A Two-Dimensional Dynamic Evaluation Method for Meridian Signals Based on Phase Space Reconstruction
By constructing a time series of the five-element functional states and combining it with a two-level characterization method based on the maximum Lyapunov exponent and fractal dimension, the problem of dynamic evaluation of short-time meridian bioelectric signals was solved, realizing a robust evaluation of the meridian-five-element functional system and objective quantification of its health status.
Patent Information
- Application Number
- CN202611151387.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-07-31
- Publication Date
- 2026-11-10
AI Technical Summary
Existing technologies lack a combined analysis method for orbit divergence and phase space geometric complexity of clinical meridian bioelectric signals with short time series and high noise levels, making it difficult to achieve quantitative characterization of the meridian-five elements functional system.
By constructing a time series of five-element functional states, robust preprocessing, sliding window robust reconstruction, and phase space embedding strategies are adopted. Combined with the maximum Lyapunov exponent and fractal dimension for dynamic evaluation, a divergent-geometric dual-level characterization system is formed, which is suitable for short-time clinical signals.
It achieves robust dynamic evaluation of the meridian-five elements functional system, reduces the interference of outliers and noise on the analysis, improves the stability and reliability of the estimation, and provides an objective quantitative evaluation of health and disease states.
Smart Images

Figure CN123072702A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the fields of TCM meridian detection technology and nonlinear signal analysis technology, specifically to a meridian signal dynamics evaluation method based on phase space reconstruction. The method uses the maximum Lyapunov exponent and fractal dimension to jointly characterize the meridian-five elements functional system from two dimensions: orbital divergence characteristics and phase space geometric complexity, respectively, for the dynamic evaluation of human physiological state. Background Technology
[0002] Understanding the dynamic organization of the human physiological regulatory system in both healthy and diseased states is a core issue in systems medicine and complex life sciences. The human body is not simply composed of independent organs and tissues, but rather a complex system with multi-level interactions, continuous feedback regulation, and time-varying coupling characteristics. In describing the dynamic behavior of such complex systems, the sensitivity of orbits to initial conditions (divergence) and the geometrical filling capacity of phase space attractors (complexity) are two fundamental and independent classical representation dimensions: the former characterizes the degree of instability in the local evolution of the system, while the latter characterizes the fine structure of the overall state space organization of the system. The maximum Lyapunov exponent and fractal dimension are representative nonlinear dynamic indicators for quantifying these two types of characteristics, and have been widely applied in the complexity analysis of heart rate variability, electroencephalograms (EEGs), and other physiological time series.
[0003] In traditional Chinese medicine theory, the meridian system is considered a crucial regulatory network connecting the various organs and tissues of the human body and coordinating overall functions. The Five Elements theory provides a structured description of the mutually promoting and restrictive relationships between different functional systems. From the perspective of complex system dynamics, the health status of the meridian-Five Elements functional system may not merely be manifested in the rise and fall of signal amplitudes at certain acupoints, but more likely in the overall differences in the divergence ability and geometric organization of the system's phase space trajectory: a healthy system may possess stronger orbital divergence ability and richer attractor geometry, while a diseased state may exhibit weakened divergence ability and a degeneration of geometric structure towards lower dimensions. However, currently, there is a lack of quantitative analytical methods for the bioelectrical signals of the meridians that can jointly characterize the aforementioned divergence and geometric complexity.
[0004] Among existing methods for estimating the maximum Lyapunov exponent, the traditional Wolf method relies on long time series and requires trajectory renormalization, both of which are difficult to apply to clinical meridian bioelectrical signals that are primarily observational data, have short time series, and high noise levels. Meanwhile, the traditional box counting method, when estimating the fractal dimension of short-time series, finite-sampling trajectories, is prone to instability due to improper scale interval selection and insufficient sampling points, making it difficult to directly apply to individualized clinical assessment scenarios. Furthermore, most existing analyses of meridian signals or acupoint bioelectrical signals still treat signals as independent measurements or static features for statistical comparison, lacking a joint and robust estimation method that considers both the divergence capability and geometric complexity of phase space trajectories.
[0005] In particular, clinical meridian bioelectric data are usually characterized by short sampling time, limited number of effective sampling points, significant individual differences, and measurement noise. Directly using traditional long-term series kinetic analysis methods is easily affected by outliers, baseline drift, local noise, and parameter selection.
[0006] Therefore, there is an urgent need to provide a trajectory divergence-geometric joint analysis method suitable for short-sequence meridian bioelectric data, which can extract stable dynamic features from two complementary dimensions: the separation velocity of adjacent trajectories and the complexity of phase space scale, so as to achieve quantitative characterization of the functional state of meridians and the five elements. Summary of the Invention
[0007] To address the shortcomings of existing technologies, this invention provides a dual-dimensional dynamic evaluation method for meridian signals based on phase space reconstruction. By constructing a time series of the five-element functional states and reconstructing the phase space, the maximum Lyapunov exponent and fractal dimension are used to quantitatively analyze the system's orbital divergence capability and the geometric complexity of the phase space, forming a divergence-geometric dual-level dynamic characterization system for evaluating the physiological state of the human meridian-five-element functional system.
[0008] The first objective of this invention is to provide a method for modeling the functional state of acupoints and meridians, which is used to robustly construct the temporal sequence of meridian functions from the original bioelectrical signals of acupoints. This method combines robust preprocessing, robust sliding window reconstruction, and phase space embedding strategy, and is suitable for short-time-series, noisy clinical signals. The specific steps are as follows:
[0009] S1. Data Organization and Meridian Object Definition: Through clinical practice, biosignals from the Yuan-points and Jing-points of the twelve meridians were collected from each subject at the same detection timescale. Let the meridian set be... Each of these meridians Each corresponds to a set of primary acupoints. Hejingxue Collection For meridians The original acupoint signal sequences were constructed respectively. and wellbore signal sequence t represents the sampling time. In this invention, each time sequence is approximately 100 sampling points long, corresponding to a 1-minute detection process.
[0010] S2. Determination of Normal Range and Calculation of Instantaneous Abnormality of Individual Acupoints: After data preprocessing and normalization, the deviation of the time-series signal of each acupoint a from the normal range is calculated to characterize the degree of abnormality of that acupoint in the current individual. The normal range is determined based on a healthy population reference database. Specifically, samples of subjects who have undergone clinical screening or whose health status is marked as healthy are selected. Using the same detection equipment, collection process, and preprocessing methods as the individuals to be tested, the bioelectric time series of the same acupoint a is obtained, and a healthy reference sample set is constructed from the samples that pass quality control. ,in, This indicates that the r-th healthy reference subject was at acupoint a and at the sampling time. The pre-processed bioelectrical signal, Indicates the number of healthy reference samples. This represents the number of valid sampling points for the corresponding sample. Based on this healthy reference sample set, calculate the lower and upper bounds of the normal range for acupoint a: , ,in, and These represent the 5th and 95th percentiles of the healthy reference sample set, respectively. Therefore, the normal range for acupoint a is defined as: The time series of pre-processed data for the individual under test at acupoint a Its instantaneous abnormality Defined as in, To prevent extremely small positive numbers with a denominator of zero, the instantaneous abnormality level is set to 0 when the measured signal is within the healthy reference range; when the measured signal is below the lower limit of the normal range or above the upper limit of the normal range, the instantaneous abnormality level increases with the magnitude of the deviation. This index is used to quantify the deviation of acupoint bioelectrical signals from the healthy reference distribution, providing a unified dimensional basis for subsequent acupoint abnormality statistics, calculation of the contribution ratio of source acupoints to well acupoints, and modeling of meridian functional states.
[0011] S3. Overall Time-Average Abnormality of Acupoints: To suppress instantaneous fluctuations and characterize the overall abnormality level of acupoints within a detection period, the time-averaged abnormality is defined as follows: in This corresponds to a 1-minute timeframe. Describes acupoints The overall degree of anomaly within a detection cycle is a coarse-grained representation on a time scale.
[0012] S4. Contribution Ratio of Yuan-Source Points to Jing-Well Points at the Meridian Level: Twelve Yuan-Source points and twelve Jing-Well points were collected clinically. Each meridian contains two acupoint sampling points. After obtaining the time-integrated abnormal index of individual acupoints, information from different acupoints within the same Five Elements was further integrated at the meridian level. For meridian m, its Yuan-Source point abnormal contribution ratio index is defined as: This proportional index characterizes the abnormal contribution weight of the Yuan-point relative to the Jing-point in the overall abnormal structure of the meridian m, thus reflecting the relative relationship between stable regulation (Yuan-point) and dynamic regulation (Jing-point) in the regulatory state of the Five Elements system.
[0013] S5. Adaptive Weighting Mechanism and Meridian Functional State: Further, the abnormal contribution ratio at the meridian level is mapped to adaptive adjustment weights, defined as: .in, The larger the value, the more the five elements' regulatory state leans towards a stable state and overall regulation (dominated by the original acupoint), while The smaller the value, the more dynamic and end-point controlled the system (well-hole dominated). This is a monotonic mapping function used to transform the contribution ratio of acupoints into regulatory weight coefficients at the Five Elements level. Without introducing additional prior assumptions, this mapping preserves the relative order of the proportional relationships, making... It can adaptively reflect the relative tendency of an individual between steady-state regulation (dominant at the original acupoint) and dynamic regulation (dominant at the well-point). Based on the above multi-scale integration and adaptive weighting mechanism, the temporal sequence of the comprehensive functional state of meridian m is defined as follows: .in, Indicates that meridian m is at time m The comprehensive functional state is obtained by fusing the signals from the original acupoints and the well-points under an adaptive weighting mechanism. Based on the theory of the twelve meridians and their corresponding five elements in Traditional Chinese Medicine, all 12 meridians are divided into five categories: wood, fire, earth, metal, and water. The set of meridians corresponding to element E is denoted as... The total number of meridians in the group is The mean aggregation of all meridian time series within the same Five Elements group is used to map the corresponding Five Elements functional state time series. The aggregation formula is: .
[0014] S6. Robust preprocessing of short-sequence meridian signals: In the implementation of this invention, the length of the five-element functional state trajectory for each subject is approximately 1 minute. (100 sampling points) represents typical short-term physiological observation data with a relatively high noise level and significant individual variability. To improve the stability of subsequent phase space structure estimation, this invention employs a multi-level robust preprocessing procedure for each of the five-element functional state trajectories. Processing: (1) Missing value and effective length constraint: After removing missing values, if the effective length is insufficient to support embedding and window reconstruction (e.g., < 30 points), then the sequence will not be subjected to dynamic analysis. (2) IQR outlier suppression: The interquartile range (IQR) method is used to filter extreme outliers and retain intervals. (3) Detrending and robust smoothing: Baseline drift in short time series can significantly affect the topology of embedded trajectories. This invention performs linear detrending on the sequence and then uses small-scale median filtering to suppress spike noise, thereby obtaining a robust basis for local structure estimation.
[0015] S7. Central Difference Derivative Characterization: Considering that the absolute amplitude in short time series is easily affected by individual physical condition, measurement contact state, etc., and that changes in the system's control state are more reflected in local rates of change, this invention uses the central difference approximate derivative to construct a dynamic change sequence: In this method, for sampling points within the sequence, the local rate of change is estimated using the central difference formed by adjacent points. For sampling points at the beginning and end, since there is a lack of adjacent points on one side, boundary compensation is performed using forward and backward differences, respectively. The resulting... The sequence length is consistent with the original sequence, facilitating subsequent robust normalization, sliding window reconstruction, and phase space embedding analysis. The derivative sequence is used to emphasize localized modulation fluctuations and reduce the impact of slow drift on the phase space structure. The derivative enhancement used in this invention is not for constructing higher-order phase space embeddings, but rather as a feature transformation emphasizing local variations. Its purpose is to improve the stability of phase space geometry estimation under short-time-series conditions without changing the intrinsic dynamic dimension of the system.
[0016] S8. A small number of outliers in short time series can significantly affect traditional Min-Max normalization. To enhance cross-individual comparability, this invention uses the 5% and 95% quantiles to construct robust scales: .in , These are the 5th and 95th percentiles of the sequence, respectively. Truncate the output to This processing preserves the overall dynamic range while avoiding extreme values that "stretch" the scale. This operation is only used for numerical scale unification and does not change the relative dynamic structure of the time series.
[0017] S9. Sliding Window Robust Reconstruction and Phase Space Embedding: Phase space reconstruction is a common method for recovering the dynamic structure of a system from a univariate time series. According to Takens' embedding theorem, the topological structure of the original system can be reconstructed in a high-dimensional space through delayed coordinate mapping. For each of the five elements... fusion sequence , build dimensional delay vector: ,in For time delay, The embedding dimension is determined by the number of samples. Under short-term physiological data conditions (approximately 100 sampling points), independent estimation of embedding parameters on a sample-by-sample basis is easily affected by noise and sample length limitations. To improve the consistency of cross-individual comparisons, this invention employs a strategy of "lead estimation + fixed-parameter master analysis" to determine the embedding parameters. The average mutual information (AMI) and pseudo-nearest neighbor (FNN) curves are calculated on representative samples. The results show that the first minimum of AMI is mainly located in... The value is approximately 1 to 3, while the proportion of FNN decreases significantly and tends to stabilize after m ≥ 3. Therefore, in the main analysis, this invention uniformly adopts: , as the embedding parameters for phase space reconstruction. To verify the robustness of the parameter selection, supplementary analysis was conducted on . ∈[1,3] and The parameter perturbations ∈[3,6] were repeatedly calculated, and the results show that the main conclusions do not depend on specific embedding parameter values. Due to the limited sequence length, directly embedding the entire sequence is easily affected by local noise. To improve the stability of trajectory estimation, a sliding window strategy is first used to locally reconstruct the time series. Let the length of the time series X(t) be N. A sliding window of length w is constructed on the time axis and moves along the time axis with a step size s, where w represents the window length and s represents the sliding step size between adjacent windows. In this study, the window length is w=30 and the step size is set to s=0.2w. Savitzky–Golay filtering and Gaussian smoothing are used to locally estimate the sequence within each window: Because of overlap between different windows, the same time point may be estimated by multiple windows. To obtain a uniform sequence representation, overlapping averaging is performed on all window results for fusion: ,in As an indicator function, when time point window The value is 1 when overwriting, and 0 otherwise. For the i-th time point on the original timeline, For the i-th sliding window, To smooth the data at various time points within window w after multiple levels of smoothing. The estimated value. This fusion strategy is equivalent to an ensemble averaging of multiple local estimates at each time point, thereby reducing the bias caused by single-window noise. The final fused sequence is obtained. Used to construct phase space trajectories.
[0018] The second objective of this invention is to provide a joint characterization method for "divergence-geometry" dual-level nonlinear dynamics based on the maximum Lyapunov exponent and fractal dimension. This method achieves robust extraction of orbital divergence capability through multi-segment estimation via a sliding window, stable estimation of phase space geometric complexity through scale interval optimization, and establishes physiological state discrimination rules based on the aforementioned dual indicators for the auxiliary identification of dynamic abnormalities related to health and disease. The specific steps are as follows:
[0019] S1. Maximum Lyapunov exponent estimation: The maximum Lyapunov exponent is used to characterize the sensitivity of a system to initial conditions and is an important indicator for identifying chaotic dynamics. The traditional Wolf method requires a long time series and relies on trajectory renormalization operations, thus limiting its application in physiological system studies that rely mainly on observational data. In contrast, the Rosenstein method only relies on the average exponential divergence behavior of neighboring trajectories in phase space, and has better robustness to short time series and noisy data. Therefore, this paper uses this method to estimate the maximum Lyapunov exponent of the five-element functional state sequence. (1) Rosenstein method: Let the time series of the five-element element E (E∈{J,M,S,H,T}) be as follows. For each element's time series, phase space reconstruction is performed according to the method given in Objective 1, followed by Lyapunov exponent estimation. The phase space state vector can be constructed through delayed coordinate embedding. Where E represents the five elements, For discrete-time indexing. For each element's phase space vector. Find the nearest neighbor vector in the phase space trajectory corresponding to the element. And satisfy the time exclusion constraint To avoid selecting adjacent points on the same trajectory as false neighbors, where This is a time-repulsion window. Let the phase space distance between two adjacent trajectories at the initial time be... As the system evolved, these two trajectories developed into... Their phase space distance can be expressed as: In chaotic systems, the distance between neighboring trajectories increases approximately exponentially with time: ,in For evolutionary step size, The sampling time interval, This represents the largest Lyapunov exponent. Taking the natural logarithm of the above relationship yields: In actual calculations, the average logarithmic distance is obtained by statistically averaging all initial neighboring trajectories: If the system exhibits chaotic behavior, then within a certain time scale... and The two sides exhibit an approximately linear relationship, and the slope of the linear fit is the maximum Lyapunov exponent. To eliminate the influence of sampling time interval on the index estimation, the calculated Lyapunov index is normalized according to the sampling time interval so that it represents the average exponential divergence rate on a unit time scale, thereby ensuring the comparability of estimation results between different individuals. (2) Sliding window robust estimation: Since the length of each five-element functional state sequence is about 100 sampling points, a single MLE estimation is easily affected by local noise. In order to improve the estimation stability, a sliding window strategy is used to estimate the sequence in multiple segments. Let the fused five-element functional state sequence be The sequence length is N. A sliding window of length L is constructed on the time axis and moves along the time axis with a step size s. Within each window, phase space embedding and the Rosenstein algorithm are performed independently to obtain a set of maximum Lyapunov exponent estimates. , where k is the number of windows. Finally, the overall estimate is obtained through statistical fusion: Simultaneously, the standard deviation is calculated to assess the uncertainty of the estimate. This sliding window strategy can reduce the impact of short-term noise on a single estimate, thereby obtaining a more robust orbit divergence rate index.
[0020] S2. Estimation of fractal dimension: After phase space reconstruction, the five-element functional state sequence can be represented as a set of embedded vectors: ,in Let M be the i-th phase space vector, and M be the number of effective embedding points. This set of points can be considered as a finite sampling approximation of the system attractor in the embedding space. The box counting method estimates the geometric complexity of the attractor by counting the number of coverage points of the trajectory point set at different spatial scales. A grid with side length ε is constructed in the embedding space, dividing the entire phase space into several hypercube boxes, and the number of boxes N(ε) containing at least one trajectory point is counted. If the system trajectory has a fractal structure, then it satisfies the power-law scaling relation within a certain scale range: Where D represents the box-counting fractal dimension. Taking the logarithm of both sides yields a linear relationship: Therefore, by different scales Calculate the number of boxes N(ε) in logarithmic coordinates. By performing linear fitting, the fractal dimension can be estimated: In actual calculations, the set of phase space trajectory points is first determined. The bounding box range is defined, and a set of decreasing scales is constructed within this range. Then, the number N of non-empty boxes was counted for each scale. ), and draw and The relationship curve is shown. Since finitely sampled data usually only exhibits an approximately linear relationship in the intermediate scale interval, the fractal dimension estimate is obtained by selecting a linear interval and performing least-squares fitting. The fractal dimension reflects the geometric filling ability of the phase space trajectory in the embedding space. When the system trajectory converges to a low-dimensional attractor, its fractal dimension is usually a non-integer value and strictly smaller than the dimension of the embedding space; if the system dynamics degenerates into a periodic or quasi-periodic orbit, the fractal dimension tends to approach an integer value.
[0021] S3. Therefore, this invention constructs a dual-level nonlinear dynamic joint characterization framework of "divergence-geometricity": at the divergence characteristic level, the Rosenstein method based on phase space reconstruction is used to estimate the maximum Lyapunov exponent, and a statistically robust divergence rate index is obtained by combining a sliding window multi-segment estimation strategy; at the geometric characteristic level, the box counting method is used to estimate the fractal dimension of the phase space trajectory, and the scale-filling characteristics of the attractor in the embedding space are used to characterize whether the system forms a low-dimensional heterogeneous attractor. The two levels correspond to different intrinsic properties of chaotic systems: the maximum Lyapunov exponent characterizes the exponential divergence rate of the trajectory; the fractal dimension describes the geometric complexity of the attractor. The two are independent yet complementary, together constituting a joint characterization of the dynamic complexity of the system.
[0022] S4. Dynamic Differences Between the Healthy and Diseased Groups: At the divergence level, the maximum Lyapunov exponent in the healthy group was significantly higher than that in the diseased group, indicating that the healthy system has a stronger sensitivity to initial conditions and its phase space trajectory exhibits higher exponential divergence. The maximum Lyapunov exponent in the diseased system was generally lower, indicating weakened system dynamic activity. At the geometric complexity level, fractal dimension analysis showed that the phase space trajectory of the healthy system exhibited a clear non-integer-dimensional fractal structure, while the diseased system showed near-integer-dimensional geometric features in some samples, suggesting a degradation in system dynamic complexity. In summary, the Five Elements functional system in a healthy state exhibits a dynamic characteristic of "high divergence—high geometric complexity," while the diseased state exhibits a dynamic evolution pattern of "weakened divergence—degraded geometric structure."
[0023] From the perspective of overall dynamic complexity, the above results reveal that the Five Elements functional system in a healthy state exhibits dynamic characteristics of "high divergence—high geometric complexity," while the disease state shows a dynamic evolution pattern of "weakened divergence—geometric structure degradation." This indicates that the human physiological regulatory system may undergo a transformation from a high-complexity dynamic structure to a low-complexity dynamic structure in a disease state. The aforementioned divergence-geometric dual-level joint characterization framework can achieve stable identification of the system's dynamic structure under short-term physiological data conditions, and reveals the potential differences in dynamic structure between healthy and disease states from the perspective of complex systems, providing a technical means for the objective quantitative evaluation of meridian functional status.
[0024] S5. Establishment and Implementation of Preset Judgment Rules: Based on the maximum Lyapunov exponent of the time series of the five elements' functional states... With fractal dimension Subsequently, this invention further sets preset discrimination rules to convert the aforementioned dynamic indicators into individual health status judgment results. First, based on the labeled set of healthy samples, the five elements are calculated for each subject. On and And calculate the health reference range. Because the healthy group was and All of the above were higher than those in the disease group, and the rules for judging abnormal single indicators are as follows: (like ), (like ),in The preset lower threshold for this indicator can be taken as the 5th percentile of the reference distribution of the healthy group, or it can be determined by the ROC curves of the healthy group and the disease group samples. Furthermore, to avoid misjudgment caused by fluctuations in a single indicator, this invention employs a multi-indicator joint judgment method to construct a comprehensive anomaly score: In one implementation, take ,at this time In the Five Elements system and The total number of abnormal indicators in both metrics. Finally, based on... Compared with the preset comprehensive threshold , The comparison results output the health status determination result: when When it is determined to be a healthy state; when When it is determined to be a sub-healthy or dynamic abnormality risk state; when This is determined to be a disease-related dynamic abnormality. If the five elements and two types of dynamic indicators form a total of 10 judgment items, then settings can be configured. , .
[0025] The present invention has the following advantages over the prior art: (1) This invention constructs a hierarchical functional state time series of "acupoint-meridian-five elements" to achieve a robust conversion from the original acupoint bioelectric signal to the five elements functional state sequence, providing a reliable data foundation for subsequent nonlinear dynamic analysis; (2) In view of the characteristics of short sampling time (about 1 minute), limited effective sampling points (about 100 points) and high noise level of clinical meridian bioelectric signals, this invention designs a multi-level robust preprocessing process and a sliding window robust reconstruction strategy, which effectively suppresses the interference of outliers, baseline drift and local noise on the phase space structure estimation, making the method operable for clinical application; (3) This invention is the first to combine the orbit divergence ability (maximum Lyapunov exponent) and the phase space geometric complexity (fractal dimension) of two independent nonlinear dynamics. The mechanical dimension is jointly characterized to form a "divergent-geometric" two-level dynamic evaluation framework, which overcomes the defect that a single index is difficult to fully characterize the dynamic state of the system and reduces the risk of misjudging the physiological state; (4) This invention improves the estimation robustness of the maximum Lyapunov exponent and fractal dimension under short time series conditions by using a sliding window multi-segment estimation strategy and scale interval optimization, respectively, and solves the instability problem of traditional methods under limited sampling conditions; (5) This invention establishes a comprehensive abnormality score and health reference range discrimination rule based on the joint judgment of dual indicators, which can output three-level judgment results of health status, sub-health / dynamic abnormality risk status and disease-related dynamic abnormal status, providing a directly applicable technical means for the objective quantitative evaluation of the meridian-five elements functional system. Attached Figure Description
[0026] Figure 1 This is a reconstructed phase space trajectory diagram of a healthy person 1 as shown in Example 1. Figure 2 This is a reconstructed phase space trajectory diagram of healthy human 2 as shown in Example 1. Figure 3 This is a reconstructed phase space trajectory diagram of healthy human 3 as shown in Example 1. Figure 4 This is a reconstruction diagram of the phase space trajectory of a diabetic patient as shown in Example 1. Figure 5 This is a reconstructed phase space trajectory diagram of a person with hypertension and hyperglycemia, as shown in Example 1. Figure 6 The image shows a reconstructed phase space trajectory of a cardiac patient as described in Example 1. Figure 7 This is a statistical distribution chart of the maximum Lyapunov index for the healthy group and the disease group, as shown in Example 2. Figure 8 The time evolution diagram of the maximum Lyapunov exponent based on the sliding window, as shown in Example 2. Figure 9 This is a plot of the exponential divergence characteristics of the average logarithmic distance, as shown in Example 2. Figure 10 This is a comparison diagram of the fractal dimensions of the five-element functional states of the healthy group and the disease group, drawn in Example 2. Detailed Implementation
[0027] The present invention will be described in detail below with reference to specific implementation examples. These examples are intended to help those skilled in the art to further understand the invention, but do not limit the invention in any way. Example 1
[0028] The data in this embodiment were collected in a clinical setting using the aforementioned platform and constitute multi-center observational data. Subjects came from multiple collaborating medical institutions, encompassing healthy controls and individuals with disease states, with a cumulative sample size exceeding 6000 and covering multiple age groups. All data were collected under standardized testing protocols to minimize the impact of different centers, operators, and short-term physiological fluctuations on the measurement results. Samples used in this analysis were further screened according to pre-defined quality control standards to ensure the reliability and comparability of subsequent short-term kinetic analyses. During the testing process, subjects sat quietly for approximately 10 minutes before the formal data collection to reduce the impact of short-term environmental disturbances and physiological fluctuations on the signal. Subsequently, the sensor was worn according to the standard acupoint positioning method, and multi-channel synchronous data acquisition was performed for approximately 1 minute. Throughout the entire testing process, subjects remained still, and the data acquisition was a non-invasive procedure, without electrical stimulation or puncture. This process ensured the consistency of signal acquisition conditions and provided a basis for subsequent cross-individual comparisons. The system primarily collects bioelectrical characteristic parameters of the skin surface at acupoints, including conductivity and impedance indicators. Such signals can characterize changes in the body's physiological regulatory state in terms of time and spatial distribution to a certain extent, and therefore can serve as an indirect observational basis for quantitative modeling of meridian-related functional activities. Based on the "acupoint-meridian-five elements" hierarchical state modeling framework proposed in this paper, the original acupoint time series is further integrated into the five-element functional state time series J, M, S, H, and T, which are used as inputs for subsequent phase space reconstruction and nonlinear dynamic analysis. The specific steps are as follows:
[0029] 1. Through clinical practice, biosignals from the source and well points of the twelve meridians were collected from each subject at the same timescale. Let the meridian set be... Each of these meridians Each corresponds to a set of primary acupoints. Hejingxue Collection For meridians The original acupoint signal sequences were constructed respectively. and wellbore signal sequence t represents the sampling time. For each acupoint a, the deviation of its temporal signal from the normal range is calculated to characterize the degree of abnormality of that acupoint in the current individual. This instantaneous degree of abnormality is denoted as: Based on the meridian deficiency / excess and cold / heat balance values, the proportion of syndrome differentiation according to the Eight Principles, and the balance values of syndrome differentiation according to the Six Channels in clinical individual detection of the multi-channel meridian intelligent detection system described in this invention, an acupoint contribution ratio index is constructed by considering the degree of abnormality, the magnitude of deviation, and the degree of stability. This is used to reflect the relative weighting of the Yuan-Source Points and Jing-Well Points within the individual's regulatory structure along the meridian m. The time-averaged degree of abnormality is defined as... in This corresponds to a 1-minute timeframe. Describes acupoints The overall degree of abnormality within a detection period is a coarse-grained representation on a time scale. Twelve meridians (source points and well points) were collected clinically, each containing two acupoint sampling points. After obtaining the time-integrated abnormality index for individual acupoints, information from different acupoints within the same Five Elements was further integrated at the meridian level. For meridian m, the contribution ratio of its source point abnormality is defined as: The abnormal contribution ratio at the meridian level is mapped to an adaptive adjustment weight, defined as: .in, The larger the value, the more the five elements' regulatory state leans towards a stable state and overall regulation (dominated by the original acupoint), while The smaller the value, the more dynamic and end-point controlled the system (well-hole dominated). This is a monotonic mapping function used to transform the contribution ratio of acupoints into regulatory weight coefficients at the Five Elements level. Without introducing additional prior assumptions, this mapping preserves the relative order of the proportional relationships, making... It can adaptively reflect the relative tendency of an individual between steady-state regulation (dominant at the original acupoint) and dynamic regulation (dominant at the well-point). Based on the above multi-scale integration and adaptive weighting mechanism, the temporal sequence of the comprehensive functional state of meridian m is defined as follows: .in, Indicates that meridian m is at time m The comprehensive functional state is obtained by fusing the original acupoint signal and the well acupoint signal under an adaptive weighting mechanism. A multi-level robust preprocessing process is adopted for each of the five-element functional state trajectories. Processing: (1) Missing value and effective length constraint: After removing missing values, if the effective length is insufficient to support embedding and window reconstruction (e.g., < 30 points), then the sequence will not be subjected to dynamic analysis. (2) IQR outlier suppression: The interquartile range (IQR) method is used to filter extreme outliers and retain intervals. For samples within the range, a dynamically changing sequence is constructed using the central difference approximation derivative: The derivative sequence is used to emphasize localized fluctuations and reduce the impact of slow drift on the phase space structure. Robust scales are constructed using the 5% and 95% quantiles. .in , These are the 5th and 95th percentiles of the sequence, respectively. Truncate the output to This processing preserves the overall dynamic range while avoiding extreme values that "stretch" the scale. This operation is only used for numerical scale unification and does not alter the relative dynamic structure of the time series. According to Takens' embedding theorem, the topology of the original system can be reconstructed in high-dimensional space through delayed coordinate mapping. For each of the five elements... fusion sequence , build dimensional delay vector: ,in For time delay, The embedding dimension is determined by the number of samples. Under short-term physiological data conditions (approximately 100 sampling points), independent estimation of embedding parameters on a sample-by-sample basis is easily affected by noise and sample length limitations. To improve the consistency of cross-individual comparisons, this paper adopts a strategy of "lead estimation + fixed-parameter master analysis" to determine the embedding parameters. The average mutual information (AMI) and pseudo-nearest neighbor (FNN) curves are calculated on representative samples. The results show that the first minimum of AMI is mainly located in... The approximation range is approximately 1 to 3, while the proportion of FNNs decreases significantly and stabilizes after m ≥ 3. Therefore, in the main analysis, this paper consistently adopts the following approach: , as the embedding parameters for phase space reconstruction. To verify the robustness of the parameter selection, supplementary analysis was conducted on . ∈[1,3] and The parameter perturbations ∈[3,6] were repeatedly calculated, and the results show that the main conclusions do not depend on specific embedding parameter values. Due to the limited sequence length, directly embedding the entire sequence is easily affected by local noise. To improve the stability of trajectory estimation, this paper first uses a sliding window strategy to locally reconstruct the time series. Let the length of the time series x(t) be N. A sliding window of length w is constructed on the time axis and moves along the time axis with a step size s, where w represents the window length and s represents the sliding step size between adjacent windows. In this study, the window length is w=30, and the step size is set to s=0.2w. Within each window, Savitzky–Golay filtering and Gaussian smoothing are used to locally estimate the sequence: Because of overlap between different windows, the same time point may be estimated by multiple windows. To obtain a uniform sequence representation, overlapping averaging is performed on all window results for fusion: ,in As an indicator function, when time point window The value is 1 when overwriting, and 0 otherwise. For the i-th time point on the original timeline, For the i-th sliding window, To smooth the data at various time points within window w after multiple levels of smoothing. The estimated value. This fusion strategy is equivalent to an ensemble averaging of multiple local estimates at each time point, thereby reducing the bias caused by single-window noise. The final fused sequence is obtained. Used to construct phase space trajectories. To intuitively compare the differences in the dynamic morphology of the five elements' functional states between healthy and diseased states, this study selected a subset that met the quality control standards for analysis, including 100 healthy individuals and 100 diseased individuals. This paper first performs Takens delayed embedding on the five elements' time series of representative individuals from both groups, and then reconstructs two-dimensional and three-dimensional phase diagrams of the five elements' phase space trajectories for representative individuals from the healthy population and the diseased population (patients with hypertension and hyperglycemia, and patients with heart disease).
[0030] 2. The phase space trajectories of the healthy group have a wider coverage and more uniform distribution, exhibiting higher state accessibility and richer trajectory geometry. The disease group, on the other hand, shows varying degrees of trajectory contraction, local aggregation, or "collapse," with a significantly reduced trajectory accessibility, suggesting a simpler system evolution. Within the disease group, several representative chronic diseases were selected, including diabetes and heart disease. The phase space trajectories of diabetes and heart disease patients show significant structural differences. In Traditional Chinese Medicine, the core of chronic diseases such as diabetes and hypertension is an imbalance of Earth (spleen system) and Water (kidney system), often accompanied by a chain reaction of imbalances in Wood (liver system), Fire (heart system), and Metal (lung system). The phase space trajectory diagram drawn by this invention shows that in diabetes and hypertension patients, the phase space trajectories corresponding to Water and Earth elements are significantly contracted, and the trajectory morphology tends to be simpler; while the structural changes in the trajectories of other elements are relatively weaker. In heart disease patients, the phase space trajectories corresponding to Water and Fire elements are significantly contracted, producing a chain reaction of imbalances in Wood and Metal elements. The phase space trajectories also show a contracted form, and their change patterns differ from those of diabetes patients, showing different Five Elements level structural characteristics. Example 2
[0031] 1. Using the divergent-geometric-structural-discriminative multi-level nonlinear dynamics characterization method provided in Objective 2, the maximum Lyapunov exponent estimation is first performed: (1) Rosenstein method: Let the time series of the five elements E (E∈{J,M,S,H,T}) be as follows This paper performs phase space reconstruction and Lyapunov exponent estimation on the time series of each element. A phase space state vector can be constructed through delayed coordinate embedding. Where E represents the five elements, For discrete-time indexing. For each element's phase space vector. Find the nearest neighbor vector in the phase space trajectory corresponding to the element. And satisfy the time exclusion constraint To avoid selecting adjacent points on the same trajectory as false neighbors, where This is a time-repulsion window. Let the phase space distance between two adjacent trajectories at the initial time be... As the system evolved, these two trajectories developed into... Their phase space distance can be expressed as: In chaotic systems, the distance between neighboring trajectories increases approximately exponentially with time: ,in For evolutionary step size, The sampling time interval, This represents the largest Lyapunov exponent. Taking the natural logarithm of the above relationship yields: In actual calculations, the average logarithmic distance is obtained by statistically averaging all initial neighboring trajectories: If the system exhibits chaotic behavior, then within a certain time scale... and The two sides exhibit an approximately linear relationship, and the slope of the linear fit is the maximum Lyapunov exponent. To eliminate the influence of sampling time interval on the index estimation, this paper normalizes the calculated Lyapunov index according to the sampling time interval, so that it represents the average exponential divergence rate on a unit time scale, thereby ensuring the comparability of estimation results between different individuals. (2) Sliding window robust estimation: Since the length of each five-element functional state sequence is about 100 sampling points, a single MLE estimation is easily affected by local noise. In order to improve the estimation stability, this paper adopts a sliding window strategy to perform multi-segment estimation of the sequence. Let the fused five-element functional state sequence be The sequence length is N. A sliding window of length L is constructed on the time axis and moves along the time axis with a step size s. Within each window, phase space embedding and the Rosenstein algorithm are performed independently to obtain a set of maximum Lyapunov exponent estimates. , where k is the number of windows. Finally, the overall estimate is obtained through statistical fusion: Simultaneously, the standard deviation is calculated to assess the uncertainty of the estimate.
[0032] 2. Next, we estimate the fractal dimension: After reconstructing the phase space, the five-element functional state sequence can be represented as a set of embedding vectors: ,in Let M be the i-th phase space vector, and M be the number of effective embedding points. This set of points can be considered as a finite sampling approximation of the system attractor in the embedding space. The box counting method estimates the geometric complexity of the attractor by counting the number of coverage points of the trajectory point set at different spatial scales. A grid with side length ε is constructed in the embedding space, dividing the entire phase space into several hypercube boxes, and the number of boxes N(ε) containing at least one trajectory point is counted. If the system trajectory has a fractal structure, then it satisfies the power-law scaling relation within a certain scale range: Where D represents the box-counting fractal dimension. Taking the logarithm of both sides yields a linear relationship: Therefore, by different scales Calculate the number of boxes N(ε) in logarithmic coordinates. By performing linear fitting, the fractal dimension can be estimated: In actual calculations, this paper first determines the set of phase space trajectory points. The bounding box range is defined, and a set of decreasing scales is constructed within this range. Then, the number N of non-empty boxes was counted for each scale. ), and draw and The relationship curve is shown. Since finitely sampled data typically exhibits an approximately linear relationship only in the intermediate scale range, the fractal dimension estimate is obtained by selecting a linear interval and performing least-squares fitting. The selection of the linear interval is determined as follows: the local slope between adjacent scale points is calculated, and the interval with a local slope change rate of less than 5% and continuously covering at least 5 scale points is selected as the linear interval for least-squares fitting to ensure the objectivity and repeatability of the estimation results.
Claims
1. A two-dimensional dynamic evaluation method for meridian signals based on phase space reconstruction, characterized in that, comprising the following steps: S1: Acupoint-meridian functional state sequence construction: acquiring bioelectric time series of a plurality of meridian-related acupoints of the human body, and performing hierarchical fusion on the bioelectric time series according to the correspondence between acupoints and meridians and between meridians and five-element functional categories, to obtain metal, wood, water, fire and earth five-element functional state time series; S2: Phase space reconstruction: performing robust preprocessing on the five-element functional state time series, and performing phase space reconstruction on the preprocessed sequence to obtain phase space trajectories corresponding to each five-element element; S3: Two-dimensional dynamic feature extraction: establishing a divergence-geometry two-level analysis system based on the phase space trajectories, and extracting indices; wherein the divergence layer uses the Rosenstein method to calculate the maximum Lyapunov exponent, and the geometry layer uses the box-counting method to estimate the fractal dimension; S4: Physiological state discrimination: outputting a physiological state assessment result of the meridian-five-element functional system based on preset discrimination rules according to the maximum Lyapunov exponent and the fractal dimension.
2. The method according to claim 1, characterized in that, In step S1, the hierarchical fusion includes: S1-1: Determine the normal range of each acupoint based on a reference database of healthy individuals, and calculate the instantaneous degree of abnormality of the bioelectric time series of each acupoint a relative to the normal range. ; S1-2: Calculate the time-averaged abnormality of each acupoint. , where T is the total number of sampling points; S1-3: For each meridian m, calculate the proportion of abnormal contribution from its original acupoint. ,in This refers to the set of acupoints corresponding to meridian m. This refers to the set of well points corresponding to meridian m. S1-4: The abnormal contribution ratio of the original acupoint Mapped to adaptive adjustment weights ,in It is a monotonic mapping function; S1-5: Based on the aforementioned adaptive adjustment weights, the source acupoint signals and well acupoint signals of the same meridian are weighted and fused to obtain the comprehensive functional state time series of each meridian. And mapped to the time series of the functional states of the five elements according to the correspondence between the twelve meridians and the five elements. .
3. The method according to claim 1, characterized in that, In step S2, the robust preprocessing includes: missing value filtering, IQR outlier suppression, linear detrending, median smoothing, central difference enhancement and quantile scaling; the central difference enhancement adopts forward difference for the first point of the sequence, central difference for the intermediate points, and backward difference for the last point; the quantile scaling uses 5% and 95% quantiles to construct a robust scale, and maps the sequence to the [0,1] interval.
4. The method according to claim 1, characterized in that, In step S2, before performing phase space reconstruction on the five-element functional state time series, the method further comprises a sliding window robust reconstruction step: constructing a sliding window with a length of W on the time axis, and moving the sliding window along the time axis at a step size S=0.2W; performing local estimation in each window by adopting Savitzky-Golay filtering and Gaussian smoothing, and obtaining a fused sequence through overlapping average fusion; the phase space reconstruction adopts delay coordinate embedding to construct m-dimensional delay vectors, with a time delay τ=2 and an embedding dimension m=3.
5. The method according to claim 1, characterized in that, In step S3, the diverging layer estimates the maximum Lyapunov exponent using the Rosenstein method, including: finding the nearest neighbor vector for each state vector in the phase space and applying a time-repulsion constraint. To avoid selecting adjacent points on the same trajectory as false neighbors, where A time-repulsion window is used; the average logarithmic distance between neighboring trajectories is calculated at different evolution step sizes k. ,right The maximum Lyapunov exponent λ1 is obtained by least-squares fitting of the linear growth interval.
6. The method according to claim 1, characterized in that, In step S3, the geometric layer estimates the fractal dimension using box counting, including: constructing a side length of... The grid is used to count the number of non-empty boxes N( ), and The linear intervals in the relationship curve are selected for least-squares fitting to estimate the fractal dimension D.
7. The method according to claim 1, characterized in that, In step S4, the preset discrimination rule includes: determining the health reference range of each indicator based on the healthy sample set. The rule for determining anomalies in a single indicator is: when... If the condition is met, it is considered abnormal; otherwise, it is considered normal. The lower limit threshold is preset; multiple indicators are used to jointly determine the anomaly score S, and the three-level judgment result is output.
8. The method according to claim 7, characterized in that, The comprehensive anomaly score ,in and These are the single-index anomaly judgment results for λ1 and D corresponding to the five elements; A healthy state is output when S<T1, a sub-health or dynamic abnormality risk state is output when T1≤S<T2, and a disease-related dynamic abnormality state is output when S≥T2.
9. The method according to claim 8, characterized in that, Said T1=2, said T2=4.
10. The method according to claim 1, characterized in that, Said acupoints include the source acupoints and well acupoints of the twelve meridians, and the acquisition duration of said bioelectric time series is 1 minute, corresponding to 100 sampling points.
