Method and system for evaluating instant curative effect of acupuncture head acupoints on treatment of cerebral apoplexy cognitive impairment based on electroencephalogram signals

Abstract

The invention provides a method and system for evaluating the instant curative effect of acupuncture head acupoints on cerebral stroke cognitive impairment based on electroencephalogram signals, and the method comprises the steps: collecting the electroencephalogram signals before, during and after acupuncture in stages, and synchronously combining a MoCA / MMSE scale and a cognitive task test to obtain subjective curative effect data; an insulating acupuncture needle is adopted to reduce interference on electroencephalogram signals, and 4-8-conduction portable dry electrode electroencephalogram equipment is matched to collect signals in an area avoiding acupuncture points of the head, so that acupuncture operation and electroencephalogram collection are ensured not to influence each other. A condition vector is generated by using acupuncture point parameters and acupuncture manipulation, and a space-time Transform-GNN model is driven to dynamically focus alpha / theta / beta frequency band time sequence characteristics and a brain network connection mode of cognitive related brain regions such as a prefrontal lobe, a parietal lobe and a temporal lobe. According to the method, a multi-modal feature vector containing electroencephalogram frequency domain features, time domain components and clinical scale scores is constructed through an electroencephalogram-acupoint-curative effect triple data training model of at least 100 patients, and an obvious / effective / invalid level is output.

Description

Technical Field

[0001] This invention relates to the fields of neuroscience and efficacy evaluation technology, and in particular to a method and system for evaluating the immediate efficacy of acupuncture at head acupoints in treating cognitive impairment following stroke based on electroencephalogram (EEG) signals. Background Technology

[0002] Stroke is a common neurological disorder that often leads to cognitive impairments such as memory loss, attention deficit, and executive dysfunction. Acupuncture, a traditional Chinese medicine therapy, has shown potential efficacy in improving cognitive function in stroke patients by stimulating acupoints on the head. However, current evaluations of acupuncture treatment efficacy primarily rely on subjective clinical assessment methods, such as the Montreal Cognitive Assessment (MoCA) and the Mini-Mental State Examination (MMSE). These methods are heavily influenced by the assessor's subjectivity and lack objective neurophysiological indicators, making it difficult to accurately quantify the neural mechanisms underlying cognitive improvement. Furthermore, traditional electroencephalogram (EEG) signal analysis methods often fail to adequately consider the specific modulation effects of different acupoints and acupuncture techniques on EEG signals, making it difficult to accurately reflect the immediate therapeutic effect of acupuncture. Existing model architectures struggle to simultaneously and effectively capture the spatial correlation and temporal dynamics of EEG signals, failing to meet the clinical need for precise assessment of acupuncture efficacy. Therefore, developing a method and system that can objectively and accurately evaluate the immediate efficacy of acupuncture at head acupoints in treating cognitive impairment following stroke has significant clinical importance and application value. Summary of the Invention

[0003] In view of this, the purpose of this invention is to provide a method and system for evaluating the immediate efficacy of acupuncture at head acupoints in treating cognitive impairment following stroke based on electroencephalogram (EEG) signals. This method integrates EEG features and treatment parameters through a conditional spatiotemporal Transformer-GNN model to achieve objective evaluation of the therapeutic effect.

[0004] To achieve the above objectives, the present invention adopts the following technical solution: a method for evaluating the immediate efficacy of acupuncture at head acupoints in treating cognitive impairment following stroke based on electroencephalogram (EEG) signals, comprising: Step 1: Collect EEG signals from patients before, during, and after acupuncture in multiple stages, including resting state and cognitive task state. At least 2 minutes of resting state signals and 3-5 minutes of cognitive task state signals should be collected in each stage before, during, and after acupuncture, and the time points of needle insertion, needle manipulation, needle retention, and needle withdrawal should be marked. Step 2: Obtain subjective cognitive evaluation data using the Montreal Cognitive Assessment Scale (MoCA) or the Mini-Mental State Examination (MMSE), and simultaneously record the task completion time and accuracy of memory, language, or executive function tests. Step 3: Using insulated acupuncture needles in conjunction with a 4-8 lead portable dry electrode EEG device, signals are collected in areas avoiding acupoints on the head; Step 4: Map the names of acupoints on the head to 50-dimensional one-hot encoded vectors through the built-in dictionary, convert the technique type to 3-dimensional category encoding, normalize the parameters to 2-dimensional numerical encoding, and concatenate them to generate 55-dimensional or 105-dimensional conditional vectors. Drive the parameter generation network to adjust the graph convolution kernel parameters and attention mechanism weights of the spatiotemporal Transformer-GNN model, so that the spatiotemporal Transformer-GNN model is sensitive to the α / θ / β frequency band features of the prefrontal, parietal, and temporal lobes. Step 5: Extract the spatiotemporal features of EEG related to the treatment method and acupoints, combine them with conditional vectors to construct an immediate efficacy evaluation model for acupuncture, input the trained model and output the efficacy prediction results.

[0005] In a preferred embodiment, collecting resting-state information and cognitive task-state information includes: Closed-eye resting-state acquisition is used to obtain the brain's basic electrical activity when there is no visual input, highlighting the occipital lobe alpha wave characteristics; open-eye resting-state acquisition is used to capture the brain's arousal level under visual input, reflecting the prefrontal lobe beta wave activity. Cognitive task state acquisition includes EEG signals in visual Oddball paradigm, n-back paradigm, and Stroop color conflict task, and simultaneous recording of task accuracy and reaction time.

[0006] In a preferred embodiment, the condition vector generation step includes: (1) Head acupoints are mapped by 50-dimensional unique thermal coding, where a single acupoint corresponds to only one brain region dimension set to 1, and multiple acupoints are spliced ​​together to generate a higher-dimensional code; (2) The technique type adopts a 3D unique thermal encoding of lifting and thrusting [1, 0, 0], twisting [0, 1, 0], and electroacupuncture [0, 0, 1]. The electroacupuncture frequency and intensity are determined according to... Normalized to 2D numerical codes, and concatenated to form a 5D method code vector.

[0007] In a preferred embodiment, the acupuncture immediate efficacy evaluation model includes: The conditional GNN layer uses the spatial distribution of electrodes as a graph structure. The graph convolution kernel parameters are dynamically adjusted by the edge weight offset generated by the conditional vector to enhance the functional connectivity weights of the prefrontal-parietal cognitive brain region. The conditionalized Transformer layer adjusts the weights of the self-attention mechanism through attention bias to highlight the temporal dependence of theta and beta waves; The multimodal fusion layer integrates spatial features, temporal features, and conditional vectors based on a cross-attention mechanism. The output layer combines the MoCA / MMSE judgment criteria to output the effective, valid, and ineffective levels.

[0008] In a preferred embodiment, the criteria for determining the efficacy prediction result are as follows: Significant effect: MoCA score improved by ≥3 points or MMSE score improved by ≥2 points, and the accuracy of cognitive tasks improved by ≥15%; Effective: MoCA score improves by 1-2 points or MMSE score improves by 1 point, and cognitive task accuracy improves by 5%-14%; Invalid: MoCA score improvement <1 point and MMSE score no change, cognitive task accuracy improvement <5%.

[0009] In a preferred embodiment, an EEG-treatment association database is constructed as a database for evaluating the immediate efficacy of acupuncture. A directed graph is used to store triplets including head acupoints, manipulation parameters, and EEG features. Conditional probabilities are calculated using a Bayesian network, and the knowledge base is updated using an incremental learning algorithm. EEG features are selected based on cosine similarity ≥ 0.6, and feature weights are adjusted in combination with model prior knowledge to form a 10-15 dimensional auxiliary feature vector.

[0010] This invention also provides a system for evaluating the immediate efficacy of acupuncture at head acupoints in treating cognitive impairment following stroke based on electroencephalogram (EEG) signals, and a method for evaluating the immediate efficacy of acupuncture at head acupoints in treating cognitive impairment following stroke based on EEG signals, comprising: User interface: Receives the name of acupoints on the head and the parameters of the manipulation, generates a conditional vector and displays the EEG thermogram and the contribution of key features in real time; EEG signal acquisition module: Uses 4-8 conductor electrodes, avoids acupoints on the head, and marks the time points of the acupuncture stage; Data processing module: Removes artifacts through independent component analysis, extracts α / θ / β time-frequency domain features, and constructs the electrode space map structure; Conditional parameter adjustment module: outputs GNN edge weight offset and Transformer attention bias, and the feature extraction controller drives the model to focus on acupoint-related brain regions and time-series frequency bands; EEG-Therapeutic Association Database: Stores triplet knowledge; Acupuncture Real-Time Efficacy Evaluation Model Outputs Efficacy Levels Based on Conditioned Spatiotemporal Transformer-GNN Architecture.

[0011] In a preferred embodiment, the condition parameter adjustment module includes: The conditional vector encoder generates 55-dimensional / 105-dimensional conditional vectors, and the parameter generation network outputs two sets of dynamic parameters, which are used to enhance the spatial correlation of the brain regions corresponding to acupoints and highlight the temporal signals of the treatment phase.

[0012] In a preferred embodiment, the training method for the acupuncture immediate efficacy evaluation model includes: Collect EEG signals, treatment parameters, and subjective evaluation data from at least 100 patients to construct a triplet training dataset; adjust model parameters by jointly optimizing classification loss and regression loss, and use L1 regularization to constrain dynamic parameter output.

[0013] Compared with existing technologies, the present invention has the following advantages: by dynamically integrating EEG spatiotemporal features and treatment parameters through a conditional deep learning model, it achieves an objective quantitative assessment of acupuncture efficacy, thus overcoming the limitations of traditional subjective evaluation. Attached Figure Description

[0014] Figure 1 This is a flowchart illustrating a method for evaluating the immediate efficacy of acupuncture at head acupoints in treating cognitive impairment following stroke, based on electroencephalogram (EEG) signals, according to an embodiment of the present invention. Figure 2 This invention provides a framework diagram for evaluating the immediate efficacy of acupuncture at head acupoints in treating cognitive impairment following stroke, based on electroencephalogram (EEG) signals. Figure 3 This is a schematic diagram illustrating the conditional vector generation method for evaluating the immediate efficacy of acupuncture at head acupoints in treating cognitive impairment following stroke, based on electroencephalogram (EEG) signals, as provided in an embodiment of the present invention. Figure 4 This is a schematic diagram of the internal structure of a conditional parameter adjustment module for evaluating the immediate efficacy of acupuncture at head acupoints in treating cognitive impairment following stroke, based on electroencephalogram (EEG) signals, according to an embodiment of the present invention. Figure 5 This is a schematic diagram of an embodiment of an electronic device for evaluating the immediate efficacy of acupuncture at head acupoints in treating cognitive impairment following stroke, based on electroencephalogram (EEG) signals, as provided in an embodiment of the present invention. Detailed Implementation

[0015] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0016] It should be noted that the following detailed descriptions are illustrative and intended to provide further explanation of this application. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains.

[0017] It should be noted that the terminology used herein is for the purpose of describing particular implementations only and is not intended to limit the exemplary implementations according to this application; as used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise; furthermore, it should be understood that when the terms “comprising” and / or “including” are used in this specification, they indicate the presence of features, steps, operations, devices, components and / or combinations thereof.

[0018] A method and system for evaluating the immediate efficacy of acupuncture at head acupoints in treating cognitive impairment following stroke based on electroencephalogram (EEG) signals, referencing Figure 1-5 .

[0019] In a first aspect of this application, a method is provided for evaluating the immediate efficacy of acupuncture at head acupoints in treating cognitive impairment following stroke based on electroencephalogram (EEG) signals, comprising: 1. Multi-stage acquisition of EEG signals Before acupuncture: EEG signals were collected in both resting and cognitive task states. During resting-state acquisition, eyes-closed rest was used to obtain baseline electrical activity of the brain without visual input, highlighting features such as occipital alpha waves; eyes-open rest was used to capture the brain's arousal level and attentional baseline under visual input, suppressing alpha waves and reflecting prefrontal cortex beta wave activity. Cognitive task-state acquisition included EEG signals during memory, language, or executive function tests, simultaneously recording task completion time and accuracy. This stage covered cognitively relevant brain regions such as the prefrontal, parietal, and temporal lobes, establishing baseline neurophysiological activity characteristics.

[0020] During acupuncture: Resting-state EEG signals were simultaneously acquired during the needle insertion, manipulation, and retention stages, and acupuncture technique parameters and time points were marked. Insulated acupuncture needles were used to reduce interference with EEG signals, and a 4-8 lead portable dry electrode EEG device was used to acquire signals in areas avoiding acupoints on the head, thus avoiding interference with the electrodes in the acupoint areas.

[0021] After acupuncture: Resting-state and cognitive task-state EEG signals were collected again and compared with baseline data to capture characteristic changes such as prefrontal theta wave inhibition and parietal alpha wave synchrony. At least 2 minutes of resting-state signals (1 minute with eyes open and 1 minute with eyes closed) and 3-5 minutes of cognitive task-state signals were collected before, during and after acupuncture, and the time points of needle insertion, needle manipulation, needle retention and needle withdrawal were marked.

[0022] 2. Subjective Efficacy Assessment: Subjective cognitive evaluation data were obtained using the Montreal Cognitive Assessment (MoCA) or the Mini-Mental State Examination (MMSE), with cognitive task test performance recorded simultaneously to form a multi-dimensional efficacy reference. The MoCA score covers eight dimensions, including attention, memory, and executive function, with the difference between baseline and recovery scores serving as the core indicator for efficacy assessment; cognitive task accuracy serves as supplementary evidence from objective behavioral studies. The MMSE also assesses orientation, memory, and other aspects, and changes in its total score are also an important basis for judging efficacy.

[0023] 3. Conditional Vector Generation Acupoint Coding: Acupoint names are mapped to high-dimensional one-hot encoded vectors using a built-in head acupoint dictionary. The system includes a dedicated dictionary of head acupoints such as Baihui, Shenting, Fengchi, and Temporal Three Needles, and uses 50-dimensional one-hot encoding for acupoint mapping. When mapping a single acupoint, only the corresponding brain region dimension is set to 1, and the rest are 0; when combining multiple acupoints, the one-hot vectors of each acupoint are concatenated to generate a higher-dimensional code.

[0024] Technique coding: Technique types are converted into category codes, and parameters are normalized to [0, 1] numerical codes. A 3D unique vector is used to represent the technique type, combined with a 2D numerical vector to achieve parameter normalization. Electroacupuncture frequency and intensity are normalized using a linear normalization formula. Mapped to the [0, 1] interval, and concatenated to form a 5-dimensional technique encoding vector.

[0025] Vector concatenation: Acupoint and manipulation technique codes are concatenated to generate conditional vectors, driving the model to focus on specific brain regions and frequency bands. A single acupoint conditional vector is generated by concatenating a 50-dimensional acupoint code and a 5-dimensional manipulation technique code into a 55-dimensional feature vector; a combination of two acupoints generates a 105-dimensional vector. This vector drives a parameter generation network, dynamically adjusting the weights of the graph convolution kernels to achieve targeted focus of the model's feature extraction on the target brain region.

[0026] 4. Conditional Spatiotemporal Feature Extraction Dynamic parameter adjustment includes GNN parameters and Transformer parameters. GNN parameters output edge weight offsets to enhance the spatial connectivity between acupoints and their corresponding brain regions. For example, the GNN edge weight offset ΔWgnn, based on acupoint encoding, enhances the functional connectivity weights of cognitively relevant brain regions such as the prefrontal-parietal lobe and temporal-hippocampus regions; for instance, the Shen Ting acupoint increases the Fp1-Fp2 edge weight by 30%. Transformer parameters output attention biases to highlight the temporal dependence of cognitively relevant frequency bands. For example, the Transformer attention bias ΔAattn, based on manipulation parameters, highlights the temporal dependence of theta and beta waves; for instance, the twisting manipulation increases the attention weight of prefrontal beta waves by 25%.

[0027] Spatiotemporal feature extraction includes a conditional GNN layer and a conditional Transformer layer. The conditional GNN layer uses the spatial distribution of electrodes as a graph structure, dynamically adjusting the graph convolution kernel parameters using conditional vectors to extract spatial features from brain regions such as the prefrontal and parietal lobes. This layer constructs a graph structure using a 10-20 system of electrode distributions, adjusting the adjacency matrix through ΔWgnn to extract spatial features such as prefrontal theta wave power and parietal alpha wave synchronicity. The conditional Transformer layer encodes temporal sequences, capturing long-range dynamics of the alpha / theta / beta frequency bands during treatment through attention mechanisms. It encodes 30-minute temporal signals, enhancing the long-range dependence of specific frequency bands during treatment phases through ΔAattn, such as increasing the attention weight of parietal P300 latency changes by 40% during needle retention.

[0028] 5. Multimodal fusion and efficacy prediction Feature fusion: Based on the cross-attention mechanism, spatial features extracted by GNN, temporal features extracted by Transformer, and conditional vectors are fused to generate multimodal feature vectors. This integrates spatial features and clinical data.

[0029] Therapeutic effect output: The therapeutic effect prediction value is output through a fully connected network, combined with the MoCA / MMSE assessment criteria for grading. Significant effect: MoCA score improved by ≥3 points or MMSE score improved by ≥2 points, and the accuracy of cognitive tasks improved by ≥15%; Effective: MoCA score improves by 1-2 points or MMSE score improves by 1 point, and cognitive task accuracy improves by 5%-14%; Invalid: MoCA score improvement <1 point and MMSE score no change, cognitive task accuracy improvement <5%.

[0030] 6. Construction of the EEG-Therapy Association Database: Data from previous clinical studies were collected, recording changes in EEG characteristics corresponding to different combinations of head acupoints and acupuncture techniques. The correlation between treatment parameters and EEG characteristics was analyzed, establishing a "Head Acupoint-Technique-EEG Feature" triplet knowledge base. The triplet relationships were stored in a directed graph format, and conditional probabilities between nodes were calculated using a Bayesian network. Clinical data was integrated to form typical association rules, and the data was continuously updated using an incremental learning algorithm to maintain the timeliness of the knowledge base. Based on the knowledge base, the most relevant EEG features for the current treatment plan were selected as auxiliary feature inputs to the model. During feature selection, semantic matching is first performed based on the cosine similarity between the conditional vector and the triples in the knowledge base to select EEG features with a similarity ≥ 0.6. Then, the feature weights are dynamically adjusted in combination with the model's prior knowledge. For example, the weight of the β wave feature corresponding to the twisting technique is increased by 30%. Finally, a 10-15 dimensional auxiliary feature vector is formed, which reduces the feature dimension by 40-60% while maintaining a correlation coefficient with the therapeutic effect ≥ 0.5 (p<0.01), effectively improving the model's inference efficiency and interpretability.

[0031] Secondly, embodiments of the present invention provide a system for evaluating the immediate efficacy of acupuncture at head acupoints in treating cognitive impairment following stroke based on electroencephalogram (EEG) signals, comprising: 1. User Interface: Receives the name of the acupoint on the head, the type of manipulation, and parameters, and generates a conditional vector using a built-in dictionary. It displays real-time EEG signal heatmaps, efficacy prediction results, and the contribution of key features, allowing doctors to intuitively understand changes in the patient's EEG and related information on treatment effectiveness.

[0032] 2. EEG Signal Acquisition Module: A 4-8 channel portable dry electrode EEG device is used, with electrodes strategically placed to avoid acupuncture points on the head. It simultaneously acquires resting-state and cognitive-task-state EEG signals before, during, and after acupuncture, marking the acupuncture time points to ensure that the acquired EEG signals accurately reflect the brain's electrical activity at each stage of treatment.

[0033] 3. Data Processing Module: This module preprocesses the EEG signals, removing oculomotor and electromyographic artifacts through independent component analysis and extracting α / θ / β time-frequency domain features. It also constructs an electrode space graph structure, defining brain region nodes and functional connectivity edges to prepare for subsequent feature extraction and analysis.

[0034] 4. Conditional Parameter Adjustment Module: Conditional vector encoder: Maps the names of acupoints on the head to unique heat codes, and normalizes the manipulation parameters to numerical codes.

[0035] Parameter generation network: Taking the conditional vector as input, it outputs two sets of dynamic parameters: GNN side weight offset and Transformer attention bias, which are used to enhance the spatial correlation of the brain regions corresponding to acupoints and highlight the temporal signals of the treatment stage.

[0036] 5. Feature Extraction Controller: Dynamic parameters are injected into the model, enabling the GNN to focus on the brain regions related to acupoints, and the Transformer to focus on the temporal dynamics of the treatment phase, thereby improving the model's ability to extract key features.

[0037] 6. EEG-Therapy Correlation Database: Stores the "head acupoint-manipulation-EEG characteristics" ternary set, including specific modulation data of different acupoints on prefrontal β waves and temporal θ waves, and the influence of manipulation parameters on brain region functional connectivity, providing prior knowledge for current treatment plans and assisting models to more accurately evaluate efficacy.

[0038] 7. Acupuncture Immediate Efficacy Evaluation Model: Based on Conditioned Spatiotemporal Transformer-GNN Architecture Input layer: Receives multichannel EEG signals, and signal preprocessing includes bandpass filtering and independent component analysis to remove electrooculogram artifacts.

[0039] Conditional GNN layers: dynamically adjust graph convolution kernel parameters to extract spatial features (such as resting-state functional connectivity strength) of brain regions such as the prefrontal-parietal lobe, and mine spatial association information between brain regions.

[0040] Conditioned Transformer layer: Adaptively adjusts attention weights, captures long-term temporal dependence of α / θ / β frequency bands before and after treatment, and analyzes the changes in EEG signals in the time dimension.

[0041] Multimodal fusion layer: By fusing spatiotemporal features and conditional vectors through a cross-attention mechanism, a multimodal vector containing EEG features, acupoint parameters, and scale scores is generated, fully integrating information from multiple aspects.

[0042] Output layer: Combining the MoCA / MMSE criteria, it outputs the levels of significant effect / effectiveness / ineffectiveness and visualizes key EEG features, providing doctors with intuitive and clear efficacy assessment results.

[0043] like Figure 3 As shown, this embodiment provides a condition vector generation method including: Acupoint Coding: The system has a built-in dictionary containing acupoints on the head such as Baihui, Shenting, Fengchi, and Temporal Three Needles, and uses 50-dimensional one-hot encoding to map acupoints. When mapping a single acupoint, only the corresponding brain region dimension is set to 1, and the rest are 0; when combining multiple acupoints, the one-hot vectors of each acupoint are concatenated to generate a higher-dimensional code.

[0044] Technique coding: A 3D unique thermal vector is used to represent the technique type, combined with a 2D numerical vector to achieve parameter normalization. Electroacupuncture frequency and intensity are normalized using a linear normalization formula. Mapped to the [0, 1] interval, and concatenated to form a 5-dimensional technique encoding vector.

[0045] Vector concatenation: The single acupoint conditional vector is concatenated from a 50-dimensional acupoint code and a 5-dimensional manipulation code to form a 55-dimensional feature vector. The combination of two acupoints generates a 105-dimensional vector. This vector is used to drive a parameter generation network, dynamically adjusting the weights of the graph convolution kernel to achieve targeted focusing of the model for extracting features from the target brain region.

[0046] like Figure 4 As shown, this embodiment provides a condition parameter adjustment module including: Dynamic parameter generation: GNN edge weight offset ΔWgnn enhances the functional connectivity weights of cognitively related brain regions such as the prefrontal-parietal-temporal-hippocampus based on acupoint encoding. For example, the Shen Ting acupoint increases the weight of the Fp1-Fp2 edge by 30%. Transformer attention bias ΔAattn highlights the temporal dependence of theta wave memory and beta wave executive function based on the technique parameters. For example, the twisting technique enhances the attention weight of the prefrontal beta wave by 25%.

[0047] Spatiotemporal feature extraction: The conditional GNN layer constructs a graph structure with a 10-20 system electrode distribution, and adjusts the adjacency matrix through ΔWgnn to extract spatial features such as the power of the theta wave in the prefrontal lobe and the synchronicity of the alpha wave in the parietal lobe; the conditional Transformer layer encodes the 30-minute time sequence signal, and enhances the long-term dependence of specific frequency bands in the treatment phase through ΔAattn, such as increasing the attention weight of the change in the P300 latency in the parietal lobe during the needle retention period by 40%.

[0048] Feature fusion: Integrating spatial features such as brain region functional connectivity matrix, temporal features such as task-state P300 latency change rate and resting-state alpha wave inhibition rate, conditional vector acupoint manipulation encoding vector, and clinical data MoCA score changes and the improvement in cognitive task accuracy.

[0049] like Figure 5 As shown, this embodiment also provides an electronic device for evaluating the immediate efficacy of acupuncture at head acupoints in treating cognitive impairment following stroke based on electroencephalogram (EEG) signals, comprising: At least one processor; At least one memory and EEG acquisition interface are provided. The memory stores an executable program to realize EEG signal processing, model inference and result output.

[0050] This embodiment provides a storage medium for evaluating the immediate efficacy of acupuncture at head acupoints in treating cognitive impairment following stroke based on electroencephalogram (EEG) signals. The medium stores a computer program, characterized in that when the program is executed by a processor, it implements the steps of the method for evaluating the immediate efficacy of acupuncture at head acupoints in treating cognitive impairment following stroke based on EEG signals, as described in any embodiment of the present invention.

Claims

1. A method for evaluating the immediate therapeutic effect of acupuncture on treating stroke cognitive impairment based on electroencephalogram signals, characterized in that, Comprise: Step 1: Multi-stage acquisition of patient's EEG signals before, during and after acupuncture, including resting state and cognitive task state, wherein each stage before, during and after acupuncture acquires at least 2 minutes of resting state signal and 3-5 minutes of cognitive task state signal, and marks the time nodes of acupuncture into the acupoint, acupuncture, needle retention and needle removal; Step 2: Obtain subjective cognitive evaluation data through Montreal Cognitive Assessment Scale MoCA or Mini-Mental State Examination MMSE, and record the task completion time and accuracy of memory, language or executive function test simultaneously; Step 3: Use insulated acupuncture needles with 4-8 lead portable dry electrode EEG equipment to collect signals in the area away from the head acupoints; Step 4: Map the head acupoint name to a 50-dimensional one-hot encoding vector through the built-in dictionary, convert the manipulation type to a 3-dimensional category encoding, normalize the parameters to a 2-dimensional numerical encoding, and concatenate to generate a 55-dimensional or 105-dimensional conditional vector to drive the parameter generation network to adjust the graph convolution kernel parameters and attention mechanism weights of the spatio-temporal Transformer-GNN model, making the spatio-temporal Transformer-GNN model sensitive to the alpha / theta / beta band features of the frontal lobe, parietal lobe and temporal lobe; Step 5: Extract EEG spatio-temporal features related to treatment methods and acupoints, and construct an acupuncture immediate efficacy evaluation model combined with the conditional vector to input the trained model to output the efficacy prediction results.

2. The method for evaluating the immediate therapeutic effect of acupuncture on cognitive impairment caused by stroke based on electroencephalogram signals according to claim 1, characterized in that, The collection of resting state information and cognitive task state information includes: The closed-eye resting state collection is used to obtain the basic electrical activity of the brain without visual input, highlighting the occipital alpha wave characteristics; the open-eye resting state collection is used to capture the brain arousal level under visual input, reflecting the frontal lobe beta wave activity; The cognitive task state collection includes EEG signals in visual Oddball paradigm, n-back paradigm and Stroop color conflict task, and the task accuracy and reaction time are recorded synchronously.

3. The method for evaluating the immediate therapeutic effect of acupuncture on cognitive impairment caused by stroke based on electroencephalogram according to claim 1, characterized in that, The conditional vector generation step includes: (1) Map the head acupoint through 50-dimensional one-hot encoding, where a single acupoint corresponds to a brain region dimension with 1, and multiple acupoints are concatenated to generate higher-dimensional encoding; (2) The manipulation type adopts 3-dimensional one-hot encoding of lifting and thrusting [1, 0, 0], twirling [0, 1, 0] and electroacupuncture [0, 0, 1], and the frequency and intensity of electroacupuncture are normalized into 2-dimensional numerical encoding to form a 5-dimensional manipulation encoding vector. The manipulation type adopts 3-dimensional one-hot encoding of lifting and thrusting [1, 0, 0], twirling [0, 1, 0] and electroacupuncture [0, 0, 1], and the frequency and intensity of electroacupuncture are normalized into 2-dimensional numerical encoding to form a 5-dimensional manipulation encoding vector.

4. The method for evaluating the immediate therapeutic effect of acupuncture on cognitive impairment caused by stroke based on electroencephalogram according to claim 1, characterized in that, The acupuncture immediate efficacy evaluation model includes: The conditional GNN layer takes the electrode spatial distribution as the graph structure, dynamically adjusts the graph convolution kernel parameters through the edge weight bias generated by the conditional vector, and enhances the functional connection weight of the frontal lobe-parietal lobe cognitive brain region; The conditional Transformer layer adjusts the self-attention mechanism weight through attention bias, highlighting the timing dependence of theta waves and beta waves; The multi-modal fusion layer fuses spatial features, temporal features and conditional vectors based on cross-attention mechanism, and the output layer outputs the levels of obvious effect, effective and ineffective combined with MoCA / MMSE judgment standard.

5. The method for evaluating the immediate therapeutic effect of acupuncture on cognitive impairment caused by stroke based on electroencephalogram according to claim 1, characterized in that, The efficacy prediction result judgment standard is: Obvious effect: MoCA score improvement ≥3 points or MMSE score improvement ≥2 points, and cognitive task accuracy improvement ≥15%; Effective: MoCA score improvement 1-2 points or MMSE score improvement 1 point, and cognitive task accuracy improvement 5%-14%; Ineffective: MoCA score improvement <1 point and MMSE score unchanged, and cognitive task accuracy improvement <5%.

6. The method for evaluating the immediate therapeutic effect of acupuncture on cognitive impairment caused by stroke based on electroencephalogram according to claim 1, characterized in that, The EEG-treatment association database is constructed as an instant efficacy evaluation model database of acupuncture, a directed graph is used to store the triplets including head acupoints, manual parameters and EEG characteristics, the conditional probability is calculated through the Bayesian network, the knowledge base is updated by using the incremental learning algorithm, the EEG characteristics are screened based on the cosine similarity ≥ 0.6, the feature weight is adjusted in combination with the prior knowledge of the model, and a 10-15-dimensional auxiliary feature vector is formed.

7. A system for evaluating the immediate therapeutic effect of acupuncture on head acupoints on stroke cognitive impairment based on electroencephalogram signals, characterized in that, The method for evaluating the instant efficacy of acupuncture in treating cognitive impairment of stroke based on the EEG signal according to any one of claims 1-6 comprises the following steps: A user interface receives the names of head acupoints and manual parameters, generates a conditional vector, and displays an EEG heat map and key feature contribution in real time; An EEG signal acquisition module uses a 4-8 lead dry electrode device, avoids electrode layout at head acupoints, and marks the time points of the acupuncture stage; A data processing module removes artifacts through independent component analysis, extracts α / θ / β time-frequency domain features, and constructs an electrode spatial graph structure; A conditional parameter adjustment module outputs GNN edge weight offset and Transformer attention bias, and a feature extraction controller drives the model to focus on acupoint-related brain regions and time sequence frequency bands; An EEG-treatment association database stores triplet knowledge, and an instant efficacy evaluation model of acupuncture outputs an efficacy grade based on a conditional spatio-temporal Transformer-GNN architecture.

8. The system for evaluating the immediate therapeutic effect of acupuncture on cognitive impairment caused by stroke based on electroencephalogram according to claim 7, characterized in that, The conditional parameter adjustment module comprises: A conditional vector encoder generates a 55-dimensional / 105-dimensional conditional vector, and a parameter generation network outputs two groups of dynamic parameters for enhancing the spatial correlation of acupoint-related brain regions and highlighting the time sequence signals in the treatment stage.

9. The system for evaluating the immediate therapeutic effect of acupuncture on cognitive impairment caused by stroke based on electroencephalogram according to claim 7, characterized in that, The training method of the instant efficacy evaluation model of acupuncture comprises the following steps: Collect EEG signals, treatment parameters and subjective evaluation data of at least 100 patients to construct a triplet training data set; adjust the model parameters by jointly optimizing the classification loss and the regression loss, and use L1 regularization to constrain the dynamic parameter output.

Images