Non-invasive information collection method and system for acupoint bioelectric signals

Abstract

This invention relates to the field of acupoint signal acquisition technology, and discloses a non-invasive method and system for acquiring bioelectrical signals from acupoints. The method first acquires initial bioelectrical signal data from acupoint regions of the human body, then optimizes the acquisition frequency and period based on historical signal change trends, and preprocesses the initial data to obtain optimized acquisition data. Subsequently, the optimized acquisition data is divided into multiple acupoint micro-regions, and signal analysis weights are assigned according to the physiological characteristics of each micro-region to generate an acupoint partition feature dataset. This method dynamically optimizes acquisition parameters, making data acquisition more adaptable to fluctuations in human physiological states, improving the comprehensiveness and efficiency of the data; targeted preprocessing enhances data reliability; and scientific partitioning and weight allocation allow the feature dataset to more accurately reflect the physiological characteristics of each micro-region, facilitating a more detailed interpretation of human physiological state information, and is suitable for scenarios such as health monitoring and physiological state tracking.

Description

Technical Field

[0001] This invention relates to the field of acupoint signal acquisition technology, specifically to a non-invasive method and system for acquiring bioelectrical signals from acupoints. Background Technology

[0002] In research combining traditional Chinese medicine theory with modern medicine, the collection of bioelectrical signals from acupoints is an important means of exploring the physiological state of the human body. Traditional methods for collecting bioelectrical signals from acupoints mostly rely on invasive procedures, such as inserting electrodes into the skin or using invasive devices to fix the collection point. These methods may not only cause discomfort to the subject but also pose a risk of infection, limiting their application in scenarios such as routine health monitoring and long-term physiological state tracking.

[0003] Even with the development of some non-invasive data acquisition methods, numerous limitations remain. Acquisition frequency and cycles are often fixed, unable to be adjusted according to dynamic changes in the human body's physiological state. The bioelectrical signals of acupoints are significantly volatile, influenced by factors such as emotions, diet, and exercise. Fixed acquisition parameters can easily lead to data omissions during periods of drastic signal changes or redundant data during periods of stable signals, reducing acquisition efficiency and data quality. Existing methods often perform holistic analysis of the acquired signals, neglecting the physiological differences between different micro-regions within the acupoint area. Different acupoint micro-regions play different roles in their connection with the body's organs and meridians, and their bioelectrical signal variation patterns also differ. Holistic analysis struggles to accurately capture this specific information, limiting the accuracy of subsequent assessments of the human body's physiological state.

[0004] Traditional preprocessing methods for noise filtering and signal enhancement are relatively simplistic, often employing uniform filtering algorithms that fail to consider the frequency characteristics of bioelectrical signals in different acupoint regions. This can easily lead to the loss of useful signals or incomplete noise removal. Furthermore, the data partitioning and weight allocation lack scientific basis, relying heavily on empirical judgment. This results in feature datasets that fail to accurately reflect the physiological state of acupoints, impacting the reliability of subsequent analysis models. These issues collectively restrict the widespread application of non-invasive acupoint bioelectrical signal acquisition technology in clinical diagnosis, health management, and other fields. Summary of the Invention

[0005] The purpose of this invention is to provide a non-invasive method and system for acquiring bioelectric signals from acupoints, in order to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides a non-invasive method for acquiring bioelectrical signals from acupoints, the method comprising:

[0007] Initial bioelectrical signal data of acupoints in the human body are collected, and the collection frequency and period are optimized based on historical signal change trends. The initial bioelectrical signal data is preprocessed to obtain optimized collection data.

[0008] The optimized collected data is divided into multiple acupoint micro-regions, and signal analysis weights are assigned according to the physiological characteristics of each acupoint micro-region to generate an acupoint partition feature dataset.

[0009] Preferably, the preprocessing of the initial bioelectrical signal data to obtain optimized acquisition data includes:

[0010] The initial bioelectric signal data was segmented using an adaptive window function, and the dispersion coefficient of the signal amplitude within each window was calculated.

[0011] The filtering threshold is dynamically adjusted based on the discrete coefficients to filter out interference signal segments that do not conform to the physiological characteristic range;

[0012] The recombined and filtered signal segments form optimized acquisition data with continuous timing.

[0013] Preferably, the generated acupoint partition feature dataset includes:

[0014] Multi-scale feature extraction was performed on the optimized data collected from each acupoint micro-region to obtain time-domain and frequency-domain feature sets.

[0015] Based on the preset acupoint physiological model, feature fusion is performed on the time-domain feature set and the frequency-domain feature set;

[0016] The dimensionality of the fused features is reduced based on the feature correlation coefficient to generate acupoint partition feature dataset.

[0017] Preferably, the method further includes:

[0018] Construct a dynamic feature analysis window and perform sliding window analysis on the acupoint partition feature dataset;

[0019] Calculate the variation index of key features within adjacent analysis windows, and mark the feature mutation point when the variation index exceeds a preset threshold;

[0020] Adjust the step length of the analysis window based on the distribution density of the characteristic mutation points.

[0021] Preferably, the method further includes:

[0022] A feature optimization rule base based on physiological constraints is established, and the parameters in the feature optimization rule base are iteratively adjusted using a heuristic search strategy.

[0023] The feature parsing accuracy is evaluated in each iteration, and the optimization is terminated when the accuracy improvement rate is lower than a set value for three consecutive iterations.

[0024] Output the final feature optimization rules and the corresponding parameter set.

[0025] Preferably, the method further includes:

[0026] The acupoint partition feature dataset is input into the optimized feature optimization rule base to generate bioelectric signal analysis results.

[0027] Compare the differences between the bioelectric signal analysis results and the standard acupoint signal template;

[0028] The signal anomaly levels are classified according to the degree of difference, and level labels are generated.

[0029] Preferably, the method further includes:

[0030] Set signal strength constraint boundaries and rate of change constraint boundaries, and use the boundary projection method to correct the bioelectric signal analysis results;

[0031] The signal segment that exceeds the constraint boundary is resampled in the time domain so that the corrected signal meets the preset physiological change constraint conditions.

[0032] Preferably, the method further includes:

[0033] Feature segments are extracted from the corrected signal, a feature segment similarity matrix is ​​constructed, and the association confidence of adjacent segments in the feature segment similarity matrix is ​​calculated; when the association confidence is lower than a set threshold, it is marked as an isolated feature segment.

[0034] Preferably, the method further includes:

[0035] Count the number and distribution location of the isolated feature fragments;

[0036] Based on the combined information of the signal anomaly level identifier and the distribution information of isolated feature fragments, a comprehensive index of acupoint status is calculated;

[0037] A status determination conclusion is generated based on the comparison between the comprehensive index of acupoint status and the health threshold.

[0038] Preferably, the present invention also includes a non-invasive acupoint bioelectric signal information acquisition system, the system including a memory, a processor, and a computer program stored in the memory and running on the processor, wherein when the processor executes the computer program, it implements the steps of the non-invasive acupoint bioelectric signal information acquisition method described above.

[0039] Compared with existing technologies, the beneficial effects of this invention are: by collecting initial bioelectrical signal data from acupoint areas of the human body and optimizing the acquisition frequency and period based on historical signal change trends, the acquisition process can dynamically adapt to fluctuations in the human physiological state. When historical signals show that the signal changes drastically within a certain period, the acquisition frequency can be automatically increased and the acquisition period shortened to capture more detailed information; while in the stage where the signal tends to be stable, the acquisition frequency is reduced and the acquisition period is extended to reduce the generation of redundant data, making the acquired data both comprehensive and efficient, avoiding the information omission or data redundancy problems that may occur under traditional fixed parameter acquisition methods.

[0040] Preprocessing the initial bioelectrical signal data yields optimized acquisition data, enabling more targeted processing methods based on the characteristics of bioelectrical signals in different acupoint regions. By distinguishing the frequency characteristics and noise sources of the signals, appropriate filtering and enhancement techniques are employed to effectively retain useful signals while removing noise interference. This allows the processed data to more accurately reflect the physiological state of the acupoints, providing a more reliable foundation for subsequent analysis.

[0041] The optimized collected data was divided into multiple acupoint micro-regions, and signal analysis weights were assigned according to the physiological characteristics of each micro-region, fully considering the functional differences among different micro-regions within the acupoint area. Different acupoint micro-regions have specific associations with different organs and meridians in the human body, and the physiological information reflected by changes in their bioelectrical signals are also different. By rationally dividing the micro-regions and assigning corresponding analysis weights, the signals of those micro-regions more closely related to specific physiological functions receive more attention during the analysis process. The generated acupoint regional feature dataset can more accurately reflect the physiological characteristics of each micro-region, which helps to interpret the human physiological state information contained in the bioelectrical signals of acupoints in a more detailed way.

[0042] This approach allows the collected bioelectrical signal data to better reflect actual physiological conditions. The pre-processed and optimized data achieves higher quality, and the partitioned feature datasets are more targeted, providing more accurate information for subsequent assessments of human physiological status and disease early warning. Simultaneously, the non-invasive acquisition method avoids the discomfort and potential risks associated with invasive procedures, increasing patient acceptance and applicability to a wider range of scenarios, including daily health monitoring and long-term physiological state tracking. By dynamically adjusting acquisition parameters, precise preprocessing, and scientifically allocating partition weights, the overall adaptability, accuracy, and reliability of acupoint bioelectrical signal acquisition are improved, contributing to the expansion of non-invasive acupoint bioelectrical signal technology across multiple fields. Attached Figure Description

[0043] Figure 1 This is a schematic diagram illustrating the working principle of the non-invasive acupoint bioelectric signal information acquisition method described in this invention.

[0044] Figure 2 A flowchart for preprocessing to obtain optimized collected data.

[0045] Figure 3 This is a flowchart for the dynamic feature analysis window and step adjustment.

[0046] Figure 4 This is a flowchart for analyzing bioelectrical signals and classifying abnormal levels.

[0047] Figure 5 This is a flowchart for calculating and determining the comprehensive index of acupoint status. Detailed Implementation

[0048] 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.

[0049] Please see Figure 1 This invention provides a non-invasive method for acquiring bioelectrical signals from acupoints, the system comprising:

[0050] By optimizing the signal acquisition and analysis process, accurate extraction and feature analysis of bioelectrical signals from acupoint regions in the human body are achieved. The specific implementation process includes: acquiring initial bioelectrical signal data from acupoint regions; dynamically adjusting the acquisition frequency and period based on historical signal change trends to ensure the timeliness and completeness of data acquisition; preprocessing the initial bioelectrical signal data to filter out noise interference and reconstructing it into optimized, time-series continuous acquisition data; further dividing the optimized acquisition data into multiple acupoint micro-regions; assigning signal analysis weights based on the physiological characteristics of each micro-region to generate an acupoint regional feature dataset containing both time and frequency domain features; iteratively optimizing the dataset through a dynamic analysis window and feature optimization rule base; finally outputting the bioelectrical signal analysis results; comparing them with standard acupoint signal templates; and generating a comprehensive acupoint status index and judgment conclusion.

[0051] Example 1: See Figure 2The preprocessing stage of the initial bioelectrical signal data employs an adaptive window function segmentation technique. The window function length is dynamically variable, with an initial default value of 200 milliseconds. Window length adjustments are automatically triggered based on signal stability indicators. When the standard deviation of the signal amplitude within the window is below a set threshold for three consecutive periods, the window length expands to 300 milliseconds; if an instantaneous amplitude mutation is detected exceeding twice the mean of adjacent signals, the window length shrinks to 100 milliseconds. The coefficient of variation is calculated based on the statistical distribution characteristics of the signal amplitude within the window, specifically quantifying the degree of signal fluctuation through the ratio of the standard deviation to the arithmetic mean. When the coefficient of variation exceeds a preset threshold range, the window is determined to contain a non-physiological interference signal segment.

[0052] The dynamic adjustment of the filtering threshold employs a nonlinear mapping mechanism, mapping the discrete coefficient values ​​to the cutoff frequency range of the bandpass filter. The mapping relationship follows a piecewise function rule: a low discrete coefficient range corresponds to a narrower filtering bandwidth, preserving the main physiological frequency band; a high discrete coefficient range corresponds to a wider filtering bandwidth, suppressing high-frequency noise components. A zero-phase digital filter is used in the filtering process to avoid introducing time-shift distortion. The signal segments after interference removal are sequentially joined using a linear interpolation method, with interpolation points calculated based on the amplitude gradient of adjacent effective signal segments. The reconstructed optimized acquisition data must meet two conditions: the timing continuity error is less than 5% of the sampling interval, and the effective signal segment loss rate does not exceed 3% of the total duration.

[0053] Acupoint micro-regions were divided using a gridded positioning method, establishing a polar coordinate system with the acupoint center point as the reference. Each micro-region covered a 5 mm radius annular area, with a 2 mm overlap zone between adjacent micro-regions to eliminate boundary effects. Physiological feature weights were determined based on the anatomical location of the micro-regions: micro-regions located in areas with dense nerve bundles had a weight increased to 1.2, while those in areas with sparse blood vessel distribution had a weight decreased to 0.8. These weighting coefficients were used in the subsequent weighted calculation process for feature analysis.

[0054] Multi-scale feature extraction includes dual-channel analysis in the time and frequency domains. Time-domain feature extraction employs a sliding window statistical method with a fixed window length of 1 second and a step interval of 200 milliseconds. Within each window, four statistics are calculated: the arithmetic mean of the signal amplitude reflects the overall intensity level; variance characterizes the degree of signal fluctuation; skewness coefficient describes the asymmetry of the amplitude distribution; and kurtosis coefficient quantifies the sharpness of the distribution pattern. Frequency-domain feature extraction uses discrete wavelet transform, with a decomposition level of 5 layers, and selects the db4 wavelet basis function. Feature parameters include: the percentage of energy in each frequency band relative to the total energy; the energy entropy value of a specific frequency band (0.5-50Hz); the amplitude ratio of the fundamental wave to the second harmonic; and the frequency coordinates of the spectral centroid.

[0055] The feature fusion stage incorporates a database of acupoint response characteristic parameters, containing typical conduction characteristic data for different acupoints. The fusion algorithm employs a combination of feature concatenation and weighted combination: time-domain statistics and frequency-domain parameters are first normalized to the same dimension, then linearly combined according to a preset ratio. The combination weights are dynamically adjusted based on the acupoint type; for example, time-domain features are given higher weight to acupoints associated with motor nerves, while frequency-domain features are emphasized for acupoints associated with endocrine disorders. The dimension of the fused feature vector is 1.5 times that of the original features.

[0056] Principal component analysis (PCA) is used for feature dimensionality reduction. The covariance matrix of the eigenvectors is calculated, and their eigenvalues ​​and eigenvectors are solved. The eigenvalues ​​are sorted from largest to smallest, and the calculation stops when the cumulative contribution rate reaches 95%. The corresponding number of principal components are selected as the final features. The dimensionality-reduced feature set is arranged in a time series, forming a structured acupoint region feature dataset. This dataset contains compressed feature vectors for each sampling time and each acupoint micro-region, and the data is stored using a dual index structure of timestamps and spatial coordinates.

[0057] In practice, the signal preprocessing module monitors key indicators in real time. When the coefficient of variation exceeds a warning threshold for 10 consecutive analysis windows, an automatic parameter adjustment mechanism is triggered, increasing the sampling frequency to 120% of the original value. A dimensionality monitoring mechanism is implemented during feature reduction; when the feature dimension after principal component analysis exceeds a preset upper limit, the contribution rate threshold is automatically increased to 97% for secondary dimensionality reduction. An anomaly handling mechanism is established throughout the implementation process, implementing data caching and breakpoint continuation functions to handle signal interruptions, feature calculation anomalies, and other situations.

[0058] The feature dataset is stored using a hierarchical structure. Raw data is stored in the bottom layer, preprocessed and optimized data is stored in the middle layer, and partitioned feature datasets are stored in the application layer. Each layer of data is appended with a timestamp, device number, and subject identification information, supporting multi-dimensional data retrieval and retrospective analysis. Data transmission employs differential coding compression technology, compressing the data volume to 40% of its original size while maintaining feature accuracy.

[0059] Quality control during implementation comprises three stages: In the signal acquisition stage, an impedance detection module is installed, automatically triggering an alarm when the electrode contact impedance exceeds 50 kΩ; in the feature extraction stage, feature stability is checked, and abnormal feature points with fluctuations exceeding three standard deviations are recalculated; in the data storage stage, cyclic redundancy check codes are used, triggering a reprocessing procedure for the corresponding data segment upon verification failure. The entire implementation system has self-diagnostic capabilities, periodically generating operational status logs for each module for maintenance and analysis.

[0060] Example 2: See Figure 3The dynamic feature analysis window is constructed using a sliding window mechanism, with an initial window length of 5 seconds, covering a time range corresponding to the typical physiological cycle of bioelectrical signals. The window moves along the time axis with a fixed step size, initially set to 1 second. Each analysis window performs key feature extraction operations, extracting signal amplitude statistics, frequency domain energy distribution parameters, and time-series variation patterns. The variance index is calculated based on the feature differences between adjacent analysis windows, specifically selecting three core indicators: the coefficient of variation of signal amplitude within the window, the offset of the dominant frequency component, and the Euclidean distance between feature vectors. The difference value is obtained by the arithmetic difference between the current window feature and the previous window feature, and the calculation result is compared with a preset threshold. The threshold setting adopts a dynamic benchmark method, with the benchmark value being the average level of the most recent ten window features, and the actual threshold being 1.5 times the standard deviation of the benchmark value. When any feature difference value exceeds the corresponding threshold, the system marks the feature mutation point at that time point.

[0061] The distribution density of mutation points is statistically analyzed using a kernel density estimation algorithm. With the time axis as the x-axis and mutation events as discrete points, the point density within a unit time window is calculated using a Gaussian kernel function. The density threshold is set to twice the average historical density. When the density values ​​of three consecutive analysis windows exceed the threshold, a window parameter adjustment mechanism is triggered. The step size is reduced from the initial value of 1 second to 0.5 seconds, while the window length remains unchanged at 5 seconds. This adjusted window configuration improves the spatiotemporal resolution of feature parsing, enhancing mutation point localization accuracy from the second level to the sub-second level. The window parameter recovery mechanism is set so that when the mutation point density of ten consecutive windows falls below 80% of the threshold, the step size automatically returns to its initial value.

[0062] The feature optimization rule base is constructed using a set of physiological constraints. These constraints are categorized into three types: signal amplitude constraints are set to 0.7 to 1.3 times the resting state reference value; frequency bandwidth constraints are limited to the 0.1-100Hz range, with a focus on monitoring the 0.5-50Hz physiological frequency band; and rate of change constraints require that the signal amplitude change between adjacent sampling points not exceed 200%. The rule base parameters include a filter coefficient matrix, feature weight vectors, and mutation detection sensitivity parameters. Parameter optimization is implemented using a genetic algorithm, with a population size of 50 individuals. Each individual has a 15-bit code containing binary codes for three parameter groups.

[0063] The heuristic search process employs iterative optimization. Each iteration comprises three steps: first, generation of offspring individuals through crossover with a crossover probability of 0.8; second, mutation with a mutation probability of 0.05; and finally, fitness evaluation. The fitness function is defined as the feature parsing accuracy index, which is a weighted average of the mean squared error and correlation coefficient terms in a 6:4 ratio. After each iteration, the top 30% of individuals by fitness are retained for the next generation. The iteration termination condition is set when the accuracy improvement rate for three consecutive iterations is less than 1%, at which point the parameters of the individual with the highest fitness in the current population are output. The optimized rule base parameter set includes: the bandpass filter cutoff frequency adjusted to the 8-45Hz range; the time-domain feature weight increased to 0.6; the frequency-domain feature weight decreased to 0.4; and the mutation detection sensitivity coefficient set to 0.75.

[0064] The rule base operates using an event-driven model. When the acupoint partition feature dataset is updated, the rule matching engine is triggered. The engine matches the input features with the rule conditions; a successful match triggers the corresponding parameter group to execute feature parsing. After each rule execution, a parsing log is generated, recording the number of rule triggers, the effect of parameter application, and any abnormal events. The log data is used for continuous updates to the rule base. If a single rule fails to improve parsing accuracy after ten consecutive triggers, the rule enters a pending revision state.

[0065] The system employs a rule version management mechanism. A new rule version is generated after each parameter optimization, and historical versions are retained for 30 days. If a new rule version experiences a decrease in accuracy for three consecutive days, it automatically rolls back to the previous stable version. The version switchover process uses a gradual migration, with the old and new rules running in parallel for 24 hours before the switch is completed.

[0066] The monitoring module tracks key indicators in real time during implementation. Window sliding status monitoring includes parameters such as window position offset and data processing latency; rule base operation monitoring records rule trigger frequency, parameter application coverage, and matching failure rate. When more than 5 sampling points of asynchronous window sliding are detected, a timestamp calibration procedure is triggered; when the rule matching failure rate continuously exceeds 10%, a rule base integrity check process is initiated.

[0067] The data analysis process establishes quality control nodes. Before a window slides, a data integrity check is performed; windows with more than 5% missing data are marked as invalid. Before rules are applied, parameter range checks are performed, and parameter values ​​exceeding reasonable ranges are automatically corrected. Before outputting parsed results, a logical consistency check is performed, and results violating physiological constraints are re-parsed.

[0068] The system maintenance mechanism includes automatic diagnostic functions. Daily scheduled module self-checks are performed, covering the timing accuracy of the window sliding mechanism, the response speed of the rule matching engine, and the data integrity of the parameter storage unit. The self-check results generate a health report; a detailed diagnostic process is triggered when the health level falls below 90%. Maintenance logs are archived weekly, supporting fault retrospective analysis.

[0069] The data processing pipeline employs a buffered design. A dual buffer is set up between the window sliding module and the rule base module: the front buffer stores data to be processed, and the back buffer stores processed results. The buffer capacity is configured to 150% of the maximum processing latency, and a flow control mechanism is activated when the buffer utilization exceeds 80%. Data flow uses timestamp synchronization technology to ensure strict alignment of time-series data.

[0070] The exception handling mechanism covers three scenarios: for data loss during window sliding, forward padding is used to complete the data; when rule matching fails, a default rule set is activated, with the default rule set based on historically optimal parameters; and after parameter optimization iteration times out, the current optimal solution is forcibly output. All exception events are recorded with detailed contextual information, including the time of occurrence, scope of impact, and handling measures.

[0071] Performance tuning employs a tiered strategy. Basic-level optimizations include parallelizing the window sliding algorithm and establishing a rule matching index; advanced-level optimizations include caching hot rules and pre-computing high-frequency features. Tuning operations are performed during system idle periods to avoid impacting real-time data processing. Tuning effectiveness is evaluated using metrics such as latency reduction rate and resource utilization.

[0072] The system interface design supports external expansion. It provides a window parameter configuration interface, allowing adjustment of the initial window length and step value; an open rule import interface, supporting the addition of custom constraint rules; and an optimization process monitoring interface, outputting iteration progress and intermediate results. Interface access employs hierarchical permission control, and modifications to critical parameters require dual authentication.

[0073] The implementation environment configuration requirements include hardware resource allocation. The window sliding module occupies one dedicated CPU core, the rule base operation is allocated two CPU cores, and the data buffer requires at least 1GB of memory. Task scheduling adopts a real-time priority strategy, with the window sliding task having the highest priority, followed by rule matching, and parameter optimization tasks running in the background. The resource monitoring module tracks CPU load and memory usage in real time, and starts a load balancer when thresholds are exceeded.

[0074] Example 3: See Figure 4The generation of bioelectric signal analysis results is performed through an optimized feature optimization rule base. The rule base receives acupoint region feature datasets as input and uses a frame-sliding processing mechanism to analyze bioelectric features frame by frame. Each frame's data processing time is 200 milliseconds, synchronized with the sampling interval of the feature dataset. The analysis process includes three core stages: feature matching, parameter calculation, and result synthesis. In the feature matching stage, the similarity between the input feature vector and the feature templates in the rule base is calculated, and the template with the highest similarity is selected as the benchmark. In the parameter calculation stage, the transformation matrix corresponding to the template is applied to map the feature vector into a set of bioelectric signal parameters, including instantaneous amplitude, dominant frequency components, harmonic distribution ratio, and temporal gradient. In the result synthesis stage, the analysis data of the current frame and the previous three frames are integrated, and the final analysis result is output through a weighted smoothing algorithm, where the current frame accounts for 70% of the weight, and the remaining weights of historical frames are distributed according to time decay.

[0075] The comparison between the analytical results and the standard acupoint signal template employs a dynamic time warping algorithm. This algorithm addresses the nonlinear deformation of the signal waveform along the time axis by constructing a cumulative distance matrix to find the optimal warping path. The warping path cost function is defined as follows:

[0076]

[0077] Where: D(i,j) represents the cumulative distance between the i-th point of the standard template and the j-th point of the analytical signal, d(s i ,t j Let be the local distance metric between the two points (using Euclidean distance here), and let i and j be the time indices of the template and the analyzed signal, respectively. After the optimal path search is completed, the average distance on the regularized path is calculated as the original dissimilarity Δ.

[0078] The difference grading introduces a time correction factor to eliminate the influence of instantaneous interference. The correction formula is:

[0079]

[0080] Where: L is the final difference level, Δ is the original difference degree, κ is the signal stability coefficient (range 0.8-1.2, calculated based on signal continuity and stationarity), T is the grading threshold base (default value 0.15), and ⌈⋅⌋ indicates rounding to the nearest integer. The grading results are divided into three levels: L=1 (Δ≤10%) is marked in dark green, indicating that the signal conforms to the normal physiological fluctuation range; L=2 (10%<Δ≤30%) is marked in amber, indicating the presence of a potential abnormal pattern; L=3 (Δ>30%) is marked in scarlet, indicating a significant deviation from the physiological benchmark.

[0081] The anomaly marking system synchronously records spatiotemporal characteristics. The time dimension records the anomaly's start time stamp, duration, and trend; the spatial dimension associates the acupoint micro-region coordinates corresponding to the anomaly signal. For anomalies with L≥2, an automatic feature backtracking mechanism is triggered: feature data from the 10 seconds prior to the anomaly's occurrence is extracted, and the feature evolution patterns during the anomaly formation process are analyzed. Evolution patterns are categorized into three types: gradual (characteristics change continuously and slowly), abrupt (characteristics change abruptly), and oscillating (characteristics fluctuate periodically).

[0082] The data storage adopts a hierarchical structure. The basic layer stores the original analysis results, including timestamps, amplitude sequences, and frequency parameters; the analysis layer records intermediate data for difference calculation, including normalized path coordinates and local distance matrices; the application layer stores the hierarchical results and labeling information, using binary encoding to compress storage space. Anomalies are indexed separately, including event ID, start time, duration, highest level, number of involved micro-regions, and evolutionary pattern classification.

[0083] The signal quality monitoring module operates in parallel. During the parsing process, three quality indicators are calculated in real time: signal-to-noise ratio (SNR) (based on high-frequency noise energy estimation), baseline stability (calculated using low-pass filter residuals), and feature consistency (distance between feature vectors in consecutive frames). When any indicator exceeds a warning threshold, a quality label is added to the parsing results. The quality label is divided into four levels: Excellent (all indicators are normal), Good (slight anomaly in one indicator), Average (severe anomaly in one indicator or slight anomaly in two indicators), and Poor (severe anomalies in multiple indicators). Results with quality labels are processed differently in subsequent analyses; data with a quality rating of "Poor" automatically triggers a re-acquisition process.

[0084] The visualization engine generates multi-dimensional analysis views. The time-series view displays three-axis synchronization curves of signal amplitude, frequency, and difference levels; the spectrum view presents the frequency domain energy distribution during abnormal periods; and the topology view maps the spatial distribution heatmap of anomalies in acupoint micro-regions. The views support dynamic focusing operations, automatically displaying precursor characteristic change curves when an abnormal event is selected.

[0085] System configuration parameters can be dynamically adjusted. The difference grading threshold T can be manually set within the range of 0.1-0.3; the alert threshold for quality monitoring is automatically calibrated based on historical data; the color mapping scheme of the visualization view provides multiple preset templates. All configuration changes are logged, including the modification time, operator ID, and parameter version number.

[0086] The exception handling mechanism includes an automatic response strategy. For exceptions with L=3, the system immediately performs three operations: freezes the current parsing results to prevent overwriting; initiates a high-density data acquisition mode (increasing the sampling rate to 200% of the baseline value); and generates a priority transmission command to upload the relevant data to the central server. The processing follows a timeout interruption principle, with a single exception handling session not exceeding 5 seconds. After the timeout, the intermediate state is saved and execution continues in the background.

[0087] Data security employs an encrypted storage strategy. Parsing result files are encrypted using the AES-256 algorithm, with the key stored in segments within a hardware security module. Access control utilizes a three-tiered permission management system: basic permissions allow viewing data with L=1; advanced permissions allow access to data with L≤2; and privileged permissions grant access to the complete parsing results. Permission changes require dual authentication, including biometric recognition and dynamic password verification.

[0088] A redundancy verification mechanism is established in the signal parsing process. The output of the main parsing channel is synchronously sent to the verification module, which operates using a simplified rule base. When the difference between the main and backup results exceeds 5%, the arbitration process is triggered: the original feature data of the disputed period is extracted, and a third parsing is performed by an independent arbitration engine, with the arbitration result as the final output. The arbitration log records the disputed time point, difference parameters, and arbitration conclusion in detail for subsequent rule base optimization.

[0089] Performance optimization employs hotspot caching technology. An LRU cache is established for frequently accessed template data, with a cache capacity of 20% of the total templates. SIMD instructions are used to accelerate vector operations during feature matching. A pre-calculated weight table is used in the result synthesis stage to reduce real-time computation. System resource monitoring shows that, under standard hardware configuration, the single-channel parsing latency is controlled within 50 milliseconds, meeting real-time processing requirements.

[0090] Example 4: The setting of signal intensity constraint boundaries is based on a physiological signal benchmark database, which contains typical signal characteristics of healthy subjects in a resting state. Taking the Zusanli acupoint as an example, the benchmark range of resting state signal amplitude is 120-180 microvolts. Based on this, the lower limit is set at 84 microvolts (70% of the benchmark lower limit), and the upper limit is set at 234 microvolts (130% of the benchmark upper limit). The calculation of the rate of change constraint uses the sliding difference method, with a window width set to 3 sampling points. The maximum allowable change amplitude between adjacent sampling points is dynamically adjusted by the average rate of change of the previous five sampling periods. The implementation process of the boundary projection method includes three steps: signal scanning, boundary violation detection, and projection correction. The signal scanning module traverses the bioelectric signal analysis results in chronological order, detecting whether the amplitude of each sampling point exceeds the constraint boundary. Boundary violation detection adopts a dual verification mechanism: first, it judges whether the instantaneous amplitude exceeds the boundary, and second, it checks whether the rate of change exceeds the standard. Signal segments that violate both constraints simultaneously are marked as serious boundary violation events.

[0091] The projection correction algorithm employs different strategies based on the type of boundary violation. For simple amplitude violations, the signal amplitude is adjusted to the nearest boundary value (upper or lower limit). For cases where the rate of change exceeds the limit, the amplitude sequence is scaled proportionally while maintaining the overall shape of the signal waveform. Time-domain resampling is performed on the corrected signal segment, using an interpolation method based on the Lanczos kernel. The resampling density is determined based on the local frequency characteristics of the original signal, with the sampling point interval in the high-frequency region reduced to 0.5 milliseconds and the low-frequency region maintaining a standard 1-millisecond interval. The resampled signal must pass four checks: waveform continuity (no abrupt changes between adjacent sampling points), frequency component integrity (energy loss in the main frequency band does not exceed 5%), physiological rationality (conforming to acupoint response patterns), and time sequence alignment (seamless connection with preceding and following signal segments).

[0092] Refer to Table 1 for a comparison of signal data before and after correction for a certain channel at the Zusanli acupoint. The table includes five fields: timestamp, original amplitude, corrected amplitude, out-of-bounds type, and resampling flag. The data spans 50 milliseconds, covering a complete out-of-bounds event cycle. As shown in the table, the amplitude exceeded the upper limit at 23.5 milliseconds, with the original signal reaching 256 microvolts, which decreased to 234 microvolts after projection correction. The subsequent three sampling points triggered proportional scaling due to excessive rate of change, controlling the signal gradient within a reasonable range. The resampling flag column shows that the sampling density increased in the high-frequency fluctuation region (30-35 milliseconds range), ensuring the complete preservation of signal details.

[0093] Table 1: Comparison data of signals from a certain channel at Zusanli acupoint before and after correction.

[0094]

[0095] The quality control module performs real-time monitoring during the correction process. Monitoring parameters include the correction amplitude ratio (the ratio of signal energy before and after correction), waveform distortion (calculated using dynamic time warping distance), and constraint compliance rate (the percentage of sampling points that meet the conditions). When the correction amplitude ratio exceeds 15% or the waveform distortion exceeds the threshold, the system automatically saves the original signal segment and issues a quality alarm. The alarm triggers a three-level response mechanism: a primary response records the event log; a secondary response initiates a backup correction algorithm; and a high-level response suspends processing of the current channel and requests manual intervention.

[0096] The signal correction log contains complete process tracking data. Each record includes a time window, channel number, original signal fingerprint (hash values ​​of the first six sampling points), correction parameters, operator identifier (automatically marked as SYSTEM), and verification result. Log files are archived daily, using an incremental backup strategy to retain detailed records for the most recent 30 days. The log analysis tool supports multi-dimensional queries by out-of-bounds type, correction magnitude, and time range, assisting in system parameter optimization.

[0097] Hardware acceleration modules improve processing efficiency. The boundary detection algorithm is deployed on an FPGA chip for parallel computing, enabling boundary judgment of 16 sampling points in a single clock cycle. Projection correction calculations are accelerated by the GPU, utilizing CUDA cores to process multi-channel signals simultaneously. The resource scheduler dynamically allocates computing units, prioritizing the processing capacity of real-time channels during peak signal periods, while background analysis tasks are automatically degraded.

[0098] The anomaly handling mechanism covers three typical scenarios. When a sudden strong interference causes a large-scale out-of-bounds situation, a signal segment replacement strategy is initiated: the abnormal segment is replaced with the most similar normal signal segment within the previous 10 seconds. When equipment drift causes a systematic offset, a baseline recalibration process is triggered to adjust the absolute values ​​of the constraint boundaries. When a software anomaly causes a correction to fail, it is rolled back to the most recent valid correction version to ensure that signal continuity is not disrupted.

[0099] The user interface provides visual monitoring of the correction process. The main view displays the superimposed curves of the original and corrected signals for the current channel, with areas of difference highlighted in red. The auxiliary panels display dynamic curves of constraint boundaries, statistical charts of out-of-bounds events, and quality monitoring indicators. The interface supports manual fine-tuning of correction parameters; all changes require double confirmation and must be logged in the operation audit log.

[0100] The system maintenance module performs regular self-checks. A signal path verification is automatically run daily at midnight to simulate various out-of-range scenarios and verify the reliability of the correction algorithm. A full-channel stress test is performed weekly to simulate high-frequency abnormal signal impacts and assess system stability. Maintenance reports record hardware resource utilization, algorithm execution efficiency, and anomaly statistics, providing a basis for preventative maintenance.

[0101] Data security measures include triple protection. Transmission is encrypted using TLS to prevent eavesdropping; stored data is encrypted using the AES-256 algorithm, with the key rotating every 24 hours; access control is based on the RBAC model, and modification permissions are limited to senior engineers. An audit trail records all data modification operations, including automatic corrections and manual interventions, preserving a complete chain of operations.

[0102] Performance optimization is ongoing. Hotspot analysis identifies high-frequency out-of-bounds patterns, and pre-generates corrective solutions for common scenarios in a cache. Memory management employs object pooling technology to reduce memory allocation overhead during real-time processing. Thread scheduling optimization ensures high-priority tasks receive stable computing resources, avoiding processing latency fluctuations. The signal processing pipeline implements load balancing, maintaining a processing latency of less than 20 milliseconds even when all eight channels are running at full capacity.

[0103] Example 5: See Figure 5Feature fragment extraction is based on the corrected bioelectrical signal, employing an adaptive length sliding window mechanism. The window length is dynamically adjusted according to the local stability of the signal, with an initial value set at 300 milliseconds. When the signal variance within the window is below a threshold for three consecutive cycles, the window expands to 500 milliseconds; when a transient pulse is detected, it shrinks to 150 milliseconds. Fragment boundaries are aligned with signal zero-crossing points to ensure waveform integrity. A 50-millisecond overlap region is set between adjacent fragments to eliminate boundary truncation effects. Each feature fragment is labeled with a start timestamp, the corresponding acupoint microregion number, and a signal quality marker.

[0104] The feature segment similarity matrix is ​​constructed using a dynamic time-warped distance metric. The matrix rows and columns correspond to feature segments arranged in chronological order, and the matrix element values ​​are the morphological similarity scores between two segments. Similarity calculation includes three dimensions: waveform contour matching is assessed through the alignment of discrete point sequences; spectral feature similarity is compared by comparing the energy distribution of major frequency bands; and the difference in temporal statistical characteristics is calculated by determining the relative deviation between the mean and variance. The final similarity score is the geometric mean of the three-dimensional scores. The confidence score of adjacent segments is defined as the weighted average of the similarity scores between the current segment and the previous segment, with higher weights for closer time intervals. The confidence threshold is set at 0.85; segments with a confidence score below this value are marked as weakly connected.

[0105] The identification of isolated feature fragments employs a density clustering algorithm. A two-dimensional feature space is established based on time-space coordinates, and the density value of each fragment within its unit neighborhood is calculated. The density threshold is taken as the median of the overall distribution; fragments with a density below 60% of the threshold are considered isolated points. The identification process involves iterative optimization: after initial labeling, the neighborhood density is recalculated, and boundary fragments are validated a second time. The finally confirmed isolated fragments are recorded with three attributes: temporal isolation (no similar fragments within 200 milliseconds), spatial isolation (no related features in adjacent micro-regions), and morphological anomaly (waveform pattern deviates from historical baselines).

[0106] The isolated fragment statistics module performs multidimensional analysis. Temporal distribution statistics generate a density heatmap of isolated events along the time axis; spatial distribution is mapped to an acupoint topology model to calculate the incidence rate of isolated events in each micro-region; morphological classification is based on waveform patterns into three categories: pulse-type, oscillating-type, and decay-type. Statistical results are stored as a timestamped distribution vector, with dimensions including hourly isolated event counts, spatial clustering coefficients, and morphological type histograms.

[0107] The calculation of the acupoint status comprehensive index integrates two types of inputs: signal anomaly level data provides L1-L3 classification identification, and isolated fragment distribution information contributes spatial anomaly pattern characteristics. The index calculation formula adopts a weighted fusion model, with the anomaly level weight coefficient set to 0.6 and the isolated fragment density weight set to 0.4. The density values ​​are logarithmically transformed to eliminate dimensional differences, and then normalized to the 0-1 interval using the sigmoid function. The index output range is 0-100, and the calculation process is executed every 5 minutes, taking the peak anomaly level and average isolated density within that time period as inputs.

[0108] The health threshold setting employs a dynamic benchmark mechanism. The base threshold is fixed at 85, with adjustments based on historical subject data: if the average index for three consecutive days is above 90, the threshold is increased by 3 points; if the lowest daily value is below 75 for several consecutive days, the threshold is decreased by 2 points. The threshold update cycle is 24 hours, with each adjustment not exceeding 5 points. The judgment conclusion is generated using a three-level classification: index ≥ threshold = "normal"; threshold - 15 ≤ index < threshold = "sub-healthy"; index < threshold - 15 = "abnormal". Each judgment simultaneously outputs a confidence score, calculated based on the completeness and quality level of the input data.

[0109] The conclusion output module integrates multi-dimensional data. The basic report includes the current index value, judgment conclusion, and analysis of major influencing factors; the trend report displays the index change curve and key event markers for the past 24 hours; the detailed report provides playback of the original signals and comparison views of characteristic segments during abnormal periods. Report transmission adopts a hierarchical push strategy: normal status reports are archived and stored hourly; sub-health status triggers real-time alerts to local terminals; abnormal status reports are encrypted and transmitted to the remote monitoring platform.

[0110] The system's self-protection mechanism includes triple protection. Data integrity verification uses cyclic redundancy check codes, with a 16-bit check value appended to each frame's characteristic fragment; timing consistency checks are performed through hardware clock synchronization, with resynchronization triggered if the deviation exceeds 2 milliseconds; storage protection implements dual-buffered alternating writes to prevent data loss due to sudden power outages. The security log records all critical operations, including isolated fragment marking events, exponent calculation time points, and conclusion generation records.

[0111] The performance monitoring system tracks and processes latency in real time. Latency during feature extraction is controlled within 50 milliseconds; the similarity matrix update cycle is 100 milliseconds; and the index calculation task is completed no later than 200 milliseconds after data is ready. Resource allocation uses dynamic priority, automatically increasing the CPU quota for feature analysis tasks when isolated events surge. System health is monitored via a watchdog timer; a soft restart is performed if there is no response for more than 500 milliseconds.

[0112] The anomaly handling process covers typical failure scenarios. When a feature fragment is lost, a forward prediction mechanism is activated to generate a replacement fragment based on historical patterns; when similarity calculation times out, a fast matching mode is enabled, using a simplified algorithm to ensure real-time performance; when an exponent calculation fails, it rolls back to the most recent valid value to avoid interrupting the conclusion. All anomaly events are recorded with detailed context snapshots, including a system state image of the 10 seconds prior to the failure.

[0113] The user configuration interface offers flexible adjustments. Feature fragment length is adjustable from 100 to 800 milliseconds; the isolation judgment density threshold supports settings from 0.4 to 0.8; and the exponential weight coefficient allows for fine-tuning within ±0.1. Configuration changes are tested in a sandbox, with new parameters taking effect after 30 minutes of parallel execution in shadow mode. Version management records every configuration change, supporting one-click rollback to previous versions.

[0114] The data visualization engine generates an interactive analysis interface. The main view displays real-time exponential curves and health threshold lines; the matrix view presents a heatmap of feature segment similarity; a 3D scatter plot maps the distribution of isolated segments in spatiotemporal coordinates; and the comparison panel displays abnormal segments and standard pattern waveforms side-by-side. View operations support time axis zooming, spatial dimension rotation, and feature filtering to meet in-depth analysis needs.

[0115] The maintenance system implements preventative maintenance. Daily automatic memory defragmentation and storage optimization are performed; temporary files are cleaned up and database indexes are rebuilt weekly; and the system clock and hardware timers are calibrated monthly. Maintenance logs generate maintenance reports, including task execution results, resource release amounts, and potential risk warnings. All maintenance operations are performed when the system is idle, prioritizing real-time data processing capabilities.

[0116] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0117] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A non-invasive method for acquiring bioelectrical signals from acupoints, characterized in that, include: Initial bioelectrical signal data of acupoints in the human body are collected, and the collection frequency and period are optimized based on historical signal change trends. The initial bioelectrical signal data is preprocessed to obtain optimized collection data. The optimized collected data is divided into multiple acupoint micro-regions, and signal analysis weights are assigned according to the physiological characteristics of each acupoint micro-region to generate an acupoint partition feature dataset. Also includes: A feature optimization rule base based on physiological constraints is established, and the parameters in the feature optimization rule base are iteratively adjusted using a heuristic search strategy. The feature parsing accuracy is evaluated in each iteration, and the optimization is terminated when the accuracy improvement rate is lower than a set value for three consecutive iterations. Output the final feature optimization rules and the corresponding parameter set; Also includes: The acupoint partition feature dataset is input into the optimized feature optimization rule base to generate bioelectric signal analysis results. Compare the differences between the bioelectric signal analysis results and the standard acupoint signal template; The signal anomaly levels are classified according to the degree of difference, and level labels are generated.

2. The non-invasive acupoint bioelectrical signal information acquisition method according to claim 1, characterized in that, The preprocessing of the initial bioelectrical signal data to obtain optimized acquisition data includes: The initial bioelectric signal data was segmented using an adaptive window function, and the dispersion coefficient of the signal amplitude within each window was calculated. The filtering threshold is dynamically adjusted based on the discrete coefficients to filter out interference signal segments that do not conform to the physiological characteristic range; The recombined and filtered signal segments form optimized acquisition data with continuous timing.

3. The non-invasive acupoint bioelectrical signal information acquisition method according to claim 1, characterized in that, The generated acupoint partition feature dataset includes: Multi-scale feature extraction was performed on the optimized data collected from each acupoint micro-region to obtain time-domain and frequency-domain feature sets. Based on the preset acupoint physiological model, feature fusion is performed on the time-domain feature set and the frequency-domain feature set; The dimensionality of the fused features is reduced based on the feature correlation coefficient to generate acupoint partition feature dataset.

4. The non-invasive acupoint bioelectrical signal information acquisition method according to claim 3, characterized in that, Also includes: Construct a dynamic feature analysis window and perform sliding window analysis on the acupoint partition feature dataset; Calculate the variation index of key features within adjacent analysis windows, and mark the feature mutation point when the variation index exceeds a preset threshold; The step length of the analysis window is adjusted according to the distribution density of the characteristic mutation points.

5. The non-invasive acupoint bioelectrical signal information acquisition method according to claim 1, characterized in that, Also includes: Set signal strength constraint boundaries and rate of change constraint boundaries, and use the boundary projection method to correct the bioelectric signal analysis results; The signal segment that exceeds the constraint boundary is resampled in the time domain so that the corrected signal meets the preset physiological change constraint conditions.

6. The non-invasive acupoint bioelectric signal information acquisition method according to claim 5, characterized in that, Also includes: Feature segments are extracted from the corrected signal. Construct a feature segment similarity matrix and calculate the association confidence of adjacent segments in the feature segment similarity matrix; when the association confidence is lower than a set threshold, it is marked as an isolated feature segment.

7. The non-invasive acupoint bioelectric signal information acquisition method according to claim 6, characterized in that, Also includes: Count the number and distribution location of the isolated feature fragments; Based on the combined information of the signal anomaly level identifier and the distribution information of isolated feature fragments, a comprehensive index of acupoint status is calculated; A status determination conclusion is generated based on the comparison between the comprehensive index of acupoint status and the health threshold.

8. A non-invasive acupoint bioelectric signal information acquisition system, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the non-invasive acupoint bioelectric signal information acquisition method according to any one of claims 1 to 7.

Images