Electroacupuncture intervention effect evaluation system based on multi-modal physiological data fusion

Abstract

The invention relates to the technical field of electro-acupuncture intervention evaluation, and discloses an electro-acupuncture intervention effect evaluation system based on multi-modal physiological data fusion, the system comprises a central processing unit, the central processing unit is used for executing: performing electro-acupuncture stimulation operation on a subject through an electro-acupuncture intervention device; the method comprises the following steps: acquiring a stimulation period physiological signal in a stimulation process by a first signal acquisition device, executing a stimulation period physiological feature extraction algorithm by a processing device, and calculating a first effect index in an electroacupuncture intervention period according to the algorithm; a second signal acquisition device acquires a post-intervention resting state physiological signal, and a processing device executes a post-intervention feature fusion algorithm and calculates a post-intervention second effect index; and the processing device performs multi-modal data fusion on the two effect indexes to generate a comprehensive intervention evaluation parameter. According to the method, the short-term instant effect and the long-term persistent effect of electroacupuncture intervention can be comprehensively captured, multi-modal physiological information is fused, and the comprehensiveness and objectivity of an evaluation result are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of electroacupuncture intervention evaluation, in particular to an electroacupuncture intervention effect evaluation system based on multi-modal physiological data fusion. BACKGROUND

[0002] As a therapy combining traditional acupuncture and modern electrical stimulation technology, electroacupuncture intervention is increasingly widely used in the fields of pain relief and functional rehabilitation. Scientific evaluation of its intervention effect has become an important direction to promote the standardization development of this therapy. Currently, the evaluation methods for electroacupuncture intervention effect mostly focus on single-dimensional data collection and analysis, which is difficult to comprehensively reflect the comprehensive influence of intervention on human physiological state.

[0003] Some evaluation methods only focus on physiological signals during electroacupuncture stimulation, such as obtaining heart rate, electroencephalogram and other data during stimulation period through single electrophysiological monitoring equipment, and calculating effect indicators based on these single signals. Although this method can capture the immediate physiological changes during intervention, it cannot reflect the sustained adjustment of human physiological state after intervention, and this sustained adjustment is often the key basis for evaluating the long-term effect of electroacupuncture intervention. Relying only on immediate signals may lead to one-sided evaluation results and make it difficult to accurately judge the actual effectiveness of intervention.

[0004] Some other evaluation methods focus on the analysis of resting state physiological signals after intervention, and evaluate the effect by monitoring the changes of physiological indicators within a certain period after intervention. However, this method ignores the dynamic response characteristics of physiological signals during intervention, and there are differences in the immediate response of different subjects to electroacupuncture stimulation. Some subjects may have obvious physiological improvement during stimulation, while some subjects may have gradual improvement after intervention. If only the signals after intervention are analyzed, it is difficult to distinguish such differences and establish a correlation between the physiological changes after intervention and the response during stimulation, resulting in incomplete evaluation logic and inability to comprehensively cover the short-term immediate effect and long-term sustained effect of intervention.

[0005] Existing evaluation methods generally lack the idea of multi-modal data fusion, and mostly rely on single type of physiological signal (such as monitoring only heart rate or analyzing only electroencephalogram) for index calculation. Human body is a complex physiological system, and electroacupuncture intervention will affect the function of multiple systems such as circulation, nervous system and endocrine system. Single modal physiological signal cannot comprehensively reflect the coordinated changes of multiple systems, and effect indicators calculated based on single signal have limited representativeness, making it difficult to comprehensively reflect the influence of intervention on the overall physiological state of human body. This leads to insufficient accuracy and reliability of evaluation results, and cannot provide comprehensive reference basis for the optimization of electroacupuncture intervention scheme and clinical application. SUMMARY

[0006] The present application aims to provide an electroacupuncture intervention effect evaluation system based on multi-modal physiological data fusion to solve the problems raised in the background art.

[0007] To achieve the above-mentioned purpose, the present application provides an electroacupuncture intervention effect evaluation system based on multi-modal physiological data fusion, which comprises a central processing unit, wherein the central processing unit is used to execute: performing an electroacupuncture stimulation operation on a subject by an electroacupuncture intervention device; acquiring a stimulation period physiological signal during the electroacupuncture stimulation operation by a first signal acquisition device; executing a stimulation period physiological feature extraction algorithm by a processing device, and calculating a first effect index during the electroacupuncture intervention based on the stimulation period physiological signal; acquiring a resting state physiological signal after the electroacupuncture intervention by a second signal acquisition device; executing a post-intervention feature fusion algorithm by the processing device, and calculating a second effect index after the intervention based on the resting state physiological signal after the electroacupuncture intervention; performing multi-modal data fusion on the first effect index and the second effect index by the processing device to generate a comprehensive intervention evaluation parameter.

[0008] Preferably, the method further comprises: constructing a stimulation response prediction model by the processing device according to a pre-set historical intervention database, wherein the historical intervention database contains reference stimulation period physiological signals of reference subjects and corresponding reference intervention effect indexes; executing the stimulation response prediction model by the processing device to predict a real-time intervention effectiveness parameter based on the stimulation period physiological signal.

[0009] Preferably, the execution of the stimulation period physiological feature extraction algorithm by the processing device comprises: identifying electroacupuncture response waveform features in the stimulation period physiological signal; extracting time domain distribution characteristics and frequency domain energy characteristics of the electroacupuncture response waveform features; generating the first effect index according to the time domain distribution characteristics and the frequency domain energy characteristics.

[0010] Preferably, the execution of the post-intervention feature fusion algorithm by the processing device comprises: separating autonomic nervous regulation features and body fluid metabolism features in the resting state physiological signal after the electroacupuncture intervention; calculating a variability index of the autonomic nervous regulation features and a concentration change index of the body fluid metabolism features; fusing the variability index and the concentration change index to generate the second effect index.

[0011] Preferably, the multi-modal data fusion of the first effect indicator and the second effect indicator by the processing device comprises: constructing a stimulus-response feature vector corresponding to the first effect indicator and a steady-state regulation feature vector corresponding to the second effect indicator; performing a feature layer fusion algorithm to map the stimulus-response feature vector and the steady-state regulation feature vector to a joint feature space; generating the comprehensive intervention evaluation parameter according to a fusion feature value of the joint feature space.

[0012] Preferably, the method further comprises: establishing an effect evaluation benchmark curve by the processing device according to the historical electroacupuncture intervention data of the reference subject; dynamically comparing the comprehensive intervention evaluation parameter with the effect evaluation benchmark curve by the processing device; outputting an intervention parameter adjustment instruction to the electroacupuncture intervention device according to the dynamic comparison result by the processing device.

[0013] Preferably, the constructing a stimulus-response prediction model comprises: extracting a plurality of groups of reference stimulus period physiological signal features and associated intervention effect indicators from the historical intervention database; performing a feature selection algorithm to screen core stimulus-response features; constructing the stimulus-response prediction model according to the mapping relationship between the core stimulus-response features and the associated intervention effect indicators.

[0014] Preferably, the performing a feature layer fusion algorithm comprises: adopting a nonlinear weighting mechanism to assign importance to the stimulus-response feature vector and the steady-state regulation feature vector; dimensionally reducing the weighted feature vectors to the joint feature space by a feature projection algorithm.

[0015] Preferably, the dynamically comparing the comprehensive intervention evaluation parameter with the effect evaluation benchmark curve by the processing device comprises: identifying a phase-sensitive interval of the effect evaluation benchmark curve; calculating a deviation degree of the comprehensive intervention evaluation parameter in the phase-sensitive interval; quantifying an intervention parameter adjustment magnitude according to the deviation.

[0016] Preferably, the method further comprises: continuously acquiring iterative stimulus period physiological signals and iterative post-intervention resting state physiological signals of multiple rounds of electroacupuncture intervention by the processing device; updating, by the processing device, a prediction weight of the stimulation response prediction model and a benchmark threshold of the effect evaluation benchmark curve; optimizing, by the processing device, generation logic of the intervention parameter adjustment instruction according to the updated prediction weight and benchmark threshold.

[0017] Compared with the prior art, the present application has the following beneficial effects: In the physiological signal acquisition link, the method not only acquires the stimulation period physiological signal in the electroacupuncture stimulation operation process through the first signal acquisition device, but also acquires the resting state physiological signal after intervention through the second signal acquisition device, realizing comprehensive capture of immediate physiological changes during intervention and sustained physiological adjustment after intervention. This phased acquisition method can completely cover the short-term effect and long-term effect of electroacupuncture intervention, avoids the evaluation one-sidedness caused by focusing on a single stage signal, and makes the subsequent effect index based on signal calculation more fully reflect the influence of intervention on the physiological state of the human body, whether it is the immediate physiological response during stimulation or the sustained optimization of the physiological state after intervention, which can be included in the evaluation system to provide more abundant basic data for subsequent comprehensive evaluation.

[0018] In the effect index calculation aspect, the processing device respectively executes corresponding algorithms for physiological signals in different stages: executes a stimulation period physiological feature extraction algorithm for the stimulation period physiological signal to calculate a first effect index, and executes a post-intervention feature fusion algorithm for the resting state physiological signal after intervention to calculate a second effect index. This targeted algorithm design can fully exploit the features of physiological signals in different stages. The dynamic change feature of the stimulation period physiological signal is more obvious, and the dedicated stimulation period physiological feature extraction algorithm can accurately capture the immediate response law in the signal. The resting state physiological signal after intervention focuses more on stability and trend changes, and the post-intervention feature fusion algorithm can effectively extract such sustained adjustment features. The combination of the two algorithms makes the calculated first effect index and second effect index respectively have clear pertinence and representativeness, which can accurately reflect the physiological effect in the corresponding stage and lay a high-quality index foundation for subsequent multi-modal fusion.

[0019] In the data fusion and comprehensive evaluation stage, the processing device performs multimodal data fusion of the first and second effect indicators to generate comprehensive intervention evaluation parameters. This fusion method breaks through the limitations of existing evaluation methods that rely on single signals or single-stage indicators, organically combining immediate effect indicators during the intervention process with sustained effect indicators after the intervention, while also integrating multi-system physiological change information reflected by different modal physiological signals. Through multimodal fusion, scattered, single-dimensional effect indicators can be integrated into comprehensive parameters with overall representativeness. These parameters not only reflect the immediate impact of electroacupuncture intervention on the human physiological state but also reflect the sustainability of the intervention effect, while encompassing the synergistic changes of multiple systems such as the circulatory and nervous systems. This makes the final evaluation results more consistent with the complex characteristics of the human physiological system, and more objectively and comprehensively reflects the actual effect of electroacupuncture intervention, providing a more comprehensive and reliable reference for the evaluation and adjustment of electroacupuncture intervention programs and the judgment of effects in clinical applications. Attached Figure Description

[0020] Figure 1 This is a schematic diagram illustrating the working principle of the electroacupuncture intervention effect evaluation system based on multimodal physiological data fusion described in this invention. Figure 2 This is a flowchart for extracting physiological characteristics during the stimulation period; Figure 3 A flowchart for feature fusion after intervention; Figure 4 A flowchart for dynamic comparison and intervention parameter adjustment; Figure 5 This is a flowchart of the model and benchmark update iterations. Detailed Implementation

[0021] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] Please see Figure 1The present application provides an electroacupuncture intervention effect evaluation system based on multi-modal physiological data fusion. The system includes a central processing unit for performing: performing electroacupuncture stimulation operation on a subject by an electroacupuncture intervention device, which adopts standard electroacupuncture output parameters such as frequency, intensity and waveform to ensure the consistency of stimulation. Through a first signal acquisition device such as a multi-channel physiological signal sensor, physiological signals during the stimulation period are obtained during the electroacupuncture stimulation operation, including electrocardiogram, electroencephalogram and electromyogram. Through a processing device such as an embedded computer system, a stimulation period physiological feature extraction algorithm is executed to calculate a first effect index during electroacupuncture intervention based on the stimulation period physiological signals, which reflects the immediate physiological response during stimulation. Through a second signal acquisition device such as an infrared spectrometer or a body fluid analyzer, resting physiological signals after electroacupuncture intervention are obtained, including heart rate variability, blood oxygen saturation and body fluid composition data. Through the processing device, a post-intervention feature fusion algorithm is executed to calculate a second effect index after intervention based on the resting physiological signals after electroacupuncture intervention, which represents the recovery state after intervention. Through the processing device, the first effect index and the second effect index are subjected to multi-modal data fusion to generate a comprehensive intervention evaluation parameter, which integrates the physiological changes during the stimulation period and after the intervention, and provides comprehensive effect evaluation.

[0023] Example 1: see Figure 2 In an embodiment of the electroacupuncture intervention effect evaluation system based on multi-modal physiological data fusion, the processing device constructs a stimulation response prediction model according to a pre-set historical intervention database. The historical intervention database is established through long-term clinical observation and data acquisition, in which a plurality of reference stimulation period physiological signals generated by reference subjects during standardized electroacupuncture intervention are systematically stored, as well as strictly corresponding reference intervention effect indexes. These reference intervention effect indexes are quantitative results calculated by previously established evaluation methods, which can reflect the effect intensity and mode of electroacupuncture intervention under different physiological states. The construction of the database follows strict data management specifications to ensure the integrity, consistency and traceability of the data. Each data entry contains the anonymous identification of the subject, the intervention parameters, the collected original physiological signal data and the verified effect index calculation results.

[0024] The process of constructing the stimulus-response prediction model is a data-driven modeling procedure. The processing device first extracts multiple sets of reference stimulus session physiological signal features and their associated intervention effect indicators from the historical intervention database. This extraction process is not a simple data retrieval, but involves reanalysis and recalculation of the original signals to ensure consistency and comparability with the current real-time processing procedure's feature extraction algorithms. Subsequently, a feature selection algorithm is executed to screen the core stimulus-response features. This feature selection algorithm evaluates the statistical association strength and nonlinear dependency between all potential features and the target effect indicators, aiming to identify the most discriminative and robust feature subset from the high-dimensional feature set, thereby reducing model complexity, avoiding overfitting, and improving model interpretability. According to the mapping relationship between the screened core stimulus-response features and the associated intervention effect indicators, the final stimulus-response prediction model is constructed. This mapping relationship is usually achieved through regression algorithms or function approximation techniques in machine learning, aiming to establish a mathematical function or computational graph that can accurately predict the effect indicators from the input physiological features. The structure and parameters of the model are trained and optimized through historical data, enabling it to have good generalization ability.

[0025] After the model is constructed, the stimulus-response prediction model is executed by the processing device, which can predict the real-time intervention effectiveness parameter based on the real-time collected stimulus session physiological signals. This prediction process is dynamic, and the physiological features of the current subject are input into the trained model, and the model output is the quantitative estimate of the current intervention effectiveness. This real-time intervention effectiveness parameter serves as a forward-looking indicator, providing immediate feedback for evaluating the effectiveness of the current stimulation operation.

[0026] In the process of executing the stimulus session physiological feature extraction algorithm by the processing device to calculate the first effect indicator, the core is to identify and quantify the specific information patterns related to the electroacupuncture response contained in the stimulus session physiological signals. The algorithm first aims to identify the electroacupuncture response waveform features in the stimulus session physiological signals. These features usually manifest as specific morphological changes in the signal in the time domain, such as evoked potentials, specific electroencephalogram rhythm synchronization or desynchronization patterns, or specific recruitment patterns in electromyographic signals after electroacupuncture stimulation is applied. The identification process may involve template matching, peak detection, or waveform morphology-based segmentation algorithms, aiming to separate the stimulus-related response components from the background physiological activity and noise in the signal.

[0027] After successfully identifying the relevant response waveform features, the algorithm further extracts the time-domain distribution characteristics of these electroacupuncture response waveform features. The analysis of time-domain distribution characteristics focuses on the occurrence regularity and statistical properties of waveform features on the time axis, such as response latency, response duration, occurrence frequency of specific waveform events, sequence of peak amplitude, and amplitude trend over time, etc. These time-domain parameters directly reflect the response speed and intensity pattern of the nervous system to electroacupuncture stimulation. At the same time, the algorithm extracts the frequency-domain energy characteristics of these electroacupuncture response waveform features in parallel. Frequency-domain analysis examines the distribution of response energy at different frequency bands by transforming the signal into the frequency domain. Common analyses include calculating the power spectral density of specific frequency bands (such as delta, theta, alpha, beta, and gamma bands of electroencephalogram), frequency band energy ratio, and dynamic migration of energy during intervention. Frequency-domain characteristics help reveal the regulatory effect of stimulation on brain oscillation activity or autonomic rhythm.

[0028] Finally, based on the extracted time-domain distribution characteristics and frequency-domain energy characteristics, the algorithm generates a first effect indicator through a predefined synthesis rule or calculation formula. This indicator is not a simple presentation of a single parameter, but often a composite indicator that integrates multi-dimensional information, and its calculation process may involve weighted combination, normalization processing, and empirical-based scaling conversion of time-domain and frequency-domain characteristic values. The generated first effect indicator aims to quantify the direct physiological effect intensity and pattern of electroacupuncture stimulation operation at the intervention period, providing key first modality data input for subsequent multi-modal data fusion.

[0029] The specific implementation process of the predefined synthesis rule: ① Feature weight determination rule: From the historical intervention database, select the time-domain / frequency-domain features with high correlation. First, calculate the correlation coefficient |r| between each time-domain / frequency-domain feature and the reference intervention effect indicator through Pearson correlation analysis, and retain the high correlation features with |r| ≥ 0.6 (exclude noise features with no significant correlation with intervention effect, such as random baseline fluctuation features in the stimulation period physiological signal); Second, use multivariate linear stepwise regression method, with reference intervention effect indicator as dependent variable and high correlation features as independent variables, to screen features with significant contribution to intervention effect (statistical test P < 0.05), and output the standardized regression coefficient of each feature. This coefficient is the feature weight, and the sum of all feature weights is 1, to eliminate the influence of different feature dimensions on weight allocation. ② Feature preprocessing rule: Perform Min-Max normalization on the selected time-domain / frequency-domain features to unify the feature values to the [0, 1] interval, avoiding calculation bias caused by feature unit differences. The normalization formula is as follows: , where, represents the original measurement value of the time-domain or frequency-domain feature, and respectively represent the minimum and maximum values of the feature in the historical intervention database, ensuring that the pre-processing benchmark is consistent with the historical data. ③ Index synthesis logic rule: the first effect index is generated by using the two-step method of weighted summation + standardization mapping, taking into account the quantitative accuracy and clinical interpretation convenience. In the first step, the weighted feature sum is calculated, that is, the sum of the product of each normalized feature value and the corresponding feature weight, and the calculation formula can be expressed as . Wherein: p is the total number of high correlation features, is the weight of the kth feature, is the normalized value of the kth feature; in the second step, the weighted feature sum is mapped to the standardization interval [0, 100] (0 represents no stimulation response, and 100 represents optimal stimulation response), and the mapping benchmark is the corresponding relationship between the weighted feature sum and the reference intervention effect index in the historical intervention database.

[0030] The specific calculation formula for generating the first effect index is: , wherein: is the first effect index, representing the quantitative result of the real-time physiological response during the stimulation period of the electro-acupuncture intervention, and the value range is [0, 100]; is the number of time-domain high correlation features, : the number of frequency-domain high correlation features, both of which are determined according to the signal type from the feature screening results of the historical intervention database; is the weight of the ith time-domain high correlation feature; is the normalized value of the ith time-domain high correlation feature; is the weight of the jth frequency-domain high correlation feature; is the normalized value of the jth frequency-domain high correlation feature.

[0031] Example 2: refer to Figure 3 The intervention post-processing feature fusion algorithm executed by the processing device is the core link for generating the second effect index. The input of the algorithm is the resting state physiological signal of the electro-acupuncture intervention obtained by the second signal acquisition device. These signals are collected in a specific resting time period after the electro-acupuncture stimulation operation is completed, and their types can include electrocardiogram, electroencephalogram, electrodermal, respiratory wave, and body fluid spectrum or electrochemical data obtained by biochemical sensors. The primary task of the algorithm is to separate the feature components representing different physiological regulation levels, namely autonomic nervous regulation features and body fluid metabolism features, from these multi-channel resting state signals.

[0032] The separation of autonomic regulation features mainly relies on the in-depth analysis of electrocardiogram and electrodermal signals. The processing device pre-processes the collected electrocardiogram signals, including R-wave detection, heart rate interval sequence extraction, and signal quality verification. From the high-quality heart rate interval sequence, a series of indicators reflecting the autonomic nervous function status are further calculated, which constitute the main body of the autonomic regulation features. For electrodermal signals, the slow fluctuation of the base conductance level and the rapid phase fluctuation are analyzed, which are closely related to sympathetic nervous activity. The separation of body fluid metabolism features is aimed at data from biochemical sensors. These data may reflect changes in the concentration of specific molecules in blood, saliva, or interstitial fluid, such as glucose, lactate, cortisol, or certain neurotransmitter metabolites. The processing device uses signal processing techniques such as spectral peak analysis, chromatographic peak identification, or electrochemical signal calibration to extract quantitative information related to metabolic activity from raw sensor data.

[0033] After successfully separating the two major features, the algorithm enters the quantification calculation phase. For autonomic regulation features, the variability index is calculated. Heart rate variability analysis is the core of calculating this index, usually including time domain and frequency domain analysis methods. Time domain analysis calculates statistical indicators of the heart rate interval sequence, such as the standard deviation of the sequence, the root mean square of adjacent interval differences, etc., which comprehensively reflect the overall regulation ability of the autonomic nervous system to heart rhythm. Frequency domain analysis calculates the power values or power ratio values of the signal in specific frequency bands such as ultra-low frequency, low frequency, and high frequency through power spectrum estimation methods, which helps to evaluate the balance state of sympathetic and parasympathetic nervous activity. The variability of electrodermal signals is quantified by calculating their fluctuation amplitude, fluctuation frequency, etc. For body fluid metabolism features, the concentration change index is calculated. The core of this index is to quantify the relative change or absolute change amount of the concentration of a specific substance before and after the intervention. The processing device compares the concentration value measured after the intervention with the individual baseline value or the group reference range, calculates the difference, percentage change, or uses standardized scores to represent it. For metabolic data with time series, it is also possible to calculate the rate or trend slope of the concentration change.

[0034] The process of generating the second effect indicator is an information fusion process. The algorithm takes the calculated variability indicator and concentration change indicator as inputs, and integrates them through a pre-defined fusion framework. This framework needs to handle indicators of different sources and different dimensions. A common practice is to first normalize or standardize each indicator to eliminate dimension differences. Subsequently, each indicator is assigned an appropriate weight according to its relative importance in evaluating the steady-state regulation effect after the intervention. The weight assignment can be based on prior knowledge, or learned from historical data using a data-driven method. Finally, all weighted indicators are combined into a single scalar value or a low-dimensional feature vector, i.e., the second effect indicator, through an aggregation function such as weighted linear combination or a nonlinear mapping function. This indicator aims to comprehensively quantify the new steady-state level of the autonomic nervous system function and the body fluid internal environment of the subject after the electroacupuncture intervention.

[0035] After the first effect indicator and the second effect indicator are calculated, the two are fused by the processing device to generate a comprehensive intervention evaluation parameter. This process begins with the construction of a feature vector. The processing device constructs the stimulation period response information represented by the first effect indicator into a stimulation response feature vector. Each dimension of this vector is composed of the various time-domain and frequency-domain feature values extracted during the calculation of the first effect indicator, or their reduced-dimension or abstracted representations. Similarly, the post-intervention steady-state regulation information represented by the second effect indicator is constructed into a steady-state regulation feature vector. The dimensions of this vector are derived from the variability indicator and the concentration change indicator described above, or their further transformed forms.

[0036] A feature layer fusion algorithm is executed. The goal of this algorithm is to effectively integrate the two feature vectors originating from different physiological processes and different time stages (stimulation period and post-intervention) into a unified mathematical space, i.e., the joint feature space, for subsequent analysis. Directly concatenating the two vectors is a simple method, but a more sophisticated approach would consider the interaction and redundancy between features. For example, a nonlinear transformation based on kernel methods can be used to map the two vectors into a high-dimensional space to capture their complex interrelationships; or a canonical correlation analysis technique can be used to find the projection direction with the highest correlation between the two sets of features, and then perform fusion based on this direction. Through this mapping, the originally heterogeneous multi-modal features are transformed into a set of fusion feature values in the joint feature space, which collectively carry information from both the stimulation response and the steady-state regulation.

[0037] The comprehensive intervention evaluation parameter is generated according to the fusion feature values obtained in the joint feature space. The generation method depends on the subsequent application requirements. If a total score is needed, the fusion feature values can be weighted and summed or input into a regression model. If a richer description is needed, the parameter itself can also be a low-dimensional vector, with different dimensions reflecting different aspects of the intervention effect. This comprehensive intervention evaluation parameter realizes the multi-angle and cross-period comprehensive evaluation of the intervention effect of electroacupuncture, and its information content surpasses any single modality or single period index.

[0038] Embodiment 3: refer to Figure 4 The processing device establishes an effect evaluation benchmark curve according to the electroacupuncture intervention history data of the reference subjects. The historical data is derived from the complete data set of multiple reference subjects after completing the complete electroacupuncture intervention process collected by the long-term system, including the stimulation period physiological signals collected in the intervention process, the resting physiological signals after intervention, and the finally calculated comprehensive intervention evaluation parameters. These data are strictly quality controlled, screened and time aligned to form a standardized time series data set. The process of establishing the benchmark curve adopts the time series modeling method, arranges the intervention effect data of each subject in time sequence, and smoothes the data by sliding window average method or Gaussian process regression to eliminate the influence of individual short-term fluctuations. Then the quantile regression algorithm is used for the group data to calculate the statistical distribution of the effect parameter at different intervention stages, and finally the trajectory of the 50th percentile (median) is taken as the main body of the benchmark curve, and the 25th and 75th percentile curves are recorded as variable range references. The benchmark curve can thus represent the expected trajectory and normal fluctuation range of the effect parameter over time under typical intervention conditions.

[0039] The processing device dynamically compares the real-time calculated comprehensive intervention evaluation parameter with the effect evaluation benchmark curve. This comparison process is continuous, and as the number of interventions increases, all historical comprehensive intervention evaluation parameters of the current subject are arranged in time sequence to form a sequence, which is matched and analyzed with the benchmark curve. The core of dynamic comparison is to quantify the similarity and difference between the current sequence and the benchmark curve. The methods adopted include calculating the difference values of the two sequences at the same time points, and analyzing the consistency of the morphological trends of the two sequences. Among them, the quantification of the morphological difference of the sequence can be realized by calculating the dynamic time warping distance, which can effectively handle the nonlinear stretching and shrinking of the two sequences on the time axis. The output of the comparison is a deviation degree index that changes over time, reflecting the matching degree between the current intervention effect and the expected benchmark.

[0040] The intervention parameter adjustment instruction is outputted to the electro-acupuncture intervention device according to the dynamic comparison result by the processing device. The output process maps the deviation degree index to a specific device parameter adjustment suggestion based on a preset decision rule. The decision rule can adopt a fuzzy logic system or a rule-based system, for example: when the deviation degree is continuously positive and exceeds a threshold value, an instruction to reduce the stimulation intensity is generated; when the deviation degree is continuously negative, an instruction to increase the stimulation frequency is generated; when the deviation degree fluctuates slightly around zero, the current parameters are maintained unchanged. The generation of the instruction takes into account the amplitude, duration and trend of the deviation degree, ensuring the gradualness and safety of the adjustment.

[0041] When constructing the stimulation response prediction model, a plurality of sets of reference stimulation period physiological signal features and associated intervention effect indicators are extracted from a historical intervention database. The extraction process includes data query, feature calculation and data alignment. The historical intervention database uses a relational database management system and stores a complete data chain including original signals, preprocessed signals, feature extraction results and effect indicators. The data subset under specific intervention conditions is extracted through structured query language to ensure the relevance and consistency of the data. The extracted features include time domain features (such as mean, variance, zero-crossing rate), frequency domain features (such as spectral energy, entropy) and nonlinear features (such as fractal dimension, Lyapunov exponent), which together constitute a multidimensional representation of the physiological response during the stimulation period.

[0042] A feature selection algorithm is executed to screen core stimulation response features. This process uses a combination of statistical tests and machine learning methods. First, the correlation coefficient, mutual information or p-value of the hypothesis test between each feature and the intervention effect indicator is calculated to evaluate the individual prediction ability of the feature. Subsequently, a recursive feature elimination algorithm is used, with the performance of the prediction model (such as mean square error or coefficient of determination) as the evaluation standard, to iteratively remove the least contributing features, finally retaining a smaller but highly predictive feature subset. This process not only reduces the data dimension, improves the model efficiency, but also enhances the interpretability and generalization ability of the model.

[0043] A stimulation response prediction model is constructed according to the mapping relationship between the core stimulation response features and the associated intervention effect indicators. The mapping relationship is established through a supervised learning algorithm, with the core features as input variables and the intervention effect indicators as target variables. Common algorithms include support vector regression, random forest regression or neural networks. Model training uses k-fold cross-validation to adjust hyperparameters and evaluate performance to prevent overfitting. The final prediction model can accurately estimate the intervention effect indicators from new stimulation period physiological signal features, providing important input for real-time effect evaluation. The loss function in the model training process is as follows: ; Where: represents the weighted average loss value, represent the total number of training samples, represents the actual intervention effect indicator value of the th sample, represents the predicted value of the model for the th sample, represents the intervention time point corresponding to the th sample, represents the mean value of the intervention time points of all samples, represents the standard deviation of the intervention time points of all samples. The characteristic of this loss function is to measure the prediction accuracy by relative error, and at the same time, introduce a time weighting factor, so that the model is more sensitive to the prediction accuracy of a specific time region.

[0044] In the implementation of the feature layer fusion algorithm, a nonlinear weighting mechanism is used to assign importance to the stimulus response feature vector and the steady-state regulation feature vector. This mechanism dynamically assigns weights based on the influence of each feature dimension on the final evaluation result in historical data. Take a specific example: suppose that after an intervention, the stimulus response feature vector extracted by the processing device contains five dimensions of values, representing the brain electrical response energy in different frequency bands during the stimulus; the steady-state regulation feature vector contains three dimensions of values, representing the time domain, frequency domain indicators of heart rate variability and the concentration change rate of a certain body fluid metabolite after the intervention. The system first calculates the correlation strength between each feature dimension and the ideal intervention effect pattern in the historical database. The stronger the correlation, the higher the basic weight of the dimension. Then an adjustment factor based on the stability of the feature value is introduced. The factor is obtained by calculating the coefficient of variation of the feature in recent interventions. The weight of the dimension with smaller fluctuation of feature value will be further enhanced, because it is considered to have higher reliability. The final weight of each feature is the product of the basic weight and the adjustment factor. This process is completely automatic and does not require human intervention.

[0045] The weighted feature vector is reduced to the joint feature space by the feature projection algorithm. Continuing the above example, the eight weighted dimensions (five from stimulus response and three from steady-state regulation) are input into a trained autoencoder neural network. The network has been trained using a large amount of historical intervention data, and the number of neurons in its hidden layer is set to three, which means that the eight weighted features will be compressed into three fusion feature values. These three fusion feature values constitute a point in the joint feature space, where each dimension is a complex nonlinear combination of the original features, which can more effectively represent the overall characteristics of the intervention effect. The projection process completely preserves the important relationships between the original features while eliminating redundant information.

[0046] The comprehensive intervention evaluation parameter is dynamically compared with the effect evaluation benchmark curve by the processing device. The effect evaluation benchmark curve is established by analyzing a large amount of historical intervention data, which describes the ideal change trajectory of the comprehensive intervention evaluation parameter over time in a typical successful intervention process. The curve is usually divided into several phase-sensitive intervals with different clinical significance. For example, there may be a rapid response interval at the beginning of the intervention, at which time the parameter should show a significant upward trend; in the middle, it may enter a platform adjustment interval, and the change tends to be stable; in the later period, it may enter the consolidation interval, and the parameter is required to be stable within a certain range. The system automatically identifies the specific phase-sensitive interval in which the current intervention time point is located, because the expectations of the parameter value are different in different intervals.

[0047] The deviation degree of the comprehensive intervention evaluation parameter in the phase-sensitive interval is calculated. This calculation not only considers the absolute difference between the parameter value and the current point of the benchmark curve, but also considers the degree of agreement between the recent parameter change trend and the expected trend of the benchmark curve. For example, in the rapid response interval, even if the current absolute difference is not large, if the parameter rising slope is significantly lower than the expected slope of the benchmark curve, it will be calculated as a positive deviation degree; on the contrary, if the parameter value fluctuates greatly in the consolidation interval, even if the mean value is close to the benchmark, it will also produce a deviation degree due to instability. The calculation result of the deviation degree is quantitative, and the numerical value directly reflects the difference between the current intervention effect and the expected effect.

[0048] The intervention parameter adjustment value is quantified according to the deviation degree. The system internally pre-stores a deviation degree-adjustment value mapping table, which is formulated by experts on the basis of a large amount of clinical experience and optimized through historical data verification. The mapping relationship is usually nonlinear, for example, for small deviations, a fine adjustment may be recommended, and for continuously increasing deviations, a larger parameter modification may be recommended. The adjustment value is finally converted into specific electric acupuncture device control instructions, such as increasing the current intensity by a certain milliampere or adjusting the stimulation frequency by a certain hertz. These instructions are sent in real time to the electric acupuncture intervention device through the digital interface, realizing online optimization and adjustment of the intervention parameter.

[0049] Table 1: Feature vector weighting table.

[0050]

[0051] Referring to Table 1, the table shows a specific numerical case of the feature weighting process. As can be seen from the table, different feature dimensions are given different weights according to their importance, for example, the alpha band energy ratio is considered to be the most discriminative, with the highest weight of 0.22; while the cortisol concentration change rate also contains useful information, but its weight is relatively low, 0.06. The weighting process changes the relative importance of the original features, preparing for the subsequent feature projection and fusion. All these calculations and processes are automatically completed by the system, ensuring the objectivity and consistency of the evaluation process.

[0052] Example 5: Referring to Figure 5 The processing device continuously acquires the iterative stimulation period physiological signals and the post-intervention resting state physiological signals for multiple rounds of electroacupuncture intervention. This process is built on the basis of continuous intervention, and each intervention session follows a standardized operational procedure. The electroacupuncture intervention device automatically records all parameter settings for each intervention, including waveform, frequency, intensity, and duration, when performing each stimulation operation. The first signal acquisition device synchronously acquires the iterative stimulation period physiological signals during each stimulation operation, which include but are not limited to electroencephalogram, electrocardiogram, and electromyogram, using uniform sampling rate and filtering settings to ensure data consistency. After each intervention session, a specified resting observation period is entered, during which the second signal acquisition device begins to work and acquire the post-intervention resting state physiological signals, which may include heart rate variability data, galvanic skin response, and certain biochemical indicators obtained through non-invasive sensors. All acquired signal data are time-stamped and stored in association with the parameter settings for this intervention.

[0053] The processing device pre-processes and quality controls the acquired iterative signal data. The pre-processing steps include signal denoising, baseline correction, and artifact removal to ensure data usability. For data segments that do not meet quality standards, the system will automatically mark and exclude them in subsequent analysis. Through this continuous multi-round data acquisition, the system accumulates a large amount of time-series physiological data of the current subject, which reflects the trend of intervention effect over time, providing a data basis for system adaptive optimization.

[0054] The processing device updates the prediction weights of the stimulation response prediction model and the baseline threshold of the effect evaluation benchmark curve. The update process uses an incremental learning mechanism, which can gradually adjust model parameters using newly acquired data without retraining the entire model. For the stimulation response prediction model, the update focuses on adjusting the prediction weights, so that the model can better adapt to the specific response patterns of individual subjects. The update algorithm calculates the weight adjustment amount by comparing the difference between the model prediction value and the actually calculated effect index, and iteratively optimizes using a small step gradient descent method. This updating method enables the prediction model to gradually adapt to changes in individual physiological characteristics, improving prediction accuracy.

[0055] The update of the effect evaluation benchmark curve mainly involves adjusting its benchmark threshold. The benchmark threshold defines the normal fluctuation range of the effect parameter at each intervention stage. The system recalculates the statistical distribution characteristics of the effect parameter, especially its central tendency and dispersion at each time point, by analyzing the newly accumulated intervention data. Based on the new statistical results, the system adjusts the threshold of the benchmark curve accordingly, so that it can reflect the latest intervention effect characteristics of the current subjects. The update process adopts a weighted average strategy, with new data having a higher weight and historical data weight gradually decaying, so as to timely reflect the latest changes and maintain the stability of the benchmark.

[0056] The generation logic of the intervention parameter adjustment instruction is optimized by the processing device according to the updated prediction weight and benchmark threshold. The optimization process involves adjusting the parameter and threshold settings in the decision rule. The system evaluates the effectiveness of the current generation logic by analyzing the execution effect of historical adjustment instructions, i.e. the change of intervention effect after the execution of the instructions. For the instruction generation rules that lead to positive effect changes, their influence will be enhanced; for the rules with poor effect, they will be weakened or modified. The optimization process also considers the stage characteristics of the intervention process, and adopts different optimization strategies in different intervention stages. For example, in the early stage of intervention, optimization may pay more attention to rapid response and parameter exploration; while in the later stage of intervention, more attention is paid to fine adjustment and stability maintenance.

[0057] The whole update and optimization process forms a closed-loop system, which can continuously adjust and improve itself according to the feedback data of multiple rounds of intervention. The system maintains an update log to record the content, time and basis of each update, so as to track the self-evolution process of the system. This iterative optimization mechanism enables the electroacupuncture intervention effect evaluation system to gradually adapt to individual differences, improve the level of individualization of intervention and the accuracy of effect evaluation. With the increase of intervention rounds, the evaluation and adjustment ability of the system is continuously improved, forming a more precise individualized intervention plan. All the update and optimization operations are automatically completed by the system without human intervention, ensuring the efficient operation and objectivity of the system.

[0058] It should be noted that in this paper, relationship terms such as first and second are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between the entities or operations. Moreover, the terms "include", "contain" or any other variant thereof are intended to cover non-exclusive inclusion, so that the process, method, article or equipment including a series of elements not only includes those elements, but also includes other elements not explicitly listed or inherent to such process, method, article or equipment.

[0059] While embodiments of the application have been shown and described, it is to be understood that the embodiments described are merely exemplary of the principles and application of the present application. Numerous modifications and adaptions can be effected without departing from the spirit and scope of the present application, which is not limited to the exact construction and arrangement described. It is intended, therefore, to cover all modifications and adaptions that fall within the scope of the claims and their equivalents.

Claims

1. A system for evaluating the effect of electroacupuncture intervention based on multimodal physiological data fusion, characterized in that, Includes a central processing unit, which is used to perform: Electroacupuncture stimulation was performed on the subject using an electroacupuncture intervention device; Physiological signals during the stimulation phase of the electroacupuncture stimulation operation are acquired through the first signal acquisition device. The processing device executes a physiological feature extraction algorithm during the stimulation period, and calculates the first effect index during the electroacupuncture intervention based on the physiological signals during the stimulation period. The resting physiological signals after electroacupuncture intervention are acquired using a second signal acquisition device. The processing device executes a post-intervention feature fusion algorithm to calculate the second post-intervention effect index based on the resting-state physiological signals after the electroacupuncture intervention. The processing device performs multimodal data fusion of the first effect index and the second effect index to generate comprehensive intervention evaluation parameters.

2. The electroacupuncture intervention effect evaluation system based on multimodal physiological data fusion according to claim 1, characterized in that, Also includes: The processing device constructs a stimulus response prediction model based on a preset historical intervention database, which includes the physiological signals of the reference stimulus period and the corresponding reference intervention effect indicators of the reference subjects. The processing device executes the stimulus response prediction model to predict real-time intervention efficacy parameters based on the physiological signals during the stimulation period.

3. The electroacupuncture intervention effect evaluation system based on multimodal physiological data fusion according to claim 2, characterized in that, The algorithm for extracting physiological features during the stimulation period, executed via a processing device, includes: Identify the electroacupuncture response waveform characteristics in the physiological signals during the stimulation period; Extract the time-domain distribution characteristics and frequency-domain energy characteristics of the electroacupuncture response waveform; The first effect index is generated based on the time-domain distribution characteristics and frequency-domain energy characteristics.

4. The electroacupuncture intervention effect evaluation system based on multimodal physiological data fusion according to claim 3, characterized in that, The post-intervention feature fusion algorithm executed by the processing device includes: Separate the autonomic nervous system regulatory features and body fluid metabolism features from the resting-state physiological signals after electroacupuncture intervention; Calculate the variability index of the autonomic nervous regulatory features and the concentration change index of the body fluid metabolic features; The second effect index is generated by combining the variability index and the concentration change index.

5. The electroacupuncture intervention effect evaluation system based on multimodal physiological data fusion according to claim 4, characterized in that, The step of performing multimodal data fusion of the first effect index and the second effect index through the processing device includes: Construct the stimulus response feature vector corresponding to the first effect index and the steady-state regulation feature vector corresponding to the second effect index; The feature layer fusion algorithm is executed to map the stimulus response feature vector and the steady-state regulation feature vector to the joint feature space; The comprehensive intervention evaluation parameters are generated based on the fused feature values ​​of the joint feature space.

6. The electroacupuncture intervention effect evaluation system based on multimodal physiological data fusion according to claim 5, characterized in that, Also includes: The processing device establishes an effect evaluation baseline curve based on the historical electroacupuncture intervention data of the reference subjects; The processing device dynamically compares the comprehensive intervention evaluation parameters with the effect evaluation benchmark curve; The processing device outputs intervention parameter adjustment instructions to the electroacupuncture intervention device based on the dynamic comparison results.

7. The electroacupuncture intervention effect evaluation system based on multimodal physiological data fusion according to claim 6, characterized in that, The construction of the stimulus-response prediction model includes: Multiple sets of physiological signal characteristics and associated intervention effect indicators during the reference stimulation period were extracted from the historical intervention database. The feature selection algorithm is used to screen core stimulus-response features. The stimulus response prediction model is constructed based on the mapping relationship between the core stimulus response characteristics and the associated intervention effect indicators.

8. The electroacupuncture intervention effect evaluation system based on multimodal physiological data fusion according to claim 7, characterized in that, The execution feature layer fusion algorithm includes: The stimulus-response feature vector and the steady-state regulation feature vector are weighted by an importance-based nonlinear weighting mechanism. The weighted feature vectors are reduced to the joint feature space using a feature projection algorithm.

9. The electroacupuncture intervention effect evaluation system based on multimodal physiological data fusion according to claim 8, characterized in that, The step of dynamically comparing the comprehensive intervention evaluation parameters with the effect evaluation benchmark curve through the processing device includes: Identify the stage-sensitive intervals of the baseline curve for evaluating the effect; Calculate the deviation of the comprehensive intervention evaluation parameters from the sensitive interval of the stage; Adjust the value of the intervention parameter based on the deviation quantification.

10. The electroacupuncture intervention effect evaluation system based on multimodal physiological data fusion according to claim 9, characterized in that, Also includes: The processing device continuously acquires physiological signals during the iterative stimulation period of multiple rounds of electroacupuncture intervention and resting physiological signals after iterative intervention. The processing device updates the prediction weights of the stimulus response prediction model and the baseline threshold of the effect evaluation baseline curve. The processing device optimizes the generation logic of the intervention parameter adjustment instruction based on the updated prediction weights and baseline thresholds.

Images