An information processing system for acupuncture clinical auxiliary decision making

Abstract

This invention relates to the field of healthcare informatics technology and discloses an information processing system for clinical decision support in acupuncture, comprising: a physiological parameter feature extraction module, a graph theory topology construction module, a decision scheme calculation module, and a feedback correction module. The system extracts the temporal statistical features of physiological impedance data streams to generate state vectors, constructs a meridian topology directed graph by combining the spatial distribution relationship of sites, and corrects the transmission edge weights using node inertia penalty terms. By solving for the minimum value of the target cost function, the target intervention site set is determined. This invention eliminates the lack of tissue fatigue perception in traditional schemes by quantifying physiological inhibition cycles and state inertia, realizes dynamic optimization of meridian topology energy, ensures intervention safety, and improves the decision accuracy of acupoint selection.

Description

Technical Field

[0001] This invention relates to the field of healthcare informatics technology, specifically to an information processing system for clinical decision support in acupuncture. Background Technology

[0002] Current conventional diagnostic and treatment assistance solutions employ a discrete acquisition architecture, collecting physiological parameters of the target area to characterize the patient's physiological state and provide objective data support for clinical diagnosis. The acupuncture intervention process relies on the mapping relationship between physiological sensing signals and meridian topological paths. Human physiological tissues exhibit obvious nonlinear conduction characteristics and dynamic compensation mechanisms, and the distribution of physiological parameters on the body surface has complex information network characteristics. This means that the intervention action not only acts on the physical target point but also induces the migration of the overall energy state of the meridian network. In addition to the insufficient discretization of sensing hardware, the decision-making algorithm logic is difficult to meet the needs of continuous intervention. For example, Chinese invention patent application with publication number CN120977509A discloses a TCM diagnostic and treatment assistance decision-making method and system, which extracts multimodal features of tongue appearance, pulse appearance, and spleen meridian impedance, and uses spatiotemporal convolutional networks and graph attention mechanisms to realize syndrome classification and dosage adjustment.

[0003] However, existing solutions treat physiological representations as transient numerical slices, ignoring the inherent high-dimensional topological constraints of the meridian system. Even when attempts are made to increase the sampling frequency or add monitoring channels to obtain more data, the lack of perception of physiological refractory periods in the underlying computational model leads to serious information redundancy. This simple expansion of data volume does not solve the problem of the lack of quantification of fatigue accumulation in biological nodes under continuous intervention, causing the model to still output intervention instructions to targets that are already in physiological fatigue or information overload in multiple iterations, resulting in clinical intervention risks.

[0004] Therefore, the technical problem to be solved by this invention is how to construct a processing architecture that senses the capacity constraints of biological networks and realizes node inertial calibration, and how to avoid the risk of physiological overload and improve decision-making accuracy by reconstructing the topological potential energy calculation logic. Summary of the Invention

[0005] This invention proposes an information processing system for auxiliary decision-making in acupuncture clinical practice, comprising:

[0006] The physiological parameter feature extraction module is used to acquire the physiological impedance data stream of the candidate intervention sites, calculate the response statistical features of the physiological impedance data stream in the time domain dimension, and generate a state vector characterizing the physiological response saturation of the candidate intervention sites.

[0007] The graph theory topology construction module is used to establish a connection matrix based on the spatial distribution relationship between each candidate intervention site, and to construct a directed meridian topology graph in combination with state vectors;

[0008] The decision-making module includes a node inertia calibration unit and a topology edge weight adjustment unit. The node inertia calibration unit is used to calculate the inertia penalty term representing the physiological inhibition cycle based on the historical response deviation of each candidate intervention site within a preset sliding time window. The topology edge weight adjustment unit is used to perform a fusion operation on the inertia penalty term and the transmission edge weight distribution in the meridian topology directed graph to construct the target cost function. By solving the energy minimum value of the target cost function, the target intervention site set is determined.

[0009] The feedback correction module is used to obtain the response recovery slope after intervention on the target intervention site set, and correct the convergence step size of the target cost function accordingly. The response recovery slope is the derivative of the physiological impedance after intervention with time.

[0010] Preferably, during the generation of the state vector, the physiological parameter feature extraction module identifies the physiological response threshold of the candidate intervention site by analyzing the local variance and cumulative energy changes of the physiological impedance data stream; when the physiological response threshold decays, the physiological parameter feature extraction module increases the penalty weight component in the state vector.

[0011] Preferably, when calculating the inertial penalty term, the node inertial calibration unit extracts the first derivative of the state vector of the candidate intervention site within a preset sliding time window to determine the physiological state evolution trend of the candidate intervention site; when the first derivative of the state vector shows a decreasing trend, the node inertial calibration unit increases the inertial penalty term to reduce the distribution weight of the candidate intervention site in the target intervention site set.

[0012] Preferably, the graph theory topology construction module includes a topology adjacency matrix construction unit; the topology adjacency matrix construction unit establishes the initial connectivity path of the meridian topology directed graph based on the biological connectivity attributes of the candidate intervention sites in the human meridian model, and dynamically corrects the conduction threshold of the initial connectivity path according to the state vector.

[0013] Preferably, the decision-making solution module determines the target intervention site set through the following steps: Step S1, extract the change characteristics of the state vector of each node in the meridian topology directed graph within a preset sliding time window; Step S11, determine whether each node is in the physiological response sensitive area based on the change characteristics; Step S12, perform energy conduction inhibition processing on nodes outside the physiological response sensitive area to guide the intervention energy to diffuse to the peripheral healthy connected nodes.

[0014] Preferably, the feedback correction module includes a residual analysis unit; the residual analysis unit calculates the logical residual between the actual physiological impedance change value after intervention and the predicted change value output by the decision scheme solution module; the feedback correction module adjusts the parameter distribution in the topology edge weight adjustment unit in reverse according to the magnitude of the logical residual.

[0015] Preferably, the physiological signals acquired by the physiological parameter feature extraction module also include infrared thermal radiation distribution feature data characterizing the metabolic level of local tissues; the physiological parameter feature extraction module maps the infrared thermal radiation distribution feature data into an auxiliary correction component of the state vector to adjust the characterization accuracy of the state vector's response to the candidate intervention site.

[0016] Preferably, the decision-making solution module utilizes a potential energy convergence algorithm based on graph neural networks to extract the topological features of the meridian topology directed graph through a multilayer sensing network, and uses the gradient descent method to solve for the global minimum of the target cost function, generating a set of target intervention sites with physical topological constraints.

[0017] Preferably, the system further includes an output interface unit; the output interface unit is connected to the decision scheme calculation module, obtains the target intervention site set, and translates the target intervention site set into auxiliary decision suggestion data output representing the intervention location coordinates and intervention intensity weights.

[0018] The beneficial effects of this invention are:

[0019] 1. In the information processing of acupuncture clinical auxiliary decision-making, the impedance sequence integral variance within a continuous sliding time window is calculated by the dynamic baseline calibration unit in the physiological state vectorization module, and compared with the baseline tolerance threshold to filter out high-frequency disturbance components that exceed the limit. This processing mechanism ensures that the input state vector entering the subsequent stage can accurately represent the real slow-changing characteristics of the physiological target area, eliminate data noise caused by skin sweat, micro-displacement or unstable epidermal contact state, provide stable and objective raw data support for subsequent high-dimensional topology mapping, and ensure the underlying reliability of the auxiliary decision-making instruction generation process.

[0020] 2. The first derivative of the state vector of the target node within the sliding time window is extracted using the dynamic penalty unit of the transmission edge weights, and it is used as the core operator to characterize the physiological deviation deterioration gradient. When the deterioration gradient exceeds the preset capacity tolerance threshold, the processor automatically performs a decay operation on the transmission edge weights of the associated path, thereby simulating the transmission saturation and physiological blockage phenomenon of biological tissue under excessive stimulation at the algorithm logic level. This processing method based on data feedback to dynamically adjust the graph structure weights avoids the drawback of traditional computing architectures blindly allocating intervention weights when facing drastically fluctuating target areas, enabling the system to have the adaptive adjustment capability to perceive the physiological bearing limit.

[0021] 3. The node inertia adaptive calibration unit extracts the cumulative variance of the impedance sequence of the target node within the global historical time window, establishes the inertia penalty coefficient characterizing the physiological fatigue level of the node, and injects it into the network equilibrium cost function in conjunction with the transmission edge weight matrix. This multi-dimensional constraint linkage mechanism enables the system to automatically identify and avoid nodes in the physiological refractory period or with drastic historical fluctuations during the process of solving the global topological potential energy convergence, and forcibly guides the compensation energy to diffuse to the healthy connected nodes on the periphery. This process does not rely on manually preset empirical rules, but spontaneously reproduces the advanced diagnostic and treatment wisdom of avoiding the real and attacking the weak and selecting acupoints at the far end in traditional Chinese medicine through the underlying logic of graph theory energy minimization, thereby improving the decision-making safety in continuous intervention environments. Attached Figure Description

[0022] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0023] Figure 1 This is a diagram showing the multi-module collaboration and topology architecture of the acupuncture clinical auxiliary decision-making system of the present invention;

[0024] Figure 2 This is a diagram illustrating the data flow and operation steps of the information processing system of the present invention. Detailed Implementation

[0025] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0026] An information processing system for auxiliary decision-making in acupuncture clinical practice, comprising:

[0027] The physiological parameter feature extraction module is used to acquire the physiological impedance data stream of the candidate intervention sites, calculate the response statistical features of the physiological impedance data stream in the time domain dimension, and generate a state vector characterizing the physiological response saturation of the candidate intervention sites.

[0028] The graph theory topology construction module is used to establish a connection matrix based on the spatial distribution relationship between each candidate intervention site, and to construct a directed meridian topology graph in combination with state vectors;

[0029] The decision-making module includes a node inertia calibration unit and a topology edge weight adjustment unit. The node inertia calibration unit is used to calculate the inertia penalty term representing the physiological inhibition cycle based on the historical response deviation of each candidate intervention site within a preset sliding time window. The topology edge weight adjustment unit is used to perform a fusion operation on the inertia penalty term and the transmission edge weight distribution in the meridian topology directed graph to construct the target cost function. By solving the energy minimum value of the target cost function, the target intervention site set is determined.

[0030] The feedback correction module is used to obtain the response recovery slope after intervention on the target intervention site set, and correct the convergence step size of the target cost function accordingly. The response recovery slope is the derivative of the physiological impedance after intervention with time.

[0031] Preferably, during the generation of the state vector, the physiological parameter feature extraction module identifies the physiological response threshold of the candidate intervention site by analyzing the local variance and cumulative energy changes of the physiological impedance data stream; when the physiological response threshold decays, the physiological parameter feature extraction module increases the penalty weight component in the state vector.

[0032] Preferably, when calculating the inertial penalty term, the node inertial calibration unit extracts the first derivative of the state vector of the candidate intervention site within a preset sliding time window to determine the physiological state evolution trend of the candidate intervention site; when the first derivative of the state vector shows a decreasing trend, the node inertial calibration unit increases the inertial penalty term to reduce the distribution weight of the candidate intervention site in the target intervention site set.

[0033] Preferably, the topological edge weight adjustment unit follows the following quantization rules when constructing the target cost function: Where E is the global topological potential minimum. Let i be the weight of the conduction edge between point i and point j in the directed graph of meridian topology. λ represents the spatial distribution distance between each candidate intervention site, and λ is a preset adjustment factor. This is the inertial penalty term determined by the nodal inertial calibration unit.

[0034] Preferably, the graph theory topology construction module includes a topology adjacency matrix construction unit; the topology adjacency matrix construction unit establishes the initial connectivity path of the meridian topology directed graph based on the biological connectivity attributes of the candidate intervention sites in the human meridian model, and dynamically corrects the conduction threshold of the initial connectivity path according to the state vector.

[0035] Preferably, the decision-making solution module determines the target intervention site set through the following steps: Step S1, extract the change characteristics of the state vector of each node in the meridian topology directed graph within a preset sliding time window; Step S11, determine whether each node is in the physiological response sensitive area based on the change characteristics; Step S12, perform energy conduction inhibition processing on nodes outside the physiological response sensitive area to guide the intervention energy to diffuse to the peripheral healthy connected nodes.

[0036] Preferably, the feedback correction module includes a residual analysis unit; the residual analysis unit calculates the logical residual between the actual physiological impedance change value after intervention and the predicted change value output by the decision scheme solution module; the feedback correction module adjusts the parameter distribution in the topology edge weight adjustment unit in reverse according to the magnitude of the logical residual.

[0037] Preferably, the physiological signals acquired by the physiological parameter feature extraction module also include infrared thermal radiation distribution feature data characterizing the metabolic level of local tissues; the physiological parameter feature extraction module maps the infrared thermal radiation distribution feature data into an auxiliary correction component of the state vector to adjust the characterization accuracy of the state vector's response to the candidate intervention site.

[0038] Preferably, the decision-making solution module utilizes a potential energy convergence algorithm based on graph neural networks to extract the topological features of the meridian topology directed graph through a multilayer sensing network, and uses the gradient descent method to solve for the global minimum of the target cost function, generating a set of target intervention sites with physical topological constraints.

[0039] Preferably, the system further includes an output interface unit; the output interface unit is connected to the decision scheme calculation module, obtains the target intervention site set, and translates the target intervention site set into auxiliary decision suggestion data output representing the intervention location coordinates and intervention intensity weights.

[0040] Example 1: In a scenario of chronic neurological function assessment and auxiliary analysis involving continuous high-frequency impedance data monitoring, the processing system continuously receives multi-channel physiological impedance data streams from candidate intervention sites transmitted by a body surface acquisition network. The data evolution exhibits a non-linear compensatory state, with specific tissue nodes showing physiological response saturation and conduction blockage after long-term information collection. The data processing logic, based on static meridian connectivity attributes and a fixed edge weight model, does not incorporate the state memory parameters of biological nodes. The computational architecture continues to allocate intervention weights to target points in a state of high-frequency fluctuation and fatigue accumulation. The solver gets stuck in a state of local high-dissipation node information overload during multiple iterations and fails to transfer the compensation energy optimization path to peripheral healthy connectivity nodes. The physiological parameter feature extraction module obtains data transmitted from the front-end acquisition device and based on system-level... The default license agreement completes the de-identification of the physiological impedance data stream of candidate intervention sites. By analyzing the local variance and cumulative energy changes of the physiological impedance data stream, the physiological response threshold of each candidate intervention site is identified. The response statistical characteristics of the physiological impedance data stream in the time domain are calculated, and a state vector representing the physiological response saturation of the candidate intervention sites is generated. The physiological parameter feature extraction module increases the penalty weight component in the state vector when the physiological response threshold decays. The graph theory topology construction module establishes the connection matrix based on the biological connectivity attributes and spatial distribution relationships of each candidate intervention site in the human meridian model. The state vector is injected into the initial connectivity path of the meridian topology directed graph, and the conduction threshold of the initial connectivity path is dynamically corrected to construct a meridian topology directed graph carrying real-time node state weights.

[0041] The decision-making module includes a node inertia calibration unit and a topology edge weight adjustment unit. The node inertia calibration unit extracts the first derivative of the state vector of each candidate intervention site within a preset sliding time window to determine the physiological state evolution trend of the candidate intervention sites. When the first derivative of the state vector shows a decreasing trend, the node inertia calibration unit determines that the candidate intervention site has entered a physiological inhibition cycle, extracts the historical response deviation of the candidate intervention site within the sliding time window, calculates the inertia penalty term representing the physiological inhibition cycle, and reduces the distribution weight of the candidate intervention site in the target intervention site set. The topology edge weight adjustment unit integrates the inertia penalty term with the transmission edge weight distribution in the directed graph of the meridian topology to construct the target cost function. Where E is the parameter for the global topological potential minimum. Let be the transmission edge weight parameter between point i and point j in the directed graph of meridian topology. The spatial distribution distance between each candidate intervention site is measured in millimeters, and λ is a preset adjustment factor. The inertial penalty term representing the state of the corresponding site is calculated for the node inertial calibration unit; the decision scheme solution module solves the energy minimum of the target cost function, causing high-fatigue nodes outside the physiological response sensitive zone to generate a state transition calculation barrier, and the intervention energy diffuses to the surrounding healthy connected nodes, determining the target intervention site set. The state transition calculation barrier and intervention energy diffusion logic generated by this system are essentially the system's allocation operation of data flow and decision weight distribution within the software solution space. Since this system itself does not have any physical energy emission device that directly acts on the human body, the suppression processing here is only reflected in artificially setting the conduction weight of the corresponding target node to infinity when constructing the directed graph adjacency matrix. By shielding the fatigue node on the algorithm path, the finally calculated command data naturally points to the surrounding area. The system identifies healthy nodes and guides externally connected physical therapy devices to perform corresponding energy transfer stimulation through a data output interface. The output interface unit connects to the decision-making solution module to obtain the target intervention site set, which is then translated into auxiliary decision-making suggestion data output representing the intervention location coordinates and intervention intensity weights. The feedback correction module obtains the derivative value of the physiological impedance change over time after intervention on the target intervention site set as the response recovery slope. It calculates the logical residual between the actual physiological impedance change value after intervention and the predicted change value output by the decision-making solution module, adjusts the parameter distribution in the topology edge weight adjustment unit based on the logical residual, corrects the convergence step size of the target cost function, and transforms the target selection into a dynamic optimization procedure in the data space based on graph theory energy minima and node memory constraints.

[0042] Example 2: In a scenario involving continuous high-frequency impedance data monitoring for chronic neurological function assessment and auxiliary analysis, to verify the quantitative accuracy of the processing system in sensing tissue fatigue and the stability of topological energy dynamic optimization, this experiment constructed a multi-channel impedance acquisition hardware test platform with a 1000Hz sampling frequency and 24-bit analog-to-digital conversion accuracy. The experimental data originated from a desensitized 500-hour continuous multi-node surface impedance open-source dataset. Gaussian white noise with a signal-to-noise ratio of 20dB and power frequency interference harmonics with a frequency of 50Hz were actively injected into the raw data stream to simulate the actual electromagnetic interference environment. The preset sliding time window length was determined based on the fluctuation frequency of the monitored physiological impedance data stream; as the fluctuation frequency increased, the preset sliding time window length decreased accordingly. Based on this logic, the baseline value for the preset sliding time window length is determined to be 10 minutes, and its underlying physical effective boundary is defined as 5 to 15 minutes. The physiological parameter feature extraction module receives the physiological impedance data stream with injected noise. This module filters out 50Hz power frequency interference and background noise through internal differential amplification and filtering links, calculates the local variance and cumulative energy change of the physiological impedance data stream in the time domain dimension, and generates a state vector characterizing the physiological response saturation of the candidate intervention site. In order to verify the global topological synergistic effect generated by the inertial penalty term calculated by the node inertial calibration unit and the physical properties of the parameter boundary, this experiment sets up an experimental group, a locally missing control group, an out-of-range lower limit control group, and an out-of-range upper limit control group. The experimental group adopts the objective cost function. The preset adjustment factor λ was set to 1.0, and the inertia penalty term was removed from the local missing control group. A simulated static edge-weighted graph neural network solver was used. The preset adjustment factor λ was set to 0.2 for the control group with the lower limit of the out-of-range and 2.5 for the control group with the upper limit of the out-of-range. The node inertial calibration unit extracted the first derivative of the state vector of each candidate intervention site within a 10-minute sliding time window. When the first derivative of the state vector showed a downward trend, the historical response deviation was extracted to calculate the inertial penalty term.

[0043] The topology edge weight adjustment unit integrates the inertia penalty term with the transmission edge weight distribution in the meridian topology directed graph. The decision scheme solution module uses the gradient descent algorithm to solve for the energy minimum of the target cost function and outputs the target intervention site set. The original input data containing perturbations shows a 15% measurement baseline drift. The state vector processed by the physiological parameter feature extraction module suppresses the baseline deviation to below 2.5%. Due to the lack of state history memory constraints, the solver continuously assigns intervention weights to sensitive nodes that enter the physiological inhibition cycle after 10 minutes in the locally missing control group. The target physiological response saturation of the locally missing control group rises to 85%, and the response recovery slope decreases to 0.12. The experimental group feeds back the inertia penalty term output by the node inertia calibration unit to the topology edge weight adjustment unit, creating a state transition calculation barrier in the early stage of target physiological inhibition. The experimental group diffuses the intervention energy optimization path to the peripheral healthy connected nodes, maintaining the physiological response saturation at 35% and making the response recovery slope... The rate rebounded to 0.78. The above data comparison confirmed the nonlinear synergistic effect generated by the fusion of dynamic state memory and static topological connectivity attributes. The physical boundary verification data set for the preset adjustment factor showed that when the preset adjustment factor was below the lower limit of 0.2, the system exhibited a node switching delay of 45 minutes; when the preset adjustment factor exceeded the upper limit of 2.5, the system generated high-frequency oscillation switching between peripheral nodes, and the spatial distribution stability of the target intervention site set dropped to 42%. The above performance inflection point data determined the parameter calibration range of the preset sliding time window length and the preset adjustment factor. The data output by the processing system verified the data space dynamic optimization mechanism based on graph theory energy minima and node memory constraints. The system quantifies the physiological inhibition cycle and state inertia, establishes a coupling model of the first derivative of the state vector and the transmission edge weights, and converts the physical state of continuous energy input and local tissue fatigue into dynamic topological edge weight reconstruction logic, outputting the target intervention site set with feedback correction of the convergence step size.

[0044] Example 3: This example combines Figures 1 to 2 This paper describes an information processing system for clinical decision support in acupuncture, such as... Figure 1As shown, the spatial distribution relationship is specifically represented by the distribution among the candidate intervention sites, which is input into the graph theory topology construction module. Simultaneously, the physiological impedance data stream, as the time-domain data of the candidate intervention sites, is input into the physiological parameter feature extraction module. This module calculates the response statistical features and generates a state vector characterizing the physiological response saturation. The physiological parameter feature extraction module outputs two branches. One branch, combined with the state vector, is input into the graph theory topology construction module to establish a connection matrix and construct a directed meridian topology graph based on the state vector. The other branch outputs the historical response deviation to the node inertial calibration unit, which then calculates the deviation based on the historical data. The deviation calculation characterizes the inertial penalty term of the physiological inhibition cycle; the meridian topology directed graph output by the graph theory topology construction module and the fusion inertial penalty term output by the node inertial calibration unit are jointly input into the topology edge weight adjustment unit, which constructs the target cost function and solves the energy minimum value to determine the intervention site set; the topology edge weight adjustment unit determines the target intervention site set to provide the final auxiliary decision result, and then the target intervention site set generates the response recovery slope after the intervention is completed and inputs it into the feedback correction module. The feedback correction module obtains the convergence step size of the target cost function after the intervention response recovery slope and feeds back the convergence step size of the target cost function to the topology edge weight adjustment unit.

[0045] like Figure 2 As shown, the front-end acquisition device executes data flow in two directions. One direction, represented by a solid line, executes the pre-calibration procedure for on-site deployment, and the auxiliary operations of injecting discrete detection signals and establishing initial biological connectivity attributes are led out via dashed lines. The other direction executes the operation of acquiring physiological impedance data stream. This operation node is associated with the processing of response statistical characteristics within a two-layer preset sliding time window via dashed lines. Simultaneously, the main flow flows downward along the solid line to the operation node for generating state vectors. This state vector generation node receives the operation input of adjusting the penalty weight component in the state vector via dashed lines. The data flow then flows downward along the solid line to the operation node for constructing the target cost function. This node directly points to the target intervention site set via solid lines. The operation process involves calculating the minimum energy value via a dashed line and then again determining the target intervention site set via a dashed line. During the determination of the target intervention site set, the dashed line also receives signal input from the energy conduction suppression processing node. The data from this stage then flows along a solid line to the output node that translates the data into auxiliary decision-making suggestions, and is finally transmitted to the output interface unit. Furthermore, the target physical entity is directly associated via a unidirectional solid line with the operation node that obtains the response recovery slope after intervention on the target intervention site set. This node executes the convergence step size of the modified target cost function downwards along the solid line and finally feeds back the modified parameters upwards to the operation node that determines the target intervention site set via a dashed line.

[0046] Example 4: Current chronic neurological function assessment and auxiliary analysis scenarios face continuous high-frequency impedance data monitoring. The multi-channel physiological impedance data stream of candidate intervention sites transmitted by the body surface acquisition network exhibits nonlinear baseline drift caused by electrode polarization and microenvironment temperature fluctuations. The static baseline model calculates historical response deviations, introducing cumulative measurement errors, leading to distortion of the inertial penalty term output by the node inertial calibration unit. The distribution of the transmission edge weights in the directed graph of the meridian topology generated by the graph theory topology construction module becomes distorted, and the decision-making solution module calculates the target cost function, which gets trapped in a pseudo-local minimum, reducing the optimization accuracy of the target intervention site set. The physiological parameter feature extraction module acquires the physiological impedance data stream and calculates the response statistical features within a double-layer preset sliding time window. Based on the nonlinear capacitance compensation characteristics of biological tissues, the cell membrane capacitance charging and discharging efficiency decays as cells continuously receive electrical signals, manifested as the convergence of the time-domain fluctuation variance of the impedance sequence. According to the physical laws of the equivalent circuit of specific biological tissues, the physiological parameter feature extraction module extracts time-domain features to construct a state vector. The state vector calculation formula sets the dimensionless variance offset component equal to... Divide by The dimensionless energy fluctuation component is equal to Divide by ,variable and These represent the actual variances of the extracted impedance data within the comparison window and the reference window, respectively. and These represent the integral energy amplitudes of the impedance sequences within the comparison window and the reference window, respectively. Based on the fundamental fact of nerve fatigue compensation in specific tissues, when the tissue state approaches its bearing limit, the incremental response to the continuously detected signal exceeds the peak value and exhibits physical convergence decay characteristics. The inertial calibration unit at each node within the system continuously monitors the state vector. When the value of the first derivative crosses zero and shows a downward trend, it is determined that the corresponding target point has entered the physiological inhibition cycle. The dual-layer preset sliding time window includes a baseline window and a comparison window.

[0047] The node inertial calibration unit extracts the mean impedance of candidate intervention sites within a 15-minute reference window as the dynamic zero point. The node inertial calibration unit calculates the absolute difference sequence between the real-time physiological impedance data stream and the dynamic zero point within a 5-minute comparison window, extracting the root mean square value of the absolute difference sequence as the historical response deviation. The node inertial calibration unit compares the historical response deviation with the physiological response threshold identified by the physiological parameter feature extraction module to determine the inertial penalty term characterizing the physiological inhibition cycle. The topology edge weight adjustment unit calculates the inverse variance of the historical response deviation of adjacent candidate intervention sites in the meridian topology directed graph, generating a conduction edge weight distribution. The topology edge weight adjustment unit fuses the inertial penalty term with the conduction edge weight distribution to construct the target cost function, as follows: Where E is the parameter for the global topological potential minimum. Let be the transmission edge weight parameter between point i and point j in the directed graph of meridian topology. The spatial distribution distance between each candidate intervention site is measured in millimeters, and λ is a preset adjustment factor. The inertial penalty term representing the state of the corresponding site is calculated for the node inertial calibration unit. The decision scheme solution module sets the initial iteration convergence step size to 0.05. The decision scheme solution module updates the propagation edge weight distribution along the negative gradient direction of the target cost function and calculates the difference in the global topological potential energy minimum parameter between two adjacent iterations. The calculation stops when the difference in the global topological potential energy minimum parameter is lower than 0.001, and the target intervention site set is determined. After the iteration stops and the potential energy minimum is output, the decision scheme solution module records the local potential energy contribution component occupied by each node in the target cost function at this time. The system performs hard threshold classification processing on the continuously distributed local potential energy contribution component according to the built-in capacity truncation threshold, removes the failed nodes whose potential energy contribution components exceed the limit, and extracts the nodes retained within the safety threshold as candidate spatial coordinates. Thus, the continuous energy network optimization result is reduced in dimension and transformed into a discrete array composed of effective node coordinates, completing the construction of the target intervention site set.

[0048] Based on the physical attenuation characteristics of energy diffusion in the meridian topology network, the conduction loss rate of intervention energy between network nodes is positively correlated with the topological connectivity distance. Controlled by the local target point capacity limitation, and according to the physical principles of energy distribution, the output interface unit obtains the target intervention point set and combines it with node coordinate data to calculate the intervention intensity weight parameter for each independent point. The formula for calculating the intervention intensity is: Target parameter Represents the dimensionless absolute intensity weight of the output to the current position i; basic variable The parameter E represents the cumulative spatial distance from point i to each connected node, and the parameter E represents the global topological potential energy value output when the iteration stops solving; the main variable is... The system receives the dimensionless inertial penalty term from the node inertial calibration unit output, and limits the upper limit of energy allocation for high-fatigue nodes based on the exponential decay characteristics. The constant coefficient α is set to 0.85 to normalize the total output intervention energy. In actual engineering configurations, this constant coefficient 0.85 is not used for rigorous algebraic proportional normalization in mathematical theory, but rather as a physical limiting protection factor designed based on the full-scale output of the peripheral execution hardware. Because the ratio of the accumulated distance value to the minimum value has a risk of divergence under extreme calculation conditions, the system uses this empirical decay multiplier of 0.85 to normalize the energy distribution of all independent points. The absolute intensity weight of the dimension is compressed to within the safe output range of the back-end digital-to-analog converter (i.e., 85% of the total load), thereby achieving normalized control of output drive to prevent overload of external equipment at the engineering application level; the dual-layer preset sliding time window and dynamic baseline tracking logic separate nonlinear baseline drift and physiological state evolution trends; the inertial penalty term output by the node inertial calibration unit reflects the physiological fatigue of the tissue; the decision scheme solution module outputs the target intervention point set by relying on gradient calculation; the system quantifies the physiological inhibition cycle and state inertia, suppresses physical measurement error disturbances, and maintains the stability of dynamic optimization of meridian topological energy.

[0049] Example 5: When facing the on-site deployment of chronic neurological function assessment auxiliary analysis, the graph theory topology construction module initiates the on-site deployment pre-calibration procedure before accessing the real-time physiological impedance data stream. The processing system drives the front-end acquisition device to inject discrete detection signals with constant amplitude into each candidate intervention site, and simultaneously measures the reference voltage response and static phase deflection parameter of adjacent candidate intervention sites under the discrete detection signal. The graph theory topology construction module calculates the static transconductance value between nodes based on the reference voltage response and static phase deflection parameter. In the specific implementation of the calculation, the front-end acquisition device is configured with a digital cross matrix that supports multi-channel signal forward and reverse switching. The system instructs the matrix to set candidate site A as the injection end and site B as the return end. The system simultaneously records the impedance magnitude and reference voltage under the forward conduction path. The system immediately reverses its polarity, setting point B as the injection end and point A as the return end, and obtains the impedance magnitude under the reverse conduction path. The graph theory topology construction module extracts the difference ratio of the impedance magnitude obtained from the two reverse measurements to establish the directional asymmetry coefficient. Finally, the coefficient is multiplied and fitted with the reference voltage response to obtain the static transconductance value with a clear spatial orientation attribute. The graph theory topology construction module sets the static transconductance value as the initial biological connectivity attribute of each candidate intervention point in the human meridian model. Based on this, the graph theory topology construction module constructs a customized connection matrix and writes it into the underlying reference database to establish the physical scale of the conduction edge weight distribution.

[0050] The physiological parameter feature extraction module acquires real-time physiological impedance data streams to generate state vectors. The topology edge weight adjustment unit extracts customized connection matrices from the underlying benchmark database. The topology edge weight adjustment unit fuses the state vectors and inertia penalty terms to update the conduction edge weight distribution. The topology edge weight adjustment unit substitutes the conduction edge weight distribution into the target cost function. The topological energy optimization process is carried out, with the physical meaning and dimensional properties of each parameter following the previous settings. The decision-making module uses the gradient descent algorithm to calculate the energy minimum and outputs the target intervention site set. The processing system establishes the benchmark of topological connectivity properties based on the on-site deployment pre-calibration procedure, converts the connectivity deviation caused by the differences in the subject's epidermal microenvironment into a definite initial edge weight parameter, and outputs auxiliary decision-making suggestion data that is adapted to the real physiological pathways of the target physical entity.

[0051] Example 6: When facing the situation of system deployment across individuals and multi-dimensional node feature analysis, the decision-making solution module establishes a calculation benchmark through standardized model construction and pre-parameter calibration procedures before processing the real-time physiological impedance data stream transmitted by the body surface acquisition network. The decision-making solution module constructs a spectrum graph convolutional network architecture using the built-in potential energy convergence algorithm based on graph neural networks. It generates the initial feature matrix of graph nodes using the authorized in vitro desensitization impedance benchmark dataset. The initial feature matrix, together with the established customized connection matrix, is input into the spectrum graph convolutional network. A local spectral filter is constructed using Chebyshev polynomial approximation kernel to extract the spatial local features of the meridian topology directed graph. A linear rectifier unit is used as the activation layer to calculate the nonlinear response of each candidate intervention site, thereby outputting a feature mapping vector that characterizes the hidden topological information of the candidate intervention site.

[0052] The physiological parameter feature extraction module completes the quantitative calibration procedure for the physiological response threshold. Using a front-end acquisition device, a swept-frequency detection signal with a frequency linearly increasing from 10kHz to 100kHz is injected into the ex vivo skin tissue sample, while simultaneously measuring the baseline impedance time series of each candidate intervention site. The physiological parameter feature extraction module calculates the impedance root mean square error (RMSE) of the baseline impedance time series on the time axis as the environmental background fluctuation baseline. It extracts the abrupt change point where the impedance RMS first exceeds three times the limit of the environmental background fluctuation baseline, and sets the absolute amplitude of the transient impedance corresponding to this abrupt change point as the physiological response threshold of the candidate intervention site. The decision-making module combines the generated feature mapping vector with the calibrated physiological response threshold to calculate the target cost function. Where E is the parameter for the global topological potential minimum. Let be the transmission edge weight parameter between point i and point j in the directed graph of meridian topology. λ represents the spatial distribution distance between each candidate intervention site, and λ is a preset adjustment factor. The inertial penalty term, which characterizes the state of the corresponding site and is determined by the node inertial calibration unit, is used to output auxiliary decision-making suggestions in the format of quantified risk parameters. In the process of calibrating the physiological response threshold using ex vivo skin tissue samples, the samples used have been pre-placed in isotonic artificial tissue fluid to maintain the physical morphology of the cell membrane bilayer. The frequency sweep detection signal mainly acts on the equivalent capacitance structure of the epidermal cell membrane. The transient impedance corresponding to the abrupt change essentially characterizes the passive dielectric breakdown critical point of the insulating lipid bilayer at a specific high frequency band. Since this dielectric critical point belongs to the inherent passive electrical structure constant of human tissue, its physical occurrence mechanism is independent of the in vivo blood flow and the active compensation system of neural electrical activity. Therefore, the system calibrates it as the absolute structural lower limit of the physiological response, aiming to provide a stable physical reference zero point that is not affected by subjective neural fatigue for the subsequent calculation of in vivo dynamic impedance.

[0053] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An information processing system for auxiliary decision-making in acupuncture clinical practice, characterized in that, include: The physiological parameter feature extraction module is used to acquire the physiological impedance data stream of the candidate intervention sites, calculate the response statistical features of the physiological impedance data stream in the time domain dimension, and generate a state vector characterizing the physiological response saturation of the candidate intervention sites. The graph theory topology construction module is used to establish a connection matrix based on the spatial distribution relationship between each candidate intervention site, and to construct a directed meridian topology graph in combination with state vectors; The decision-making module includes a node inertia calibration unit and a topology edge weight adjustment unit. The node inertia calibration unit is used to calculate the inertia penalty term representing the physiological inhibition cycle based on the historical response deviation of each candidate intervention site within a preset sliding time window. The topology edge weight adjustment unit is used to perform a fusion operation on the inertia penalty term and the transmission edge weight distribution in the meridian topology directed graph to construct the target cost function. By solving the energy minimum value of the target cost function, the target intervention site set is determined. The feedback correction module is used to obtain the response recovery slope after intervention on the target intervention site set, and correct the convergence step size of the target cost function accordingly. The response recovery slope is the derivative of the physiological impedance after intervention with time. The topology edge weight adjustment unit follows the following quantization rules when constructing the target cost function: Where E is the global topological potential minimum. Let i be the weight of the conduction edge between point i and point j in the directed graph of meridian topology. λ represents the spatial distribution distance between each candidate intervention site, and λ is a preset adjustment factor. This is the inertial penalty term determined by the nodal inertial calibration unit.

2. The information processing system for clinical decision support in acupuncture according to claim 1, characterized in that, During the generation of the state vector, the physiological parameter feature extraction module identifies the physiological response threshold of candidate intervention sites by analyzing the local variance and cumulative energy changes of the physiological impedance data stream. When the physiological response threshold decays, the physiological parameter feature extraction module increases the penalty weight component in the state vector.

3. The information processing system for auxiliary decision-making in acupuncture clinical practice according to claim 1, characterized in that, When calculating the inertial penalty term, the node inertial calibration unit extracts the first derivative of the state vector of the candidate intervention site within a preset sliding time window to determine the physiological state evolution trend of the candidate intervention site. When the first derivative of the state vector shows a decreasing trend, the node inertial calibration unit increases the inertial penalty term to reduce the distribution weight of the candidate intervention sites in the target intervention site set.

4. The information processing system for auxiliary decision-making in acupuncture clinical practice according to claim 1, characterized in that, The graph theory topology construction module includes a topological adjacency matrix construction unit. Based on the biological connectivity attributes of candidate intervention sites in the human meridian model, the topological adjacency matrix construction unit establishes the initial connectivity path of the meridian topological directed graph and dynamically corrects the conduction threshold of the initial connectivity path according to the state vector.

5. The information processing system for clinical decision support in acupuncture according to claim 1, characterized in that, The decision-making module determines the target intervention site set through the following steps: Step S1, extract the change characteristics of the state vector of each node in the meridian topology directed graph within a preset sliding time window; Step S11, determine whether each node is in the physiological response sensitive area based on the change characteristics; Step S12, perform energy conduction inhibition processing on nodes outside the physiological response sensitive area to guide the intervention energy to diffuse to the peripheral healthy connected nodes.

6. The information processing system for auxiliary decision-making in acupuncture clinical practice according to claim 1, characterized in that, The feedback correction module includes a residual analysis unit; the residual analysis unit calculates the logical residual between the actual physiological impedance change value after intervention and the predicted change value output by the decision scheme solution module; The feedback correction module adjusts the parameter distribution in the topology edge weight adjustment unit in reverse according to the magnitude of the logical residual.

7. The information processing system for auxiliary decision-making in acupuncture clinical practice according to claim 1, characterized in that, The physiological signals obtained by the physiological parameter feature extraction module also include infrared thermal radiation distribution feature data characterizing the local tissue metabolic level; The physiological parameter feature extraction module maps infrared thermal radiation distribution feature data into auxiliary correction components of the state vector to adjust the characterization accuracy of the state vector's response capability to candidate intervention sites.

8. The information processing system for clinical decision support in acupuncture according to claim 1, characterized in that, The decision-making module utilizes a potential energy convergence algorithm based on graph neural networks to extract topological features of the directed graph of meridian topology through a multilayer sensing network, and uses gradient descent to solve for the global minimum of the target cost function, generating a set of target intervention sites with physical topological constraints.

9. The information processing system for clinical decision support in acupuncture according to claim 1, characterized in that, The system also includes an output interface unit; The output interface unit connects to the decision scheme solution module, obtains the target intervention site set, and translates the target intervention site set into auxiliary decision suggestion data output representing the intervention location coordinates and intervention intensity weights.

Images