However, existing research still has many limitations. First, some methods over-rely on high-dimensional but difficult-to-obtain neuroimaging data, which limits their promotion and application in primary clinical settings; second, most current models are black-box structures, lacking the interpretability of the contribution of key features, making it difficult to guide clinical practice; in addition, existing methods fail to fully consider the dynamic changes in acupuncture response and its interaction with individual patient characteristics, resulting in poor stability and generalization of prediction results. Therefore, there is an urgent need for an acupuncture response modeling method based on clinically accessible features that is both interpretable, adaptable, and predictive, in order to promote the precise development of acupuncture treatment for central nervous system diseases. Summary of the Invention

[0005] The main purpose of the present invention is to provide a method for predicting acupuncture responses to central nervous system diseases based on clinical characteristics. The core of the present invention is to design a spectrum embedding network module and an adaptive decoding module of spectrum features to enhance the feature perception and prediction capabilities of the model. The spectrum embedding network module captures frequency changes at different levels through a multi-level feature embedding method, and then introduces dynamic spectrum perturbations to each feature, evaluates its importance in the frequency domain, and dynamically adjusts the model's learning weights for the features, and finally extracts key frequency components in combination with spectrum perturbation analysis. The adaptive decoding module of spectrum features constructs a weighted matrix through the perturbation response energy, weights the spectrum components channel by channel, and then inputs them into an adaptive spectrum decoding mechanism to achieve decoupling transformation and aggregation at the frequency band level, and through a spectrum partitioning weighted fusion mechanism, fuses low-frequency and high-frequency sub-bands with the energy ratio dominated by the perturbation response, thereby achieving efficient analysis and integration of multi-frequency information, and significantly enhancing the structural understanding and discrimination capabilities of the acupuncture response prediction model for central nervous system diseases.

[0006] To achieve the above objectives, the technical solution of the present invention is: a method for predicting acupuncture response of central nervous system diseases based on clinical characteristics, the method comprising: S1. Collect relevant information on acupuncture responses to central nervous system diseases, including: patient-related demographic information, clinical signs and neurological function assessments, and acupuncture treatment response assessments, which together constitute the central nervous system disease acupuncture response dataset; S2. Preprocessing of the acupuncture response dataset for central nervous system diseases. Methods include: missing value filling, outlier detection and removal, data normalization, and label consistency verification. S3. A spectrum embedding network module is proposed. The spectrum embedding network module first captures the frequency changes at different levels through multi-level feature embedding. Secondly, each feature is dynamically perturbed. The value obtained by the perturbation mechanism is used to evaluate the fit of the feature for prediction and dynamically adjust the learning weight of the model for the feature. Finally, the spectrum perturbation analysis is performed according to the threshold. Screening the frequency components of the disturbance response; S4. An adaptive decoding module for spectral features is proposed. This module uses the disturbance response energy to construct a weighting matrix and dynamically weights each spectral component channel by channel. The weighted spectral features are then fed into an adaptive spectral decoding mechanism, which performs a hierarchical analysis of multi-frequency information through band-level decoupling transformation and aggregation. Finally, a spectrum partition weighted fusion mechanism is used to fuse low-frequency and high-frequency sub-bands according to the energy ratio dominated by the disturbance response. S5. Construct a prediction model for acupuncture response to central nervous system diseases, which includes data preprocessing, a spectrum embedding network module, an adaptive decoding module of spectrum features, and loss function optimization; wherein the optimization of the loss function is the training stage of the prediction model. After the prediction model is trained, the test set is input into the prediction model for acupuncture response to central nervous system diseases, and finally the prediction model for acupuncture response to central nervous system diseases outputs the type of acupuncture response.

[0007] Furthermore, in step S3, for the training set ,in For the samples, is the number of features of the sample; each feature vector Mapped to the frequency domain, a multi-level spectrum representation is obtained, each feature After being converted into frequency components through discrete Fourier transform, its mathematical model is: ; Where, For the In the sample The value of the feature, For the In the sample The characteristic nth frequency component, For the In the sample The frequency components of the features, is the index of the frequency component, is the dimension of embedded frequency; the component of each feature in the frequency domain reflects its contribution to the frequency distribution in the time domain. In order to capture low-frequency and high-frequency information, the frequency representation is divided into multiple frequency bands. The mathematical model of the multiple frequency bands is: ; Where M is the number of frequency bands, and each frequency band corresponds to a feature representation of a different scale; In order to evaluate the suitability of each feature for prediction in different frequency bands, an adaptive spectrum perturbation analysis mechanism is introduced to introduce perturbations to each frequency component and quantify its impact on model prediction by calculating the energy change after perturbation. The frequency components of the features Introducing a perturbation vector , its mathematical model is: ; in, is the disturbance amplitude, which controls the impact of the disturbance on the spectrum. is the disturbance direction, is the spectrum representation after the disturbance is added; then the energy change of the spectrum after the disturbance is calculated to obtain the disturbance response energy, and its mathematical model is: ; Where, is the disturbance energy response of the rth feature, reflecting the impact of the disturbance on the prediction result, is the energy of the original spectrum, is the energy of the spectrum after disturbance, which generally reflects the impact of the disturbance on the frequency components; Through spectrum perturbation analysis, the frequency components that have the greatest impact on the model are selected; the perturbation energy response It is used to evaluate the fit of features to predictions, and selects frequency components with stronger disturbance responses as feature inputs. Based on the disturbance response energy, the frequency components are screened out. Its mathematical model is: ; Where, , is a threshold used to select the frequency component with larger disturbance response. is the feature set after perturbation screening.

[0008] Furthermore, in step S3, the spectrum embedding network module captures the potential frequency change patterns in the acupuncture response data features of central nervous system diseases through multi-level feature embedding, and then introduces a dynamic spectrum perturbation mechanism to perform perturbation evaluation on the response sensitivity of each feature in the frequency domain, thereby quantifying its importance in the prediction task; finally, the key frequency components in the acupuncture response data of central nervous system diseases are screened through the spectrum perturbation response intensity, so that the model focuses on the high-contribution information dimension; the spectrum embedding network module enhances the recognition ability of the central nervous system disease acupuncture response prediction model for nonlinear changes, improves the modeling ability of implicit structures in multidimensional complex features, and realizes the coordinated optimization of feature selection and representation learning.

[0009] Furthermore, in step S4, firstly, the normalized weighting coefficient of the disturbance energy response is calculated, and its mathematical model is: ; in, is the spectral attention weight after normalization of the perturbation response; then the spectral weighting matrix is ​​calculated , for features Each frequency component in is weighted, The mathematical model is: ; Where, For operations that generate diagonal matrices with one-dimensional vectors as diagonals, is the spectral feature weighting matrix, corresponding to the sample , used to adjust the amplitude of each characteristic spectrum component; Using a weighting matrix For input features Perform weighting to obtain the weighted spectrum feature representation , The mathematical model is: ; Then, set a frequency division threshold , which is used to divide the frequency into low frequency band and high frequency band, where the low frequency band index is , the high frequency segment index is , for the index According to the dynamic selection of spectral energy distribution, this application sets 70% of the frequency as the low-frequency upper bound and 30% of the frequency as the high-frequency lower bound; then constructs low-frequency and high-frequency spectrum subbands, and extracts the perturbation-weighted spectrum components in segments. The sub-vectors of the low-frequency band are: , the sub-vector of the high frequency band is: ; Finally, the spectrum characteristics after band division are weighted to obtain , its mathematical model is: ; Where, 、 They are the low-frequency and high-frequency feature weighting coefficients, which are used to control the importance of the two frequency bands and meet the requirements of ; The weighted spectral features are input into the structural response decoder of frequency auto-cointegration transformation. The structural response decoder of the frequency auto-cointegration transform The contribution of each frequency band is dynamically adjusted through the adaptive mechanism. First, the spectrum cointegration residual transformation is required. For the sample , first calculate any two dimensions of its spectrum characteristics and The linear residual of , its mathematical model is: ; Where, For the spectral features, For the spectral features, is the cointegration coefficient obtained by minimum residual fitting; then the cointegration structure constraint is constructed, and all cointegration residuals are combined into a structural matrix, and its mathematical model is: ; Where, For samples The cointegration residual combination matrix contains the cointegration deviation between all spectral features and is used to measure the stable dependence structure across frequency bands. Subsequently, a nonlinear cointegration response function is introduced, and the cointegration residual combination matrix is ​​used as input to extract its structural response characteristics. Its mathematical model is: ; Where, 、 is a trainable parameter matrix used to adjust the spectrum cointegration combination, To enhance the nonlinear response to disturbance, is the structural response characteristic; the structural response decoder of the final frequency auto-cointegration transformation Output final value , its mathematical model is: ; Where, is the output weight vector, is the bias term, For encoder For samples The output result;

[0010] Furthermore, in step S4, the adaptive decoding module of the spectral features constructs a weighted matrix through the disturbance response energy to achieve modeling and dynamic weighting of the importance of each spectral channel, thereby enhancing the expression ability of key frequency components; subsequently, the adaptive decoding module of the spectral features performs band-level decoupling and aggregation on the weighted spectrum, hierarchically analyzes multi-frequency information structures such as low-frequency trends and high-frequency mutations, and improves the ability to identify complex spectral patterns; finally, through the spectrum partitioning weighted fusion mechanism, the low-frequency and high-frequency information are optimized and fused according to the disturbance dominant energy ratio, thereby effectively improving the frequency domain expression accuracy of the adaptive decoding module of the spectral features and the structural sensitivity of the prediction results.

[0011] Compared with the prior art, the present invention has the following beneficial effects: The present invention proposes a method for predicting acupuncture responses to central nervous system diseases based on clinical characteristics, which includes a spectrum embedding network module and an adaptive decoding module of spectrum features. The spectrum embedding network module captures frequency changes at different levels through a multi-level feature embedding method, and then introduces dynamic spectrum perturbations to each feature, evaluates its importance in the frequency domain, and dynamically adjusts the model's learning weights for the features, and finally extracts key frequency components in combination with spectrum perturbation analysis; the adaptive decoding module of spectrum features constructs a weighted matrix through the perturbation response energy, weights the spectrum components channel by channel, and then inputs them into an adaptive spectrum decoding mechanism to achieve frequency band-level decoupling transformation and aggregation, and through a spectrum partitioning weighted fusion mechanism, fuses low-frequency and high-frequency sub-bands with an energy ratio dominated by the perturbation response, thereby achieving efficient analysis and integration of multi-frequency information, significantly enhancing the model's ability to discriminate acupuncture responses, and solving the problem of predicting acupuncture responses to central nervous system diseases. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Figure 1 Flowchart of the steps of a method for predicting acupuncture response in central nervous system diseases based on clinical characteristics.

[0013] Figure 2 This is the structural diagram of the spectrum embedding network module.

[0014] Figure 3 This is the structural diagram of the adaptive decoding module of spectral features.

[0015] Figure 4 This is a diagram of the training structure of the acupuncture response prediction model for central nervous system diseases.

[0016] Figure 5 This is the training graph for the acupuncture response prediction model for central nervous system diseases.

[0017] Figure 6 This is a comparison chart of the predicted values ​​and true values ​​of the acupuncture response prediction model for central nervous system diseases. DETAILED DESCRIPTION

[0018] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0019] See also Figures 1-6The present invention provides a technical solution: a method for predicting acupuncture response of central nervous system diseases based on clinical characteristics, the method steps including: collecting relevant information of acupuncture response of central nervous system diseases, preprocessing acupuncture response data set of central nervous system diseases, constructing a spectrum embedding network module, constructing an adaptive decoding module of spectrum features, and constructing a prediction model for acupuncture response of central nervous system diseases.

[0020] Please refer to Figure 1 As shown in the embodiment of the present application, a method for predicting acupuncture response of central nervous system diseases based on clinical characteristics is described, and the specific steps are as follows:

[0021] S1. Collect relevant information on acupuncture response for central nervous system diseases, including: patient-related demographic information, clinical signs and neurological function assessment, and acupuncture treatment response assessment, which together constitute the central nervous system disease acupuncture response dataset; among which, patient-related demographic information includes: gender, age, height, weight, body mass index, disease type, duration of disease, first onset or recurrence, and previous acupuncture history; clinical signs and neurological function assessment data include: blood pressure, heart rate, respiratory rate, NIHSS, UPDRS, MMSE, Barthel index, VAS score, and temperature pain threshold test results; acupuncture treatment response assessment results serve as label columns in the dataset, and label types include: high response on the day of treatment, low response within 1 hour after treatment, and no response 1 week after treatment.

[0022] S2. Preprocessing the acupuncture response dataset for central nervous system diseases, including the following steps: filling missing values, detecting and removing outliers, normalizing data, and checking label consistency; there are few missing values ​​in the acupuncture response dataset for central nervous system diseases, so this application will remove the data with missing values ​​to address the missing value problem; then use Methods: Outlier detection was performed on the acupuncture response dataset of central nervous system diseases. Normalization, calculate the The mathematical model of the score is: ; Where, and They are The mean and standard deviation of is the characteristic variable No. Sample values, for Sample value Score, if , it is determined to be an outlier and the sample is removed. The value is 3; for the data set after outlier processing Normalization processing, its mathematical model is: ; Where, is the characteristic variable The minimum value in is the characteristic variable The maximum value in is the characteristic variable Middle Normalize all variable values ​​to map all eigenvalues ​​to the interval \left [ {0,1} \right ] ; Finally, the labels were checked for consistency. Samples with high response evaluation results were marked as 0, low response samples were marked as 1, and no response samples were marked as 2. This constituted the acupuncture response prediction dataset for central nervous system diseases. The acupuncture response prediction dataset for central nervous system diseases was divided into a training set and a test set in a ratio of 8:2. The number of samples in the training set was 16,000, and the number of samples in the test set was 4,000.

[0023] S3, build spectrum embedding network module, such as Figure 2 As shown in the figure, first, a multi-level feature embedding method is used to capture frequency changes at different levels. Second, each feature is dynamically perturbed in the spectrum, and the perturbation mechanism is used to evaluate the importance of the feature and dynamically adjust the model's learning weight for the feature. Finally, through spectrum perturbation analysis, important frequency components are selected based on the response strength. The specific steps are as follows: S301, for the training set ,in For the samples, is the number of features of the sample, which is 19; each feature vector Mapped to the frequency domain, a multi-level spectrum representation is obtained, each feature After being converted into frequency components through discrete Fourier transform, its mathematical model is: ; Where, For the In the sample The value of the feature, For the In the sample The characteristic nth frequency component, For the In the sample The frequency components of the features, is the index of the frequency component, ranging from 0 to , is the dimension of the embedded frequency, with a value of 512. The component of each feature in the frequency domain reflects its contribution to the frequency distribution in the time domain. In order to capture low-frequency and high-frequency information, the frequency representation is divided into multiple frequency bands. The mathematical model of the multiple frequency bands is: ; Where M is the number of frequency bands, and each frequency band corresponds to a feature representation of a different scale; S302. In order to evaluate the importance of each feature in different frequency bands, an adaptive spectrum perturbation analysis mechanism is introduced to introduce perturbations to each frequency component and quantify its impact on model prediction by calculating the energy change after perturbation. First, The frequency components of the features Introducing a perturbation vector , its mathematical model is: ; in, is the disturbance amplitude, which controls the influence of the disturbance on the spectrum, and its value is 0.3. is the disturbance direction, is the spectrum representation after the disturbance is added; then the energy change of the spectrum after the disturbance is calculated to obtain the disturbance response energy, and its mathematical model is: ; Where, is the disturbance energy response of the rth feature, reflecting the impact of the disturbance on the prediction result, is the energy of the original spectrum, is the energy of the spectrum after disturbance, which generally reflects the impact of the disturbance on the frequency components; S303. Select the frequency component that has the greatest impact on the model through spectrum perturbation analysis; perturbation energy response It is used to evaluate the fit of features to predictions, and selects frequency components with stronger disturbance responses as feature inputs. Based on the disturbance response energy, the frequency components are screened out, and their mathematical model is: ; Where, , is a threshold value, which is 0.8, and the frequency component with larger disturbance response is selected. is the feature set after perturbation screening.

[0024] S4, construct an adaptive decoding module of spectrum features, such as Figure 3As shown in the figure, the perturbation response energy is used to construct a weighting matrix, and each spectral component is dynamically weighted channel by channel. The weighted spectral features are then input into the adaptive spectrum decoding mechanism, and the multi-frequency information is hierarchically analyzed through band-level decoupling transformation and aggregation. Finally, the spectrum partition weighted fusion mechanism is used to fuse the low-frequency and high-frequency sub-bands according to the energy ratio dominated by the perturbation response. The specific steps are as follows: S401. First, based on the normalized weighting coefficient of the disturbance energy response, the mathematical model is: ; in, is the spectral attention weight after normalization of the perturbation response; then the spectral weighting matrix is ​​calculated , for features Each frequency component in is weighted, The mathematical model is: ; Where, For operations that generate diagonal matrices with one-dimensional vectors as diagonals, is the spectral feature weighting matrix, corresponding to the sample , is a A diagonal matrix is ​​used to adjust the amplitude of each characteristic spectrum component; S402: Using a weighted matrix For input features Perform weighting to obtain the weighted spectrum feature representation , The mathematical model is: ; Then, set a frequency division threshold , which is used to divide the frequency into low frequency band and high frequency band, where the low frequency band index is , the high frequency segment index is , for the index According to the dynamic selection of spectral energy distribution, this application sets 70% of the frequency as the low-frequency upper bound and 30% of the frequency as the high-frequency lower bound; then constructs low-frequency and high-frequency spectrum subbands, and extracts the perturbation-weighted spectrum components in segments. The sub-vectors of the low-frequency band are: , the sub-vector of the high frequency band is: ; Finally, the spectrum characteristics after band division are weighted to obtain , its mathematical model is: ; Where, 、 They are the low-frequency and high-frequency feature weighting coefficients, which are used to control the importance of the two frequency bands and meet the requirements of ; S403: Input the weighted spectral features into the structural response decoder of frequency auto-cointegration transformation The structural response decoder of the frequency auto-cointegration transform The contribution of each frequency band is dynamically adjusted through the adaptive mechanism. First, the spectrum cointegration residual transformation is required. For the sample , first calculate any two dimensions of its spectrum characteristics and The linear residual of , its mathematical model is: ; Where, For the spectral features, For the spectral features, The cointegration coefficient is obtained by fitting the minimum residual, which is initialized to 1 and optimized iteratively. Then, the cointegration structure constraint is constructed, and all cointegration residuals are combined into a structure matrix. Its mathematical model is: ; Where, For samples The cointegration residual combination matrix contains the cointegration deviation between all spectral features and is used to measure the stable dependence structure across frequency bands. Subsequently, a nonlinear cointegration response function is introduced, and the cointegration residual combination matrix is ​​used as input to extract its structural response characteristics. Its mathematical model is: ; Where, 、 is a trainable parameter matrix used to adjust the spectrum cointegration combination, To enhance the nonlinear response to disturbance, is the structural response characteristic; the structural response decoder of the final frequency auto-cointegration transformation Output final value , its mathematical model is: ; Where, is the output weight vector, is the bias term, For encoder For samples The output result is:

[0025] S5. Construct a prediction model for acupuncture response to central nervous system diseases. The prediction model for acupuncture response to central nervous system diseases is constructed from multiple aspects including data preprocessing, spectrum embedding network module, adaptive decoding module of spectrum features and loss function optimization. The prediction model for acupuncture response to central nervous system diseases is built under Linux system and implemented using PyTorch deep learning framework. During the model training phase, a dynamic feedback adjustment mechanism and a joint loss function calculation mechanism are set to ensure that the model automatically adjusts the learning strategy at each stage to improve the prediction accuracy. The training structure is as follows: Figure 4 As shown, the specific implementation method is: S501. During the training process, each sample in the acupuncture response dataset for central nervous system diseases is normalized and then input into the model. The spectral embedding network module is used to perform frequency domain mapping on the input samples, and the learning weight of each frequency band feature is dynamically adjusted through frequency perturbation analysis. The adaptive decoding module of the spectral features is further used to achieve multi-level analysis of low-frequency and high-frequency features through band-level decoupling and fusion operations. In terms of loss function optimization, the learning rate is set to 0.0001, the batch size is 64, the optimizer is Adam, the weight decay coefficient is 1e-4, the gradient clipping threshold is set to 5, the ReLU activation function is used, the dropout ratio is set to 0.3 in the feature decoding stage, the training epochs is 200, the batch size is 128, and the training set is loaded in each round for forward propagation to calculate the loss and backpropagation to update the model parameters. A joint loss function is constructed. The joint loss function uses a standard prediction error term and introduces an additional adjustable structural regularization term to enhance the controllability of the model output. The mathematical model of the loss function is: ; Where, For the The true acupuncture response label value of the sample, Acupuncture response prediction model for central nervous system diseases The predicted value of the sample, is the regularization strength parameter, with a value of 0.1, which is the regularization term. For Perform regularization processing, is the loss function value; Figure 5 The figure shows the training loss value of the acupuncture response prediction model for central nervous system diseases. As the number of epochs increases, The value continues to decrease, indicating that the model accuracy continues to improve and the training effect is good; S502: After the model is trained using the training set in step S501, the trained model is verified, and the test set is input into the trained central nervous system disease acupuncture response prediction model to obtain the model's prediction output for the acupuncture response type; Figure 6 As shown in the figure, the horizontal axis is the sample number, the vertical axis is the acupuncture response prediction value, the dotted line is the actual response value, and the solid line is the model prediction value; it can be seen from the figure that the prediction result is consistent with the actual response value trend, and the deviation is small. The experimental results show that the constructed acupuncture response prediction model for central nervous system diseases can accurately predict the response type of acupuncture treatment for central nervous system diseases, and has good application effect.

[0026] The specific embodiments described above further illustrate the objectives, technical solutions, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. In particular, it should be noted that any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included within the scope of protection of the present invention for those skilled in the art.

Claims

1. A method for predicting acupuncture response in central nervous system diseases based on clinical characteristics, characterized in that: include: Collect information related to acupuncture responses in patients with central nervous system diseases and construct a dataset required for acupuncture response prediction in central nervous system diseases; Preprocessing the data set, including filling missing data in the data set, identifying and removing abnormal data, and normalizing the data set; Construct a spectrum embedding network module, extract the characteristic components of different frequency levels through multi-layer feature embedding, combine the dynamic spectrum perturbation mechanism, measure the fit of each feature to the response, and adjust the feature learning weight, and finally calculate the threshold value. The frequency components of the disturbance response were screened for subsequent analysis; Based on the disturbance energy, a channel weighting matrix is ​​constructed to weight the spectral features channel by channel. Hierarchical reconstruction of multi-frequency information is achieved through band-level structural decoupling and fusion operations. Low-frequency and high-frequency features are weighted and fused according to the dominant energy ratio of the disturbance, and an adaptive decoding module for spectral features is designed. A prediction model for acupuncture response to central nervous system diseases is constructed. The prediction model for acupuncture response to central nervous system diseases integrates a data preprocessing module, a spectrum embedding network, a spectrum feature adaptive decoding mechanism and a loss function optimization method. The parameters of the prediction model for acupuncture response to central nervous system diseases are optimized through a training phase. After training is completed, test samples are input into the model to output the prediction results of the acupuncture response type.

2. A method for predicting acupuncture response of central nervous system diseases based on clinical characteristics according to claim 1, characterized in that: The dataset is divided into a test set and a training set. For each original feature of each sample in the training set, its corresponding numerical sequence is first mapped to the frequency domain to extract the structural contribution of different frequency components to the original feature. This is achieved through Fourier transform, and its mathematical model is: ； Where, For the In the sample The value of the feature, For the In the sample The characteristic nth frequency component, For the In the sample The frequency components of the features, is the index of the frequency component, is the dimension of embedded frequency; the sequence of each eigenvalue in the time domain is converted into a complex representation in the frequency domain, thereby capturing its response intensity on multiple frequency components; the component of each feature in the frequency domain reflects its contribution to the frequency distribution in the time domain, and the frequency representation is divided into multiple frequency bands to capture low-frequency and high-frequency information. The mathematical model of the multiple frequency bands is: ； Where M is the number of frequency bands, and each frequency band corresponds to a feature representation of a different scale.

3. A method for predicting acupuncture response of central nervous system diseases based on clinical characteristics according to claim 2, characterized in that: An adaptive spectrum perturbation engine is introduced to add a small perturbation in a specific direction to each spectral component. The small perturbation is obtained by introducing a perturbation vector to obtain the spectrum representation after the perturbation. The influence of the frequency component on the model output is measured by calculating the change in energy before and after the perturbation. For each frequency component, the difference between its original energy and the energy after the perturbation is calculated, and the energy difference is used as the perturbation response intensity in the frequency dimension. Its mathematical model is: ； Where, is the disturbance energy response of the rth feature, reflecting the impact of the disturbance on the prediction result, is the energy of the original spectrum, is the energy of the spectrum after disturbance, which generally reflects the impact of disturbance on frequency components. is the disturbance amplitude, which controls the impact of the disturbance on the spectrum. It is the spectrum representation after adding disturbance; after completing the calculation of the spectrum disturbance response energy, the model uses the response intensity of each frequency component as a metric, and then, according to whether the energy change of the disturbance response exceeds the set threshold Filter each component in the spectrum, and the filtered frequency components together constitute the final spectrum feature set , used as input for subsequent models.

4. A method for predicting acupuncture response of central nervous system diseases based on clinical characteristics according to claim 3, characterized in that: The adaptive decoding module of the spectrum characteristics includes: according to the disturbance response strength of each frequency component Normalize it to obtain the weight of each frequency band; then construct a diagonal weighted matrix , used to analyze the spectrum characteristics Perform dynamic weighting at the channel level to obtain the weighted spectral feature representation , The mathematical model is: ; Then, set a frequency division threshold , which is used to divide the frequency into low frequency band and high frequency band, where the low frequency band index is , the high frequency segment index is , for the index According to the dynamic selection of spectral energy distribution, low-frequency and high-frequency spectrum sub-bands are then constructed, and the perturbation-weighted spectrum components are extracted segmentally. The sub-vector of the low-frequency band is: , the sub-vector of the high frequency band is: The model extracts low-frequency and high-frequency sub-vector representations respectively, and performs a weighted combination of the two according to the set fusion coefficient, and finally forms the feature representation after spectrum-level fusion. .

5. A method for predicting acupuncture response of central nervous system diseases based on clinical characteristics according to claim 4, characterized in that: The weighted spectral features are input into the structural response decoder of frequency auto-cointegration transformation. The structural response decoder of the frequency auto-cointegration transform The contribution of each frequency band is dynamically adjusted through the adaptive mechanism. First, the spectrum cointegration residual transformation is required. For the sample , first calculate any two dimensions of its spectrum characteristics and The linear residual of , its mathematical model is: ； Where, For the spectral features, For the spectral features, is the cointegration coefficient obtained by minimum residual fitting; then the cointegration structure constraint is constructed, and all cointegration residuals are combined into a structural matrix, and its mathematical model is: ； Where, For samples The cointegration residual combination matrix includes the cointegration deviation between all spectral features and is used to measure the stable dependence structure across frequency bands; Then, a nonlinear cointegration response function is introduced, which is composed of a set of trainable linear transformation matrices and nonlinear activation functions. The cointegration residual combination matrix is ​​used as input to extract its structural response characteristics. Finally, the extracted structural response characteristics are input to the output layer of the decoder and corrected by linear weighting and bias terms to generate samples. The prediction results , to achieve accurate prediction of acupuncture response based on spectral cointegration structure.

Images