A method for constructing a whole-cycle complication risk atlas for patients with chronic diseases
By constructing a dynamic hypergraph of physiological indicators and using pharmacokinetic parameters to remove intervention perturbations, the problem of overlapping physiological perturbations and pathological degeneration trajectories caused by medical interventions was solved. This enabled the accurate construction of a full-cycle complication risk map and reflection of pathological trends, thereby improving the accuracy of risk assessment and predictive ability for patients with chronic diseases.
Patent Information
- Application Number
- CN202610787942.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-06-03
- Publication Date
- 2026-07-03
AI Technical Summary
When constructing complication risk maps, existing healthcare information systems suffer from overlapping physiological disturbances caused by medical interventions with the pathological degeneration trajectories of organ networks. This leads to false improvement signals output by the risk assessment system, which fails to accurately reflect the pathological state of patients with chronic diseases and affects the reliability of medical prediction system decisions.
By acquiring multimodal vital sign monitoring sequences and diagnosis and treatment event sequences of patients with chronic diseases, the regulation coefficient is calculated using pharmacokinetic parameters, a dynamic hypergraph of physiological indicators is constructed, medical intervention disturbances are removed, the pathological evolution trend in the complication risk map is decoupled, and a full-cycle complication risk map is generated using topological deformation parameters.
It achieves accurate reflection of the structured degradation trend of the patient's multi-system metabolic network under external intervention perturbation, eliminates false improvement signals, improves the stability and predictive accuracy of the risk assessment model, and can detect the deterioration of the disease in the multi-organ coupling network in advance.
Smart Images
Figure CN122334431A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to a method for constructing a risk map of complications throughout the entire life cycle for patients with chronic diseases, belonging to the field of healthcare informatics technology. Background Technology
[0002] Current healthcare information systems typically collect longitudinal physiological monitoring data from patients when constructing complication risk maps, and use time-series analysis models to reflect the evolution of the physiological state of organ systems. Under ideal conditions, this approach can capture the trend changes of physiological indicators.
[0003] In the context of real-world chronic disease management, patients are typically in a continuous and irregular cycle of medical intervention. Existing data analysis logic attributes fluctuations in physiological indicators to the natural evolution of pathological states. This leads to an overlap between the instantaneous physiological disturbances caused by medical interventions and the intrinsic pathological degeneration trajectory of organ networks in the time-series signals. Since the indicator pullbacks generated by interventions often induce the risk assessment system to output false improvement signals, they mask the true degenerative trend of the underlying pathological topology. Increasing data collection density or simply optimizing data alignment algorithms cannot decouple external intervention noise at the physical mechanism level. These problems cause the output results of the risk map to deviate from the patient's true pathological state, limiting the decision-making reliability of medical prediction systems in complex clinical situations.
[0004] Therefore, the technical problem to be solved by this invention is how to construct a risk map of complications throughout the entire life cycle for patients with chronic diseases by stripping medical intervention disturbances from a dynamic physiological feature network and restoring the intrinsic pathological topological evolution trend of multiple organs. Summary of the Invention
[0005] To address the problems in the background art, the technical solution of the present invention is as follows: A method for constructing a risk map of complications throughout the entire life cycle for patients with chronic diseases, comprising the following steps: Step S101: Obtain the multimodal vital sign monitoring sequence and treatment event sequence of the patient with chronic disease throughout the entire cycle. The multimodal vital sign monitoring sequence includes continuous vital sign monitoring data and non-uniform biochemical follow-up data. The treatment event sequence records discrete medication instructions and treatment dosages. Step S102: Match the clinical pharmacokinetic parameters corresponding to each diagnosis and treatment event in the diagnosis and treatment event sequence, and calculate and generate a regulation coefficient that monotonically decreases with time based on the pharmacokinetic parameters. The regulation coefficient is used to quantify the instantaneous contribution of each diagnosis and treatment event to the fluctuation of physiological indicators at different times. Step S103: Construct a dynamic hypergraph of physiological indicators reflecting the state of multi-organ association using multimodal vital sign monitoring sequences. The dynamic hypergraph of physiological indicators uses physiological indicators as nodes and represents the nonlinear coupling relationship between different physiological indicators through hyperedges. Step S104: The adjustment coefficient is introduced as a correction weight into the local weight calculation path of the dynamic hypergraph of physiological indicators. The adjustment coefficient is used to offset the deviation of physiological data caused by diagnosis and treatment events, so as to decouple and extract the atlas evolution trajectory representing the baseline trend of pathological evolution from the dynamic hypergraph of physiological indicators. Step S105: Calculate the topological deformation parameters of the atlas evolution trajectory in the pathological feature space relative to the preset standard template, and output the full-cycle complication risk atlas using the topological distribution morphology generated by the topological deformation parameters.
[0006] Preferably, step S101 further includes: step S1011, using a resampling operator to asynchronously align continuous vital sign monitoring data and non-uniform biochemical follow-up data in time series, and constructing a physiological state observation matrix under a unified time reference; step S1012, identifying data missing points in the observation matrix, and extracting the mean, fluctuation range, and trend slope of physiological parameters within a preset window prior to the data missing points to provide prior distribution parameters for the topological reconstruction of the dynamic hypergraph of subsequent physiological indicators.
[0007] Preferably, the process of constructing the dynamic hypergraph of physiological indicators in step S103 is as follows: Step S1031, calculate the dynamic cross-correlation coefficient between physiological indicators in the multimodal vital sign monitoring sequence, and determine the initial connection weight between physiological indicator nodes; Step S1032, establish a mapping table between physiological indicators and pathological mechanisms of complications, and cluster physiological indicator nodes with common pathological significance to the same hyperedge according to the mapping table, so as to cover the multi-dimensional physiological function interaction state.
[0008] Preferably, the rule for calculating the adjustment coefficient in step S102 is as follows: ,in, For adjustment coefficients, This is the preset half-life constant for the corresponding drug in the diagnosis and treatment event. This represents the time difference between the current sampling time and the time the diagnostic event occurred. These are the preset sensitivity correction parameters.
[0009] Preferably, step S104 further includes: step S1041, obtaining the real-time adjacency matrix of the dynamic hypergraph of physiological indicators at the sampling time; step S1042, scaling the weights of nodes affected by diagnosis and treatment events in the real-time adjacency matrix using adjustment coefficients to correct the computational contribution of physiological indicator mutations caused by drug intervention to the evolution of the global pathological state, thereby achieving logical separation of intervention noise and pathological trends.
[0010] Preferably, step S104 further includes: step S1043, determining whether there are missing non-uniform biochemical follow-up data in the current sampling window; step S1044, if it is determined that there are missing data, extracting the spectral evolution trajectory in the preceding adjacent effective window as the local structural basis, and combining it with the fluctuation range of continuous vital sign monitoring data to complete the bias compensation, so as to maintain the topological continuity of the dynamic hypergraph of physiological indicators.
[0011] Preferably, the process of calculating the topological deformation parameter in step S105 is as follows: Step S1051, calculate the Laplace eigenvalue distribution of the dynamic hypergraph of physiological indicators on a continuous time series; Step S1052, quantify the topological contraction rate of the dynamic hypergraph of physiological indicators by comparing the offset of the Laplace eigenvalue distribution at adjacent time points, and determine the topological contraction rate as the topological deformation parameter to reflect the deterioration acceleration of the pathological state.
[0012] Preferably, after outputting the full-cycle complication risk map in step S105, the method further includes: step S106, calculating the fluctuation entropy value of the risk signal output by the full-cycle complication risk map within a preset time period; step S107, if the fluctuation entropy value exceeds a preset stability threshold, automatically increasing the correction intensity for the fluctuation of physiological indicators in step S104 until the risk signal recovers to a smooth convergence state.
[0013] Preferably, step S102 further includes collaborative processing of medical intervention events: step S1021, identifying whether there are multiple drugs intervening concurrently in the diagnosis and treatment event sequence; step S1022, if there are multiple drugs intervening concurrently, then the adjustment coefficients corresponding to each drug are weighted and superimposed according to the preset drug interaction coefficients to generate a comprehensive adjustment weight and feed it back to the weight calculation path.
[0014] Preferably, the method further includes a model calibration step: Step S108, obtaining the patient's biochemical follow-up endpoint label, and verifying the deviation between the biochemical follow-up endpoint label and the prediction results of the full-cycle complication risk profile; Step S109, correcting the sensitivity correction parameter in the formula based on the deviation value generated by the verification. This is to achieve dynamic closed-loop optimization of the complication prediction logic.
[0015] Compared with the prior art, the beneficial effects of the present invention are: 1. In the construction of the full-cycle complication risk map, the logic decoupling between external medical intervention disturbances and the intrinsic pathological degeneration trajectory of human organs is achieved. By translating the medical intervention event sequence in the electronic medical record into a topological damping factor with pharmacokinetic characteristics, and injecting it as a multiplicative constraint into the weight calculation process of the dynamic hypergraph of physiological features, the algorithm isolates the instantaneous large fluctuations of physiological indicators caused by drug dosage adjustment or treatment intervention at the algorithm level. This eliminates the phenomenon of false improvement or noise false alarm in risk assessment caused by drug effects, which is common in clinical scenarios. It ensures that the output complication risk map can objectively reflect the most real structured degeneration trend of the patient's multi-system metabolic network.
[0016] 2. To address the technical bottleneck of highly sparse and asynchronous clinical follow-up data in the time dimension, the adjacency matrix of the preceding effective time window is extracted as the structural basis within the time window where low-frequency biochemical indicators are missing, and the bias compensation amount is calculated in combination with the real-time fluctuation difference of high-frequency vital signs nodes. Thus, without changing the global topological connectivity of the hypergraph, the local weights are dynamically corrected, maintaining the continuous convergence output of the risk assessment model during the data gap period and improving the system's operational stability under non-ideal sampling conditions.
[0017] 3. The monitoring of chronic disease risk is upgraded from linear numerical tracking of a single indicator to deformation monitoring of a multidimensional topological space. By constructing a dynamic hypergraph structure that associates multimodal physiological indicators, the nonlinear coupling relationship between each physiological node is quantified. The Riemannian distance shortening rate between the current risk topological state vector and the standard complication topological attractor template is used to characterize the acceleration of disease deterioration. This allows for a longer lead time for complication intervention by capturing the microscopic collapse of the underlying connection relationship of the multi-organ coupling network before patients show obvious clinical deterioration. Attached Figure Description
[0018] Figure 1 This is a flowchart of the method for constructing a risk map of complications throughout the entire life cycle for patients with chronic diseases according to the present invention; Figure 2 This is a state evolution diagram of the process of constructing the full-cycle complication risk map of the present invention.
[0019] The objectives, features, and advantages of this invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation
[0020] The technical solutions of the embodiments of this application will be clearly described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of this application are within the scope of protection of this application.
[0021] A method for constructing a risk map of complications throughout the entire life cycle for patients with chronic diseases includes the following steps: Step S101: Obtain the multimodal vital sign monitoring sequence and treatment event sequence of the patient with chronic disease throughout the entire cycle. The multimodal vital sign monitoring sequence includes continuous vital sign monitoring data and non-uniform biochemical follow-up data. The treatment event sequence records discrete medication instructions and treatment dosages. Step S102: Match the clinical pharmacokinetic parameters corresponding to each diagnosis and treatment event in the diagnosis and treatment event sequence, and calculate and generate a regulation coefficient that monotonically decreases with time based on the pharmacokinetic parameters. The regulation coefficient is used to quantify the instantaneous contribution of each diagnosis and treatment event to the fluctuation of physiological indicators at different times. Step S103: Construct a dynamic hypergraph of physiological indicators reflecting the state of multi-organ association using multimodal vital sign monitoring sequences. The dynamic hypergraph of physiological indicators uses physiological indicators as nodes and represents the nonlinear coupling relationship between different physiological indicators through hyperedges. Step S104: The adjustment coefficient is introduced as a correction weight into the local weight calculation path of the dynamic hypergraph of physiological indicators. The adjustment coefficient is used to offset the deviation of physiological data caused by diagnosis and treatment events, so as to decouple and extract the atlas evolution trajectory representing the baseline trend of pathological evolution from the dynamic hypergraph of physiological indicators. Step S105: Calculate the topological deformation parameters of the atlas evolution trajectory in the pathological feature space relative to the preset standard template, and output the full-cycle complication risk atlas using the topological distribution morphology generated by the topological deformation parameters.
[0022] Preferably, step S101 further includes: step S1011, using a resampling operator to asynchronously align continuous vital sign monitoring data and non-uniform biochemical follow-up data in time series, and constructing a physiological state observation matrix under a unified time reference; step S1012, identifying data missing points in the observation matrix, and extracting the mean, fluctuation range, and trend slope of physiological parameters within a preset window prior to the data missing points to provide prior distribution parameters for the topological reconstruction of the dynamic hypergraph of subsequent physiological indicators.
[0023] Preferably, the process of constructing the dynamic hypergraph of physiological indicators in step S103 is as follows: Step S1031, calculate the dynamic cross-correlation coefficient between physiological indicators in the multimodal vital sign monitoring sequence, and determine the initial connection weight between physiological indicator nodes; Step S1032, establish a mapping table between physiological indicators and pathological mechanisms of complications, and cluster physiological indicator nodes with common pathological significance to the same hyperedge according to the mapping table, so as to cover the multi-dimensional physiological function interaction state.
[0024] Preferably, the rule for calculating the adjustment coefficient in step S102 is as follows: ,in, For adjustment coefficients, This is the preset half-life constant for the corresponding drug in the diagnosis and treatment event. This represents the time difference between the current sampling time and the time the diagnostic event occurred. These are the preset sensitivity correction parameters.
[0025] Preferably, step S104 further includes: step S1041, obtaining the real-time adjacency matrix of the dynamic hypergraph of physiological indicators at the sampling time; step S1042, scaling the weights of nodes affected by diagnosis and treatment events in the real-time adjacency matrix using adjustment coefficients to correct the computational contribution of physiological indicator mutations caused by drug intervention to the evolution of the global pathological state, thereby achieving logical separation of intervention noise and pathological trends.
[0026] Preferably, step S104 further includes: step S1043, determining whether there are missing non-uniform biochemical follow-up data in the current sampling window; step S1044, if it is determined that there are missing data, extracting the spectral evolution trajectory in the preceding adjacent effective window as the local structural basis, and combining it with the fluctuation range of continuous vital sign monitoring data to complete the bias compensation, so as to maintain the topological continuity of the dynamic hypergraph of physiological indicators.
[0027] Preferably, the process of calculating the topological deformation parameter in step S105 is as follows: Step S1051, calculate the Laplace eigenvalue distribution of the dynamic hypergraph of physiological indicators on a continuous time series; Step S1052, quantify the topological contraction rate of the dynamic hypergraph of physiological indicators by comparing the offset of the Laplace eigenvalue distribution at adjacent time points, and determine the topological contraction rate as the topological deformation parameter to reflect the deterioration acceleration of the pathological state.
[0028] Preferably, after outputting the full-cycle complication risk map in step S105, the method further includes: step S106, calculating the fluctuation entropy value of the risk signal output by the full-cycle complication risk map within a preset time period; step S107, if the fluctuation entropy value exceeds a preset stability threshold, automatically increasing the correction intensity for the fluctuation of physiological indicators in step S104 until the risk signal recovers to a smooth convergence state.
[0029] Preferably, step S102 further includes collaborative processing of medical intervention events: step S1021, identifying whether there are multiple drugs intervening concurrently in the diagnosis and treatment event sequence; step S1022, if there are multiple drugs intervening concurrently, then the adjustment coefficients corresponding to each drug are weighted and superimposed according to the preset drug interaction coefficients to generate a comprehensive adjustment weight and feed it back to the weight calculation path.
[0030] Preferably, the method further includes a model calibration step: Step S108, obtaining the patient's biochemical follow-up endpoint label, and verifying the deviation between the biochemical follow-up endpoint label and the prediction results of the full-cycle complication risk profile; Step S109, correcting the sensitivity correction parameter in the formula based on the deviation value generated by the verification. This is to achieve dynamic closed-loop optimization of the complication prediction logic.
[0031] Example 1: In a long-term management environment for diabetes mellitus with chronic kidney disease (CKD) with multimodal vital sign monitoring and regular biochemical follow-up records, the system faces the challenge of the transient physiological index fluctuations caused by high-frequency medication interventions masking the intrinsic pathological degeneration trajectory of the multi-organ network. For CKD patients with risk of complications, the processing unit acquires their full-cycle multimodal vital sign monitoring sequence and treatment event sequence. The multimodal vital sign monitoring sequence includes continuous vital sign monitoring data collected by continuous blood glucose monitoring devices and non-uniform biochemical follow-up data generated by non-uniform biochemical follow-up. The treatment event sequence records discrete medication instructions and treatment doses. By using a resampling operator to asynchronously align the continuous vital sign monitoring data and non-uniform biochemical follow-up data in time series, a physiological state observation matrix is constructed under a unified time reference.
[0032] To correct for disturbances caused by external medical interventions, the system matches the clinical pharmacokinetic parameters corresponding to each diagnostic event in the sequence of diagnostic events, and calculates and generates a regulation coefficient that monotonically decreases over time based on the pharmacokinetic parameters. This adjustment coefficient is used to quantify the instantaneous contribution of each diagnostic and treatment event to the fluctuation of physiological indicators at different times. The specific calculation formula is as follows: ,in, For adjustment coefficients, This is the preset half-life constant for the corresponding drug in the diagnosis and treatment event. This represents the time difference between the current sampling time and the time the diagnostic event occurred. The sensitivity correction parameter k is a preset parameter. To address the prevalent nonlinear hysteresis effect between pharmacokinetic concentrations in the blood and the actual physiological responses of pharmacodynamics in the organ network, the sensitivity correction parameter k in the above formula is configured as a nonlinear conversion medium carrying the receptor saturation mechanism. The processing unit uses drug concentration as an intermediate transfer quantity and establishes a pharmacodynamic saturation feedback mapping using the classical Hill equation. This ensures that within the onset window when the blood drug concentration is higher than the receptor saturation threshold, the value of k is suppressed, thus maintaining the damping effect of the regulation coefficient D in a high-level shift state. When the time difference... When the drug concentration is prolonged to the linear elimination region, the actual offsetting effect of the drug effect on physiological fluctuations follows the exponential decline shown by the calculation formula, thus truly restoring the physical degradation process of metabolic elimination in vivo on intervention disturbances. A dynamic hypergraph of physiological indicators reflecting the state of multi-organ correlation is constructed using multimodal vital sign monitoring sequences. This dynamic hypergraph of physiological indicators uses physiological indicators as nodes and uses hyperedges to represent the nonlinear coupling relationship between different physiological indicators. The adjustment coefficient is introduced as a correction weight into the real-time adjacency matrix of the dynamic hypergraph of physiological indicators at the sampling time. The adjustment coefficient is used to scale the weight of nodes affected by diagnosis and treatment events in the real-time adjacency matrix to correct the calculation contribution of physiological indicator mutations caused by drug intervention to the evolution of the global pathological state. In this way, medical intervention noise is removed at the data level and the intrinsic pathological topological evolution trend of multi-organs is restored.
[0033] The Laplace eigenvalue distribution of the dynamic hypergraph of physiological indicators on a continuous time series is calculated. The topological contraction rate of the dynamic hypergraph is quantified by comparing the offset of the Laplace eigenvalue distribution between adjacent time points. This topological contraction rate is determined as a topological deformation parameter reflecting the acceleration of pathological deterioration. The generated topological distribution morphology is then used to output a full-cycle complication risk map, thereby eliminating spurious risk convergence signals caused by drug effects in clinical scenarios. This ensures that the output objectively reflects the structural degradation trend of the metabolic network. During this process, to ensure that the map output is not distorted by extreme drug doses, the processing unit also implements... The system collects complication risk signals from the most recent ten sampling periods and calculates the fluctuation entropy value of the sequence using the information entropy algorithm. When the fluctuation entropy value exceeds the preset stability threshold of 0.2, the system initiates closed-loop control, progressively increasing the correction weight coefficient of the drug-receiving node in the adjacency matrix with an absolute mathematical step size of 0.05. Simultaneously, after each step adjustment, the system calculates the first-order differential slope of the risk spectrum evolution trajectory until the absolute value of the slope falls within the range of less than 0.01 for five consecutive sampling periods. At this point, the system can determine that the risk signal has recovered to a smooth convergence state and latch the current correction intensity.
[0034] Example 2: In a test environment integrating a multi-source heterogeneous electronic medical record simulator and a physiological dynamics simulation module, the system faces the challenge of evaluating whether the transient numerical decline caused by frequent medication interventions masks fluctuations in pathological risk. The experimental platform uses anonymized historical datasets of clinical chronic disease patients as the initial input stream. The computational accuracy of the experimental platform is set to 0.001, and the sampling frequency is set to 1Hz to generate the output characteristics of the physiological sensors; the sampling period... The settings are determined by balancing the real-time performance of data acquisition with the computational load of the processing unit. When the coefficient of variation of the monitored signal is greater than the preset variation threshold of 0.15, in order to ensure that the sampling meets the signal reconstruction requirements, the sampling period tends to the lower limit of its value range, which is set to 5 minutes. In order to simulate the sensor noise in the clinical environment, Gaussian white noise with a signal-to-noise ratio of 20dB is superimposed in the experimental signal source, and power frequency interference harmonics with a frequency of 50Hz are introduced.
[0035] The experiment was divided into a control group and the experimental group of the present invention. The control group used a moving average filtering algorithm to process physiological indicators and did not introduce pharmacokinetic parameters for intervention correction, while the experimental group of the present invention used an adjustment coefficient. The real-time adjacency matrix of the dynamic hypergraph of physiological indicators is weighted and scaled. During the baseline setting phase, a raw blood glucose monitoring sequence containing medication intervention is input, with an initial blood glucose value of 12.5 mmol / L. After 120 minutes of hypoglycemic drug intervention, the raw blood glucose value drops to 6.8 mmol / L. At this point, the system matches the preset half-life constant of the drug. The time difference between the sampling time and the medication time is 4 hours. Under the condition of 2h, according to the formula Determine the adjustment coefficient ,in, For adjustment coefficients, To adjust the sensitivity parameters, To preset the half-life constant, This represents the time difference between the current sampling time and the time of the diagnosis / treatment event. In the control group, due to the lack of intervention correction, the risk index identified by the system decreased from 0.76 before medication to 0.42, generating an inaccurate risk convergence signal. However, in the sample group of this invention, the adjustment coefficient was used... Weight compensation was applied to the dynamic hypergraph of physiological indicators, and the calculated topological deformation parameters fluctuated between 0.72 and 0.74. This value reflects the persistence of the structural degeneration of the metabolic network and does not indicate improvement due to the decline in indicator values caused by drugs.
[0036] To verify the rationality of the parameter boundaries, the sensitivity correction parameter was examined in the experiment. The effect of the value of on system stability is shown in experimental results. When the value is within the range of 0.8 to 1.2, the accuracy of the topological deformation parameter in identifying pathological degeneration remains at approximately 94.7%; when After exceeding the upper limit of 1.5, due to overcompensation for the intervention disturbance, the growth rate of key performance indicators slows down and tends to plateau, indicating that the system's correction gain has entered the saturation region. The value cannot improve the accuracy of pathological trend extraction; meanwhile, in the comparative experiment where the core problem variable, namely the pathological degradation rate, presents low, medium, and high gradients, the topological contraction rates output by the sample group of this invention are 0.022, 0.051, and 0.114, respectively, showing a monotonous linear correlation consistent with the severity of the pathology; by comparing the offset of the Laplace eigenvalue distribution at adjacent time points, the sample group of this invention, under the condition of including 20dB background noise, logically decouples the short-term physiological disturbances caused by external medical intervention from the irreversible degradation trajectory of the organ network, proving that the adjustment coefficient The multiplicative constraint mechanism of dynamic hypergraph weights eliminates false alarms in risk assessment caused by drug effects in clinical scenarios, enabling the full-cycle complication risk map to reflect the metabolic network health status of patients with chronic diseases.
[0037] Example 3: This example combines Figures 1 to 2 This document describes a method for constructing a risk map of complications throughout the entire lifecycle of patients with chronic diseases. Figure 1 As shown, step S101 acquires the sequence of vital signs monitoring and treatment events, step S102 matches pharmacokinetic parameters and calculates the regulation coefficient, then step S103 constructs a dynamic hypergraph of physiological indicators, and on this basis, step S104 introduces the regulation coefficient to extract the evolution trajectory of the graph, and finally step S105 calculates the topological deformation parameters and outputs the risk graph.
[0038] like Figure 2 As shown, the physiological state observation matrix includes asynchronous time series alignment. This physiological state observation matrix points to the regulation coefficient generation state, which includes pharmacokinetic parameter matching, by identifying the sequence of diagnostic and treatment events. At the same time, this physiological state observation matrix points to the dynamic hypergraph construction state, which includes nonlinear coupling mapping, by calculating the dynamic cross-correlation coefficient. The regulation coefficient generation state points to the graph evolution trajectory extraction state, which includes intervention noise logic separation, by introducing a correction weight. The dynamic hypergraph construction state points to the graph evolution trajectory extraction state by scaling the real-time adjacency matrix. The graph evolution trajectory extraction state points to the risk graph output state, which includes topological contraction rate determination, by comparing the Laplace eigenvalue distribution offset. The risk graph output state points to the graph evolution trajectory extraction state by calculating the fluctuation entropy value and increasing the correction intensity feedback, and points to the regulation coefficient generation state by inversely correcting the sensitivity correction parameter feedback.
[0039] Example 4: In a diabetic nephropathy management environment integrating a continuous blood glucose monitoring terminal and a discrete biochemical testing system, for continuous vital sign monitoring data with a sampling period of 10 seconds and non-uniform biochemical follow-up data with a sampling interval of 30 days, the processing unit acquires the multimodal vital sign monitoring sequence and diagnosis and treatment event sequence of the chronic disease patient throughout the entire cycle. The resampling operator opens an alignment window at the timestamp node of each non-uniform biochemical follow-up data, and uses linear interpolation logic to map the mean of the continuous vital sign monitoring data within the alignment window to generate a time-axis aligned physiological state observation matrix, thereby eliminating the phase deviation of heterogeneous data streams in the dynamic hypergraph construction process.
[0040] The system constructs a dynamic hypergraph of physiological indicators reflecting the interconnected states of multiple organs using a physiological state observation matrix. This dynamic hypergraph defines blood glucose, blood pressure, and serum creatinine as nodes, and uses hyperedges to represent the metabolic coupling relationships between different physiological indicators. The system matches the preset half-life constants of corresponding drugs in the sequence of diagnostic and treatment events. And based on the time difference between the current sampling time and the drug administration time Determine the adjustment coefficient The specific calculation formula is as follows: ,in, For adjustment coefficients, To adjust the sensitivity parameters, To preset the half-life constant, This is the time difference; for non-intravenous injection interventions involving oral medications and other drugs with an absorption phase, the system will adjust the coefficient. The calculation logic is switched to a one-chamber open dynamics double-exponential model, based on the formula. Calculate instantaneous weighted damping, where the variables are... Represents a dimensionless adjustment coefficient, a variable. Represents sensitivity correction parameters, variables The metabolic elimination constant, representing the rate at which a drug is cleared from the body, is a variable. The absorption rate constant, representing the rate at which a drug enters the bloodstream, is numerically defined as follows: ,variable The time difference between the sampling point and the issuance of the dosing command is represented by the above two kinetic constants for a specific patient and a specific drug. When acquiring these two kinetic constants, the processing unit accesses the clinical pharmacokinetic baseline database via the intranet interface, retrieves and extracts the population pharmacokinetic prior data of the oral drug in standard adult trials, and interfaces it with the patient's current biochemical follow-up records. Creatinine clearance and liver enzyme activity indicators, used to characterize the true metabolic capacity, are extracted. Based on these individual biochemical characteristics, the system uses Bayesian parameter estimation to perform individualized scaling and conversion of the metabolic elimination constant and absorption rate constant of the above population, thereby obtaining individual constant values that perfectly match the current patient's specific internal environment. To avoid dimensional drift caused by the significant difference in characteristic time scales between high-frequency physical examination and sparse biochemical follow-up, the processing unit establishes a cutoff time window on the time axis equivalent to five metabolic elimination half-lives. When the time difference... When the cutoff time window is exceeded, the processing unit stops outputting the adjustment coefficient. It also forces a zeroing out, causing the corresponding hypergraph node weight calculation path to automatically fall back to the baseline of the dominant structural evolution based on long-term biochemical follow-up data, and the sensitivity correction parameter. The value is controlled by the signal-to-noise ratio of the monitoring equipment. When the signal-to-noise ratio When the interference threshold is 15dB below the threshold, the system adjusts the signal-to-noise ratio accordingly. The sensitivity correction parameter is increased based on the positive correlation with the preset interference intensity. The processing unit uses the adjustment coefficient to calculate the value. The feature weights of drug-received nodes in the real-time adjacency matrix of the dynamic hypergraph for scaling physiological indicators are adjusted by the coefficient. The eigenvalue decomposition process of the Laplace matrix is introduced to correct the fluctuations in physiological indicators caused by external drug intervention.
[0041] The system calculates the Laplace eigenvalue distribution offset of the dynamic hypergraph of physiological indicators at adjacent sampling times and determines it as a topological deformation parameter reflecting the rate of deterioration of the pathological state. When the increment of the topological deformation parameter exceeds the deterioration threshold of 0.05 in three consecutive alignment windows, the system marks the corresponding organ node as a warning state in the output full-cycle complication risk map. To achieve this logical closed-loop transformation from pure topological mathematical structure to specific clinical biological quantities, the system has a built-in entity mapping mechanism. The processing unit extracts target organ degenerative indicators confirmed by medical institutions, such as estimating the annualized decline of glomerular filtration rate over six consecutive months. The system uses ultrasound increments to measure the slope and thickness of the intima-media layer of blood vessels, and introduces a nonlinear polynomial regression algorithm to establish a numerical conversion equation between the abstract contraction change of the Laplacian eigenvalue spectral density matrix calculated based on continuous time series and the actual metabolic degeneration rate of the physical organs. This establishes an objective causal chain at the underlying level that characterizes the acceleration of clinical disease deterioration by structural deformation parameters. Before processing multimodal vital sign monitoring sequences, the system performs hash desensitization processing on the identity identifiers of patients with chronic diseases to ensure that the data flow process meets privacy and security standards, so that the output complication risk map objectively reflects the structural degeneration trend of the multi-organ network.
[0042] Example 5: In a multimodal chronic disease management platform environment deployed across institutions, when the system faces situations where there are differences in the signal noise floor of monitoring devices and inconsistent sensitivity of patient groups to drug responses, the processing unit determines the initial boundary values of the physiological state observation matrix through baseline calibration logic. This calibration process includes collecting 512 continuous vital sign monitoring sampling points in a resting state to determine the signal-to-noise ratio. And according to the calculation relationship Calibration sensitivity correction parameters The initial value of , where, To adjust the sensitivity parameters, For signal-to-noise ratio, For equipment characteristic coefficients, The system uses environmental compensation constants; it inputs the collected non-uniform biochemical follow-up data into the pharmacokinetic parameter matching module, and extracts the preset half-life constant of the corresponding drug by searching the desensitized clinical drug attribute database. Furthermore, a resampling operator is used to perform a linear mapping at the timestamp nodes of the biochemical follow-up, generating a physiological state observation matrix with a uniform sampling step size.
[0043] When the system enters a continuous operation phase and encounters a situation where drug residues and new dosing events overlap, the processing unit performs gain adjustment on the hyperedge weights of the dynamic hypergraph of physiological indicators by extracting the topological energy distribution of the real-time adjacency matrix. After identifying a new dosing instruction in the sequence of diagnostic and treatment events, the system calculates the dosing instructions according to the formula... Generate adjustment coefficient The system retrieves the node degree values of physiological indicators at the current moment, calculates the Laplace eigenvalue spectral density shift before and after the superimposed drug effect within the alignment window determined by the resampling operator, quantifies the morphological evolution path of the dynamic hypergraph of physiological indicators within the drug effect period, and determines the morphological evolution path as a topological deformation parameter reflecting the pathological degradation rate. It then outputs the topological distribution morphology reflecting the stability of the organ network metabolic structure in the full-cycle complication risk map. The processing unit executes the online calibration procedure of the risk assessment model. The system extracts the measured values of the patient's clinical biochemical indicators within the current alignment window and converts them into biochemical follow-up endpoint labels. The processing unit uses the mean squared error algorithm to calculate the gradient between the biochemical follow-up endpoint labels and the predicted quantified values output by the full-cycle complication risk map at the same observation node. The system updates the sensitivity correction parameters based on the gradient descent backpropagation algorithm. The parameter update mathematical expression is as follows: , where variables The sensitivity correction parameter represents the value updated in the current iteration cycle. The sensitivity correction parameter representing the previous calculation period, variable Represents the learning rate step size, a variable This represents the calculation of the error objective function relative to the variable based on the aforementioned deviation gradient. The first-order partial derivative is used to periodically perform supervised parameter iterative calculations to compensate for individual differences in metabolic enzyme activity or baseline drift caused by long-term aging of sensors used to collect specific vital signs.
[0044] Example 6: In an offline parameter calibration environment integrating a discrete biochemical testing system and a multi-source vital sign acquisition terminal, when the system faces initial configuration issues due to inconsistencies between the noise floor characteristics of different hardware sensors and the sensitivity of specific drug responses, the processing unit determines the equipment characteristic coefficients through a controlled experimental procedure. Environmental compensation constant The value of the parameter is specified; the processing unit extracts a historical medication response dataset containing 200 sample points, where each sample point has a known signal-to-noise ratio. Drug pre-defined half-life constant In addition, the real-time observation values of physiological indicators at the sampling time were used to construct sensitivity correction parameters by employing the least squares fitting method. With signal-to-noise ratio The regression model between them, under the functional specifications of a device sampling frequency of 1Hz and a measurement accuracy of 0.001, calculates and makes the relationship... The deviations between the calculated values and the actual responses in the historical medication response dataset tend to converge. To adjust the sensitivity parameters, For signal-to-noise ratio, This is the equipment characteristic coefficient, calibrated to 0.125. The environmental compensation constant is set to 0.452, thus providing a basis for real-time calculation of the adjustment coefficient. The above-mentioned calibrated coefficient values are based on the specific impedance-type vital sign sensing hardware used in this embodiment. They are absolute physical boundaries established through cross-validation within a typical clinical working range with a signal-to-noise ratio of 5dB to 30dB. The calibration experiment shows that when the device characteristic coefficient 'a' exceeds the upper limit of 0.125, the logarithmic gain will cause the erroneous rejection of high-frequency drug effect pulses, resulting in excessive smoothing of pathological features. When the environmental compensation constant 'b' is below the lower limit of 0.452, the model will not be able to provide sufficient initial damping to resist the baseline shift failure caused by low-frequency background thermal noise. These two parameters together define the engineering feasibility threshold for this system to effectively filter out drug effect noise without damaging the topological features of organ degeneration in complex electromagnetic interference environments.
[0045] When the system faces the task of constructing a dynamic hypergraph of physiological indicators for patients with diabetic nephropathy, the processing unit determines the composition logic of the hyperedges and the calibration benchmark of the deterioration threshold by searching the medical ontology library. The system defines a hyperedge as a set of nodes with pathological synergistic characteristics. When blood glucose monitoring values, systolic blood pressure readings, and serum creatinine concentrations simultaneously deviate from their respective preset physiological benchmark values, the system establishes a hyperedge association for the above nodes in the real-time adjacency matrix. Before establishing the above association, the processing unit preloads and maintains a mapping table, which is stored using a key-value pair data structure. The key is defined as the specific pathological mechanism category of the complication, such as diabetic nephropathy microvascular disease or hypertension target organ damage, while the corresponding value is defined as the set of underlying physiological indicators that trigger the mechanism and its quantitative mapping rules. In actual operation, when multiple physiological indicator nodes... When the data fluctuation amplitude of a point exceeds 1.2 times the pre-set medical baseline threshold within the table within a unified observation time window, and the signs of its first derivative are consistent, the system will establish a medical judgment on the pathological mechanism of the corresponding complication from the physiological indicators. This will drive the nonlinear coupling of the hyperedge weights. The processing unit will collect data within the first 7 days of the chronic disease patient's management cycle to establish the probability density function of the topological deformation parameters. The 95th percentile of the probability density function will be determined as the patient's personalized deterioration threshold. In clinical deployment, by setting this personalized deterioration threshold to 0.052 and using the incremental offset of the topological deformation parameters as the judgment index, the processing unit will quantify the pathological degradation rate and generate the corresponding topological distribution pattern in the full-cycle complication risk map. This will allow the output to reflect the structured degradation trend of the multi-organ network while eliminating drug intervention noise.
[0046] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.
[0047] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.
Claims
1. A method for constructing a risk map of complications throughout the entire life cycle for patients with chronic diseases, characterized in that, Includes the following steps: Step S101: Obtain the multimodal vital sign monitoring sequence and treatment event sequence of the patient with chronic disease throughout the entire cycle. The multimodal vital sign monitoring sequence includes continuous vital sign monitoring data and non-uniform biochemical follow-up data. The treatment event sequence records discrete medication instructions and treatment dosages. Step S102: Match the clinical pharmacokinetic parameters corresponding to each diagnosis and treatment event in the diagnosis and treatment event sequence, and calculate and generate a regulation coefficient that monotonically decreases with time based on the pharmacokinetic parameters. The regulation coefficient is used to quantify the instantaneous contribution of each diagnosis and treatment event to the fluctuation of physiological indicators at different times. Step S103: Construct a dynamic hypergraph of physiological indicators reflecting the state of multi-organ association using multimodal vital sign monitoring sequences. The dynamic hypergraph of physiological indicators uses physiological indicators as nodes and represents the nonlinear coupling relationship between different physiological indicators through hyperedges. Step S104: The adjustment coefficient is introduced as a correction weight into the local weight calculation path of the dynamic hypergraph of physiological indicators. The adjustment coefficient is used to offset the deviation of physiological data caused by diagnosis and treatment events, so as to decouple and extract the atlas evolution trajectory representing the baseline trend of pathological evolution from the dynamic hypergraph of physiological indicators. Step S105: Calculate the topological deformation parameters of the atlas evolution trajectory in the pathological feature space relative to the preset standard template, and output the full-cycle complication risk atlas using the topological distribution morphology generated by the topological deformation parameters.
2. The method for constructing a risk map of complications throughout the entire life cycle for patients with chronic diseases according to claim 1, characterized in that, Step S101 further includes: Step S1011, using a resampling operator to asynchronously align continuous vital sign monitoring data and non-uniform biochemical follow-up data in time series, and constructing a physiological state observation matrix under a unified time reference; Step S1012, identifying data missing points in the observation matrix, and extracting the mean, fluctuation range, and trend slope of physiological parameters within a pre-set window preceding the data missing points to provide prior distribution parameters for the topological reconstruction of the dynamic hypergraph of subsequent physiological indicators.
3. The method for constructing a full-cycle complication risk map for patients with chronic diseases according to claim 1, characterized in that, The process of constructing the dynamic hypergraph of physiological indicators in step S103 is as follows: Step S1031, calculate the dynamic cross-correlation coefficient between physiological indicators in the multimodal vital sign monitoring sequence and determine the initial connection weight between physiological indicator nodes; Step S1032, establish a mapping table between physiological indicators and pathological mechanisms of complications, and cluster physiological indicator nodes with common pathological significance to the same hyperedge according to the mapping table to cover the multidimensional physiological function interaction state.
4. The method for constructing a risk map of complications throughout the entire life cycle for patients with chronic diseases according to claim 1, characterized in that, The rules for calculating the adjustment coefficient in step S102 are as follows: ,in, For adjustment coefficients, This is the preset half-life constant for the corresponding drug in the diagnosis and treatment event. This represents the time difference between the current sampling time and the time the diagnostic event occurred. These are the preset sensitivity correction parameters.
5. The method for constructing a full-cycle complication risk map for patients with chronic diseases according to claim 1, characterized in that, Step S104 further includes: Step S1041, obtaining the real-time adjacency matrix of the dynamic hypergraph of physiological indicators at the sampling time; Step S1042, scaling the weights of nodes affected by diagnosis and treatment events in the real-time adjacency matrix using adjustment coefficients to correct the computational contribution of physiological indicator mutations caused by drug intervention to the evolution of the global pathological state, thereby achieving logical separation of intervention noise and pathological trends.
6. The method for constructing a risk map of complications throughout the entire life cycle for patients with chronic diseases according to claim 5, characterized in that, Step S104 further includes: Step S1043, determining whether there are missing non-uniform biochemical follow-up data in the current sampling window; Step S1044, if it is determined that there are missing data, extracting the spectral evolution trajectory in the preceding adjacent effective window as the local structural basis, and combining it with the fluctuation range of continuous vital sign monitoring data to complete the bias compensation, so as to maintain the topological continuity of the dynamic hypergraph of physiological indicators.
7. The method for constructing a full-cycle complication risk map for patients with chronic diseases according to claim 1, characterized in that, The process of calculating the topological deformation parameter in step S105 is as follows: Step S1051, calculate the Laplace eigenvalue distribution of the dynamic hypergraph of physiological indicators on the continuous time series; Step S1052, quantify the topological contraction rate of the dynamic hypergraph of physiological indicators by comparing the offset of the Laplace eigenvalue distribution at adjacent time points, and determine the topological contraction rate as the topological deformation parameter to reflect the deterioration acceleration of the pathological state.
8. The method for constructing a risk map of complications throughout the entire life cycle for patients with chronic diseases according to claim 1, characterized in that, After outputting the full-cycle complication risk map in step S105, the method further includes: step S106, calculating the fluctuation entropy value of the risk signal output by the full-cycle complication risk map within a preset time period; step S107, if the fluctuation entropy value exceeds the preset stability threshold, automatically increasing the correction intensity for the fluctuation of physiological indicators in step S104 until the risk signal recovers to a smooth convergence state.
9. The method for constructing a risk map of complications throughout the entire life cycle for patients with chronic diseases according to claim 1, characterized in that, Step S102 also includes collaborative processing of medical intervention events: Step S1021, identify whether there are multiple drugs intervening concurrently in the diagnosis and treatment event sequence; Step S1022, if there are multiple drugs intervening concurrently, then the adjustment coefficients corresponding to each drug are weighted and superimposed according to the preset drug interaction coefficients to generate a comprehensive adjustment weight and feed it back to the weight calculation path.
10. A method for constructing a full-cycle complication risk map for patients with chronic diseases according to claim 4, characterized in that, The method also includes a model calibration step: Step S108, obtaining the patient's biochemical follow-up endpoint label, and verifying the deviation between the biochemical follow-up endpoint label and the prediction results of the full-cycle complication risk profile; Step S109, correcting the sensitivity correction parameter in the formula based on the deviation value generated by the verification. This is to achieve dynamic closed-loop optimization of the complication prediction logic.