A method and system for chronic disease risk prediction

Abstract

The present application relates to the technical field of physiological signal risk prediction, in particular to a chronic disease risk prediction method and system, comprising: receiving a user physiological waveform sequence continuously collected by a sensing network, synchronously dividing energy metabolism sub-waveforms and blood circulation sub-waveforms with different rhythms, performing cross-cause inspection on the morphological evolution and dynamic fluctuation of the two types of sub-waveforms to generate a physiological state correlation network; tracing the source of non-stationary disturbance events, injecting network iteration to update the sub-waveform topology connection strength, and forming a user-specific physiological state evolution map; deducing the occurrence trend of a target chronic disease within a preset time to obtain a risk quantization trajectory, and reversely screening the metabolism and circulation sensitive waveform segments with the highest contribution degree. The method can accurately analyze the correlation of physiological signals, realize chronic disease risk deduction and key waveform positioning.

Description

Technical Field

[0001] This invention relates to the field of physiological signal risk prediction technology, and in particular to a method and system for predicting the risk of chronic diseases. Background Technology

[0002] Existing technologies for predicting chronic disease risk mainly rely on data analysis of user physiological waveform sequences collected by sensor networks. They typically extract features and construct risk models directly from complete physiological waveforms without breaking down the signals within the physiological waveforms according to rhythmic differences. They use only a single overall physiological signal or discrete physiological indicators as the basis for analysis and determine chronic disease risk through static statistical methods. They do not conduct targeted analysis on the dynamic evolution process of physiological signals and the intrinsic correlation between signals of different physiological mechanisms.

[0003] Existing technologies cannot simultaneously segment energy metabolism sub-waveforms and blood circulation sub-waveforms with rhythmic differences within the same physiological waveform sequence. They also cannot perform cross-causal tests on the morphological evolution paths of energy metabolism sub-waveforms and the dynamic fluctuations of blood circulation sub-waveforms, making it difficult to construct a correlation network that reflects the interconnected relationships of physiological states. Furthermore, existing solutions lack the ability to trace and label non-stationary perturbation events in physiological waveform sequences, and cannot use these labeled perturbation events to iteratively update the topological connection strength between sub-waveforms, thus failing to generate a physiological state evolution map adapted to individual users.

[0004] Existing technologies cannot extrapolate the occurrence trend of target chronic pathological states within a preset time span based on physiological state evolution maps, cannot generate corresponding risk quantification trajectories, and cannot reverse-select the metabolically sensitive waveform segments and circulatory sensitive waveform segments that contribute the most to the trajectory direction based on the risk quantification trajectory. This invention proposes a solution to the technical problems existing in the above-mentioned existing technologies. Summary of the Invention

[0005] The purpose of this invention is to address the shortcomings of existing technologies by proposing a method and system for predicting the risk of chronic diseases.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: a method for predicting the risk of chronic diseases, comprising:

[0007] Receives user physiological waveform sequences continuously acquired by a sensor network;

[0008] In the physiological waveform sequence, energy metabolism sub-waveforms and blood circulation sub-waveforms with rhythmic differences are synchronously divided;

[0009] Cross-causal tests were performed on the morphological evolution path of the energy metabolism sub-waveform and the dynamic fluctuation process of the blood circulation sub-waveform to generate a physiological state correlation network.

[0010] Based on the physiological state association network, non-stationary disturbance events detected in physiological waveform sequences are traced and labeled.

[0011] Non-stationary perturbation events with traceability tags are injected into the physiological state association network, and the topological connection strength of the energy metabolism sub-waveform and the blood circulation sub-waveform is iteratively updated to form a user-specific physiological state evolution map.

[0012] Based on the physiological state evolution map, the trend of users developing target chronic pathological states within a preset time span is inferred, and a risk quantification trajectory is obtained;

[0013] Based on the risk quantification trajectory, the metabolically sensitive waveform segment and the circulatory sensitive waveform segment that contribute the most to the trajectory direction in the physiological waveform sequence are screened in reverse.

[0014] As a further aspect of the present invention, the physiological waveform sequence is synchronously divided into energy metabolism sub-waveforms and blood circulation sub-waveforms with rhythmic differences, including the following steps:

[0015] Multi-scale fluctuation pattern recognition was performed on physiological waveform sequences to separate the metabolic dominant pattern component with a fluctuation period of minutes.

[0016] In the physiological waveform sequence, the cyclic dominant mode component with a fluctuation period of seconds is simultaneously separated;

[0017] Construct a sequence of instantaneous phase differences between the metabolic-dominant mode component and the circulatory-dominant mode component, and calculate the phase-locking index;

[0018] When the phase-locked index remains below the dynamic separation threshold, the metabolic-dominant mode component is identified as the energy metabolism sub-waveform, and the circulatory-dominant mode component is identified as the blood circulation sub-waveform.

[0019] As a further aspect of the present invention, the step of performing a cross-causal test on the morphological evolution path of the energy metabolism sub-waveform and the dynamic fluctuation process of the blood circulation sub-waveform to generate a physiological state correlation network includes the following steps:

[0020] Locate the metabolic pathway inflection point where the waveform slope abruptly changes along the morphological evolution path of the energy metabolism sub-waveform.

[0021] During the dynamic fluctuations of the blood circulation sub-waveform, locate the critical point of the cyclic process where the amplitude jumps.

[0022] To examine the Granger causal effects of metabolic pathway inflection points on critical points of subsequent cyclic processes, and to examine the Granger causal effects of critical points of cyclic processes on Granger causal effects of metabolic pathway inflection points.

[0023] Inflection points and critical points with bidirectional or unidirectional Granger causal effects are constructed as network nodes;

[0024] A physiological state association network is generated by using the time delay and intensity of causal influence between nodes as weighted edges.

[0025] As a further aspect of the present invention, the step of tracing and labeling non-stationary disturbance events detected in physiological waveform sequences based on physiological state association networks includes the following steps:

[0026] Real-time monitoring of the local complexity of physiological waveform sequences; when the complexity index exceeds the adaptive warning line, a non-stationary perturbation event is triggered.

[0027] Extract the local morphological vector of the energy metabolism sub-waveform and the local dynamic vector of the blood circulation sub-waveform when the non-stationary disturbance event occurs;

[0028] In the physiological state association network, search for the historical network node with the highest similarity to the local morphological vector and local dynamic vector;

[0029] The causal influence types of the historical network nodes found are marked as the source type of non-stationary disturbance events.

[0030] As a further aspect of the present invention, the step of injecting the non-stationary perturbation events after source tracing and labeling into the physiological state association network, and iteratively updating the topological connection strength between the energy metabolism sub-waveform and the blood circulation sub-waveform to form a user-specific physiological state evolution map includes the following steps:

[0031] Transform a non-stationary disturbance event with a source-tracing type into a virtual network node with initial weights;

[0032] The virtual network node is connected to the physiological state association network, and the time delay of the causal influence between it and the original network node is calculated.

[0033] Based on the time lag of causal influence, dynamically adjust the connection weights between virtual network nodes and nodes related to energy metabolism sub-waveforms and blood circulation sub-waveforms;

[0034] After multiple iterations, until the network connection weights converge, a physiological state evolution map containing the dynamics of metabolic-circulatory interactions is formed.

[0035] As a further aspect of the present invention, the step of inferring the trend of a user developing a target chronic pathological state within a preset time span based on a physiological state evolution map to obtain a risk quantification trajectory includes the following steps:

[0036] In the physiological state evolution map, a map termination state corresponding to the target chronic pathological state is defined;

[0037] Starting from the current state of the graph, enumerate all paths that reach the final state of the graph through the graph edges;

[0038] Based on the connection weights and time delays of each edge on the path, calculate the transition probability and cumulative time cost of each path;

[0039] By integrating the transition probabilities and cumulative time costs of all paths, a discrete probability distribution of the arrival at the map termination state at different future time points is generated, and the discrete probability distribution constitutes the risk quantification trajectory.

[0040] As a further aspect of the present invention, the step of reversely screening the metabolically sensitive waveform segment and the circulatory sensitive waveform segment that contribute the most to the trajectory direction in the physiological waveform sequence according to the risk quantification trajectory includes the following steps:

[0041] In the risk quantification trajectory, identify the key time points that lead to a sharp increase in the probability distribution;

[0042] By tracing back the evolutionary map of physiological states, we can locate key network edges whose connection weights have changed significantly before critical time points.

[0043] Based on the metabolic pathway inflection points or cyclic process critical points associated with the key network edges, the corresponding time windows are located in the original physiological waveform sequence.

[0044] The original waveform data within the time window are extracted and used as the metabolism-sensitive waveform segment and the circulation-sensitive waveform segment, respectively.

[0045] As a further aspect of the present invention, after examining the Granger causal effects of metabolic pathway inflection points on subsequent cyclic process critical points, and after examining the Granger causal effects of cyclic process critical points on subsequent metabolic pathway inflection points, before constructing inflection points and critical points with significant bidirectional or unidirectional Granger causal effects as network nodes, the following steps are further included:

[0046] When Granger causality is examined, the efficiency of information transfer from the metabolic pathway inflection point to the critical point of the cyclic process is further analyzed.

[0047] Simultaneously analyze the information transfer efficiency from the critical point of the cyclic process to the turning point of the metabolic pathway;

[0048] The process of constructing inflection points and critical points with significant bidirectional or unidirectional Granger causal effects as network nodes includes the following steps:

[0049] Metabolic pathway inflection points and cyclic process critical points with information transmission efficiency exceeding a preset transmission threshold are identified as network nodes with physiological significance.

[0050] As a further aspect of the present invention, after identifying metabolic pathway inflection points and cyclic process critical points with information transmission efficiency higher than a preset transmission threshold as network nodes with physiological significance, the invention further includes the following steps:

[0051] Cluster analysis was performed on the network nodes with physiological significance to identify node clusters that aggregate in the time domain;

[0052] Calculate the community cohesion coefficient for the information transmission efficiency between nodes within each node community;

[0053] The node community with the highest cohesion coefficient is identified as the core physiological regulatory community, which is used to simplify the initial structure of the physiological state association network.

[0054] As a further aspect of the present invention, the present invention also includes a chronic disease risk prediction system, the system including a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor, when executing the computer program, implements the steps of the chronic disease risk prediction method described above.

[0055] Compared with the prior art, the advantages and positive effects of the present invention are as follows:

[0056] By simultaneously segmenting energy metabolism and blood circulation sub-waveforms with rhythmic differences within physiological waveform sequences, the mixed and fused overall physiological signals can be precisely decomposed according to physiological operating mechanisms, allowing signal components corresponding to different physiological functions to be presented independently. Cross-causal testing of the morphological evolution path of the energy metabolism sub-waveform and the dynamic fluctuation process of the blood circulation sub-waveform clarifies the interaction and influence logic between the two types of sub-waveforms. The resulting physiological state correlation network can fully reflect the intrinsic linkage characteristics between different physiological signal components, overcoming the inherent limitations of single-signal-dimensional analysis and achieving refined multi-mechanism, multi-dimensional analysis of physiological signals, making the physiological signal analysis process closely aligned with the actual physiological operating patterns of the human body.

[0057] By performing source tracing and labeling operations on non-stationary disturbances detected in physiological waveform sequences, the location and corresponding source of the disturbances in the physiological signals can be determined. These source-labeled non-stationary disturbances are then injected into a physiological state correlation network, which continuously iteratively adjusts the topological connection strength between energy metabolism and blood circulation sub-waveforms. This allows the network structure to adapt to real-time changes in the user's physiological state, thereby forming a physiological state evolution map that fits the individual user's physiological characteristics. Based on this physiological state evolution map, the trend of target chronic pathological states within a preset time span can be predicted, forming a corresponding risk quantification trajectory. Following this trajectory, the metabolically sensitive and circulatory sensitive waveform segments that contribute most to the trajectory can be directly selected, achieving precise location of disturbances and effective identification of key risk signals. This makes the chronic disease risk prediction process dynamic and targeted, ensuring a high degree of match between risk assessment results and the individual user's physiological state. Attached Figure Description

[0058] Figure 1 This is a flowchart of a chronic disease risk prediction method according to the present invention;

[0059] Figure 2 A flowchart for simultaneously dividing the energy metabolism sub-waveform and the blood circulation sub-waveform;

[0060] Figure 3 This is a flowchart for tracing and labeling non-stationary disturbance events. Detailed Implementation

[0061] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0062] In the description of this invention, it should be understood that the terms "length," "width," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicating orientation or positional relationships, are based on the orientation or positional relationships shown in the accompanying drawings and are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the invention. Furthermore, in the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0063] See Figure 1 This invention provides a method for predicting the risk of chronic diseases, the specific method including:

[0064] The system receives a series of user physiological waveforms continuously acquired by a sensor network. This series consists of mixed waveform data containing multiple types of physiological signals. Within the physiological waveform sequence, energy metabolism sub-waveforms and blood circulation sub-waveforms with rhythmic differences are synchronously segmented. These two sub-waveforms reflect physiological rhythms at different time scales. A cross-causal test is performed on the morphological evolution path of the energy metabolism sub-waveform and the dynamic fluctuation process of the blood circulation sub-waveform. This step aims to analyze the causal relationship of the dynamic interaction between the metabolic and circulatory systems and, based on this, generate a network structure characterizing the associations between physiological subsystems, namely, a physiological state association network. Based on the constructed physiological state association network, non-stationary disturbances detected in real-time in the physiological waveform sequence are traced and labeled to determine the types of physiological associations they may induce. The traced and labeled non-stationary disturbances are injected into the physiological state association network. The topological connection strength between the relevant network nodes of the energy metabolism sub-waveform and the blood circulation sub-waveform is updated through iterative calculation, thereby forming a user-specific atlas reflecting the dynamic evolution of an individual's physiology, namely, a physiological state evolution atlas. Based on this physiological state evolution map, the trend of a user developing a specific target chronic pathological state within a preset time span is deduced. This trend is presented as a trajectory of risk probability changing over time, resulting in a risk quantification trajectory. According to the obtained risk quantification trajectory, the original physiological waveform sequences are analyzed and screened in reverse to locate the metabolically sensitive waveform segments and the circulatory sensitive waveform segments that contribute the most to the direction of the risk trajectory.

[0065] In one embodiment of the present invention, see [reference] Figure 2 Multi-scale fluctuation pattern recognition is performed on physiological waveform sequences. Based on the characteristic time scales of different physiological processes, a metabolic-dominant pattern component with a fluctuation period of minutes is separated from the original sequence. Simultaneously, a circulatory-dominant pattern component with a fluctuation period of seconds is also separated from the physiological waveform sequence. An instantaneous phase difference sequence between the metabolic-dominant and circulatory-dominant pattern components is constructed, and a phase-locking index is calculated based on the stability of this phase difference sequence. When the phase-locking index remains below a preset dynamic separation threshold, the metabolic-dominant pattern component is confirmed as an energy metabolism sub-waveform, and the circulatory-dominant pattern component as a blood circulation sub-waveform, thus completing the division of two sub-waveforms with rhythmic differences.

[0066] A cross-causal test was performed on the morphological evolution path of the energy metabolism sub-waveform and the dynamic fluctuation process of the blood circulation sub-waveform to generate a physiological state association network. The process is as follows: On the morphological evolution path of the energy metabolism sub-waveform, points where the waveform slope changes abruptly were identified as metabolic pathway inflection points. During the dynamic fluctuation process of the blood circulation sub-waveform, points where the amplitude undergoes a step change were identified as critical points of the circulation process. It was tested whether each metabolic pathway inflection point constituted a Granger causal influence on subsequent critical points of the circulation process, and vice versa. Those metabolic pathway inflection points and circulation critical points that were verified to have bidirectional or unidirectional Granger causal influences were constructed as network nodes in the physiological state association network. The time lag of the causal influence between nodes and the strength of the statistical test were used as weighted edges to connect the corresponding network nodes, thereby generating a weighted, directed physiological state association network.

[0067] In practice, a mixed physiological waveform sequence containing ECG, chest impedance, and body movement signals, collected continuously for 24 hours, is used as input. This waveform sequence reflects the user's comprehensive physiological state during daily activities and sleep. Simultaneously separating the energy metabolism sub-waveform and blood circulation sub-waveform with rhythmic differences from the physiological waveform sequence is the basis for subsequent analysis. During implementation, multi-scale fluctuation pattern recognition is performed on the physiological waveform sequence, and the intrinsic mode function (IMF) is separated from the original sequence using an empirical mode decomposition algorithm. One IMF with a significant fluctuation period of 45 to 120 minutes is identified as the metabolic dominant mode component, and this period is associated with the rhythm of gastric emptying and basal energy consumption. Simultaneously, the circulatory dominant mode component with a fluctuation period of seconds is separated from the physiological waveform sequence. This is achieved by extracting the RR interval sequence from the original ECG signal, and after resampling and filtering, obtaining a waveform component dominated by high-frequency heart rate variability.

[0068] In practical implementation, an instantaneous phase difference sequence is constructed between the metabolic-dominant mode component and the circulatory-dominant mode component. The instantaneous phases are obtained through Hilbert transform, and the phase-locking exponent is calculated. The phase-locking exponent quantifies the stability of phase synchronization between the two sub-waveforms, and its calculation can be expressed as:

[0069]

[0070] in: Indicates the phase-locked index. Indicates a point in time The instantaneous phase difference between the metabolic-dominant and circulatory-dominant modes of operation. To calculate the total number of data points within the window, It is an imaginary unit. The phase-locking index ranges from [0,1], with a value closer to 1 indicating stronger phase locking and higher rhythmic synchronicity. In continuous monitoring, when the calculated phase-locking index remains below the preset dynamic separation threshold of 0.3 for 6 consecutive hours, the previously separated metabolic-dominant mode component is confirmed as the energy metabolism sub-waveform, while the circulatory-dominant mode component is confirmed as the blood circulation sub-waveform. To effectively distinguish the rhythmic states of phase synchronization and desynchronization between the energy metabolism sub-waveform and the blood circulation sub-waveform, the dynamic separation threshold is set to 0.3. When the phase-locking index remains below this threshold for 6 consecutive hours, it indicates that the two physiological subsystems have entered a stable and persistent decoupled operation mode. At this point, the division between them exhibits a high degree of temporal independence, laying a reliable foundation for subsequent accurate analysis of the causal interaction between them. This data-driven division method contrasts with the method of simply specifying a specific signal source as a sub-waveform based on prior knowledge, which may not be able to adaptively capture the dynamic coupling and decoupling states between different physiological systems within an individual.

[0071] In some embodiments, a cross-causal test is performed on the morphological evolution path of the energy metabolism sub-waveform and the dynamic fluctuation process of the blood circulation sub-waveform to generate a physiological state correlation network. On the morphological evolution path of the energy metabolism sub-waveform, the metabolic pathway inflection point where the waveform slope abruptly changes is located by calculating the first-order difference of the waveform and finding its local extrema. For example, 90 minutes after a meal, the energy metabolism sub-waveform shows a sustained rise followed by a flattening inflection point, which is recorded as a metabolic pathway inflection point. During the dynamic fluctuation process of the blood circulation sub-waveform, the critical point of the circulatory process where the amplitude undergoes a step change is located by detecting abrupt changes in the waveform envelope, such as a sudden drop in the amplitude of the blood circulation sub-waveform during REM sleep at night.

[0072] In practice, the Granger causal effects of metabolic pathway inflection points on subsequent cyclic process critical points were examined, as were the Granger causal effects of cyclic process critical points on subsequent metabolic pathway inflection points. A time-series Granger causality test model was used, with a maximum lag order of 5. The model was tested to determine whether the data sequences before and after each metabolic pathway inflection point significantly predicted the occurrence of subsequent cyclic process critical points, and vice versa. One analysis showed that for a metabolic pathway inflection point at 3:00 AM, the F-statistic corresponding to a p-value of less than 0.01 in the Granger causality test indicated a significant Granger causal effect of this inflection point on a cyclic process critical point observed at 3:15 AM. However, the p-value of the reverse test was greater than 0.05, indicating that this cyclic process critical point had no significant causal effect on the metabolic pathway inflection point. In the cross-causality test, a Granger causality test model with a maximum lag order of 5 was used. This order range was designed to cover the possible causal delays in physiological signals. The significance level for determining the existence of significant Granger causal influence was set at a p-value of less than 0.01. This stringent standard aims to minimize the probability of misjudging random fluctuations as causal relationships, thereby ensuring that the connections between nodes in the generated physiological state association network are statistically significant, thus improving the network's authenticity and credibility. Inflection points and critical points with bidirectional or unidirectional Granger causal influences are constructed as network nodes. Taking the aforementioned 3 AM event as an example, the metabolic pathway inflection point and the circulatory process critical point are respectively constructed as two nodes in the network. The physiological state association network is generated using the time lag and strength of the causal influence between nodes as weighted edges. The time lag of the causal influence, i.e., the time difference from the cause node to the effect node, such as 15 minutes, is recorded as an edge attribute; the strength of the causal influence is quantified by the F-value of the Granger causality test or a standardized value, and used as the edge weight. The final generated physiological state association network is a directed graph containing multiple such nodes and weighted edges. Compared with undirected association networks constructed by simply calculating correlation coefficients, directed Granger causal networks can reveal the potential temporal driving relationships between physiological events.

[0073] In one embodiment of the present invention, see [reference] Figure 3The system monitors the local complexity index of physiological waveform sequences in real time. When this index exceeds an adaptive warning line dynamically calculated based on historical data, a non-stationary disturbance event is immediately triggered. Local morphological feature vectors of the energy metabolism sub-waveform within a specific time window before and after the non-stationary disturbance event are extracted, along with local dynamic feature vectors of the blood circulation sub-waveform. In the established physiological state association network, historical network nodes with the highest similarity to the currently extracted local morphological and dynamic vectors are searched in parallel. The causal influence type of the searched historical network nodes in the physiological state association network, such as "metabolism affects circulation" or "circulation affects metabolism," is marked as the source type of the non-stationary disturbance event.

[0074] In practice, the local complexity of the physiological waveform sequence is monitored in real time. This local complexity is quantified by calculating the approximate entropy of the waveform within a sliding time window. For a sliding window of 5 minutes with a step size of 10 seconds, the approximate entropy value of the physiological waveform sequence within each window is calculated continuously. In practice, a non-stationary disturbance event is triggered when the complexity index exceeds the adaptive warning threshold. The adaptive warning threshold is not a fixed value but is dynamically set based on the sum of the mean and three standard deviations of the approximate entropy values ​​of the user's historical physiological waveform sequences over the same time period. In the example scenario, the historical mean approximate entropy is 0.32, and the historical standard deviation is 0.05, so the adaptive warning threshold is set to 0.47. When the latest window's approximate entropy value, calculated in real time, reaches 0.53, it exceeds the adaptive warning threshold of 0.47, and the system triggers a non-stationary disturbance event record.

[0075] In some embodiments, the local morphological vector of the energy metabolism sub-waveform and the local dynamic vector of the blood circulation sub-waveform are extracted when a non-stationary disturbance event occurs. Specifically, 90 seconds of energy metabolism sub-waveform data are extracted before and after the trigger time point. Eight features, including waveform skewness, kurtosis, sample entropy, and the root mean square value of the first derivative sequence, are extracted from this waveform to form an eight-dimensional local morphological vector. In another embodiment, 15 seconds of blood circulation sub-waveform data are extracted before and after the same time point. Six features, including mean amplitude, amplitude standard deviation, zero-crossing rate, and dominant frequency power, are extracted from this waveform to form a six-dimensional local dynamic vector.

[0076] Optionally, the historical network node with the highest similarity to the local morphological vector and local dynamic vector can be searched in the physiological state association network. In specific implementation, each historical network node in the physiological state association network stores the local morphological vector of the energy metabolism sub-waveform and the local dynamic vector of the blood circulation sub-waveform associated with it when it is created. The search process is performed by calculating cosine similarity. For the eight-dimensional local morphological vector extracted from the triggering event, the cosine similarity is calculated with the eight-dimensional local morphological vectors stored in all historical network nodes in the physiological state association network. For a historical network node, its comprehensive similarity S with the current event can be calculated by the following formula:

[0077]

[0078] in: This represents the overall similarity score between the current event and a certain historical network node; This is the weight assigned to the local morphological vector similarity, set to 0.6; This is the weight assigned to the local dynamic vector similarity, set to 0.4; Represents the local shape vector of the current event Historical local morphological vector of a certain historical network node Cosine similarity between them; The local dynamic vector representing the current event Historical local dynamic vectors of the same historical network node The cosine similarity between them. In specific implementation, after calculating the overall similarity S between the current event and the 150 historical network nodes in the physiological state association network, the historical network node with the highest overall similarity S with the current event is obtained, with an overall similarity S value of 0.87, while the second highest node has an overall similarity S value of 0.72.

[0079] In some embodiments, the causal influence type of the searched historical network nodes is marked as the source type of the non-stationary perturbation event. Specifically, the historical network node with the highest overall similarity found above is recorded in the physiological state association network as arising from a Granger causal relationship of "metabolism influencing circulation," meaning that this historical network node represents a metabolic pathway inflection point and has a unidirectional causal influence on a critical point of a cyclic process. It can be understood that the system then marks this causal influence type "metabolism influencing circulation" as the source type of the triggered non-stationary perturbation event. In another optional example, if the searched historical network node is defined by a causal relationship of "circulation influencing metabolism," the source type of the event is marked as "circulation influencing metabolism"; if the node is associated with bidirectional causality, it is marked as "metabolism and circulation mutually influencing each other." It can be understood that through this step, each detected non-stationary perturbation event is assigned a clear causal type label originating from the historical physiological state association network.

[0080] In one embodiment of the present invention, non-stationary perturbation events with traceability tags are injected into a physiological state association network. The topological connection strength between the energy metabolism sub-waveform and the blood circulation sub-waveform is iteratively updated to form a user-specific physiological state evolution map. The process is as follows: Each non-stationary perturbation event with a clear traceability type is transformed into a virtual network node with initial weights. This virtual network node is tentatively connected to the existing physiological state association network, and the time delay of the possible causal influence between it and all existing network nodes in the network is calculated. Based on the calculated causal influence time delay and the traceability type of the virtual node, the connection weights between the virtual network node and the nodes related to the energy metabolism sub-waveform and the nodes related to the blood circulation sub-waveform are dynamically adjusted. As new non-stationary perturbation events are continuously recorded and injected, the process of virtual node connection and connection weight adjustment is repeated. After multiple iterations, the connection weight distribution of the entire network tends to stabilize and converge, ultimately forming a user-specific physiological state evolution map containing the dynamic evolution history of individual metabolic-circulatory interactions.

[0081] In practical implementation, non-stationary disturbance events with a source tracing type are transformed into virtual network nodes with initial weights. Taking a non-stationary disturbance event triggered at timestamp "2023-10-26 14:30:05" and with a source tracing type marked as "metabolic influence on circulation" as an example, this event is transformed into a virtual network node with the node identifier "VN_014" and assigned an initial connection weight value, uniformly set to 0.1. In practical implementation, the virtual network node is connected to the physiological state association network, and the time lag of causal influence between the virtual network node and the original network nodes is calculated. For a virtual network node "VN_014", the system retrieves all original network nodes in the physiological state association network that occurred earlier than it in time, calculates the absolute time difference between the timestamp of the virtual network node "VN_014" and the timestamp recorded by each original network node, and uses this as the potential causal time lag for estimation. In a simplified example, the time delay calculation results for the virtual network node "VN_014" and the three original network nodes "Node_A", "Node_B", and "Node_C" are shown in Table 1:

[0082] Table 1: Time Delay Calculation Table for Virtual Network Nodes and Some Existing Network Nodes

[0083] Virtual network node Existing network nodes Existing node timestamp Time delay (seconds) VN_014 Node_A 2023-10-2614:28:20 105 VN_014 Node_B 2023-10-2614:25:10 295 VN_014 Node_C 2023-10-2614:29:55 10

[0084] In some embodiments, the connection weights between virtual network nodes and nodes related to the energy metabolism sub-waveform and the blood circulation sub-waveform are dynamically adjusted based on the time lag of causal influence. In specific implementations, the adjustment of connection weights considers not only the time lag but also the tracing type of the virtual network node and the type of the original network node. For the virtual network node "VN_014", its tracing type is "metabolic influence on circulation," meaning it is more likely to have a causal relationship with nodes related to the energy metabolism sub-waveform and further influence nodes related to the blood circulation sub-waveform. The dynamic adjustment of connection weights follows the formula:

[0085]

[0086] in: This represents the adjusted connection weight between a virtual network node and a certain original network node; This is the initial connection weight of the virtual network node, with a value of 0.1; This is the weighting adjustment factor, used to control the sensitivity of the effect of time delay deviation, and is set to 0.01; It is the calculated time delay, in seconds; This is the expected typical physiological response lag. For the type of "metabolic effects on circulation," the typical physiological response lag is... Set to 60 seconds. (In the formula) This represents the absolute deviation between the actual time delay and the typical time delay. Weighting adjustment factor. Set to 0.01, this value is used to adjust the impact of the deviation between the actual and typical time delays on the update of connection weights, making the weight adjustment process both sensitive and smooth. For events of the "metabolic effect on circulation" type, the typical physiological response time delay... Set to 60 seconds; for events of the "cyclical effect on metabolism" type, the typical time lag is... The time lag is set to 30 seconds. These two different typical time lag values ​​are set based on the different physiological characteristics and response speeds of the interaction between metabolic and circulatory processes. This allows the connection weights of non-stationary disturbances injected into the network to be dynamically adjusted more closely to physiological realities, thus ensuring that the final physiological state evolution map more accurately reflects the individual user's metabolic-circulatory interaction dynamics. It can be understood that when the time lag... The closer to typical time delay At that time, the adjusted connection weights The closer the weights are to the initial weights, the greater the deviation, and the more exponentially the weights decay after adjustment. For example, for the original network node "Node_A" in the table above, the time delay... The time is 105 seconds. Substituting this into the formula, the adjusted weight is calculated. For the existing network node "Node_C", the time delay is... The adjusted weights are calculated over a period of 10 seconds. In practice, the system calculates an adjusted connection weight for each existing network node in the physiological state association network of the virtual network node "VN_014", and uses this as the initial connection strength to add corresponding directed edges to the network. The direction of the edges is inferred by causal logic. For virtual nodes of the "metabolic influence on circulation" type, by default, edges are added from the virtual node to the blood circulation sub-waveform related nodes, and edges are added from the energy metabolism sub-waveform related nodes to the virtual node.

[0087] In practice, after multiple iterations until the network connection weights converge, a physiological state evolution graph containing the dynamics of metabolic-circulatory interactions is formed. Each time a new non-stationary perturbation event is detected, marked, and transformed into a virtual network node, the aforementioned access and weight adjustment process is executed. This not only adds new nodes and edges but may also trigger a re-evaluation and update of existing edge weights. Optionally, the convergence of the network connection weights is determined by monitoring the rate of change of the Frobenius norm of the global weight matrix of the entire physiological state association network. When the rate of change of the global weight matrix after five consecutive injections of new events is below the threshold of 0.5%, the network connection weights are considered to have converged. It can be understood that after a period of continuous data injection and iterative updates, the initial static physiological state association network evolves into a dynamic graph structure with stable connection weights that reflects the individual's long-term physiological interaction patterns—this is the user-specific physiological state evolution graph.

[0088] In one embodiment of the present invention, based on a physiological state evolution graph, the trend of a user developing a target chronic pathological state within a preset time span is deduced to obtain a risk quantification trajectory. The specific deduction method includes: In the physiological state evolution graph, a graph termination state corresponding to the target chronic pathological state is defined according to medical knowledge; this state is described by a set of network node features. Starting from the current state of the graph, all potential paths that can reach the graph termination state through directed edges in the graph are enumerated. Based on the connection weights and time delay parameters of each directed edge on the path, the transition probability of each path and the cumulative time cost required to reach the termination state from the current state are calculated. The transition probabilities and cumulative time costs of all possible paths are integrated to generate a discrete probability distribution of the user's physiological state evolving to the graph termination state at different future time points; this discrete probability distribution constitutes the risk quantification trajectory.

[0089] Based on the risk quantification trajectory, the metabolically sensitive and circulatory sensitive waveform segments that contribute most to the trajectory's direction are selected in reverse from the physiological waveform sequence. The reverse selection process is as follows: In the generated risk quantification trajectory, one or more key time points that cause a sharp increase in the probability distribution leading to the termination state are identified. By tracing back the structure and historical state of the physiological state evolution map, one or more key network edges whose connection weights have changed significantly before these key time points are located. Based on the metabolic pathway turning points or circulatory process critical points associated with the key network edges in the physiological state evolution map, these are back-mapped to the original physiological waveform sequence to locate the time windows corresponding to these turning points or critical points. The original waveform data within the time windows are extracted and designated as the metabolically sensitive and circulatory sensitive waveform segments that significantly contribute to the risk trajectory's direction.

[0090] In practical implementation, risk extrapolation and reverse localization of key physiological signals are performed based on the physiological state evolution map. The following example scenario, using specific data, illustrates how the target chronic pathological state is extrapolated to be "compensated heart failure." In practice, a map termination state corresponding to the target chronic pathological state is defined in the physiological state evolution map. The definition process is based on clinical guidelines and historical case data. For example, the map termination state is defined as a set that simultaneously contains the following network node features: node feature A (average slope of the energy metabolism sub-waveform is below the threshold of 0.1 for three consecutive cycles), node feature B (variance coefficient of amplitude of the blood circulation sub-waveform is greater than 0.4 for two consecutive cycles), and node feature C (average time lag of the metabolic-circulatory causal network edges is greater than 120 seconds). When a node or combination of nodes simultaneously satisfies these three features appears in the physiological state evolution map, the map is considered to have entered the defined termination state.

[0091] In the specific implementation, starting from the current state of the graph, all paths leading to the final state of the graph via graph edges are enumerated. The current state of the graph is represented by a set of active network nodes, such as nodes N1 and N2. The enumeration process is implemented using a graph traversal algorithm, with a search depth of 5 steps. In the example, starting from the currently active nodes N1 and N2, a total of three paths leading to the final state of the graph are enumerated. The path enumeration results are shown in Table 2.

[0092] Table 2: Enumeration of Paths from Current State to Terminal State of the Graph

[0093] Path number Path node sequence Number of edges in the path Path 1 N1->N3->N5->Termination State 3 Path 2 N1->N4->Termination State 2 Path 3 N2->N5->Termination State 2

[0094] In some embodiments, the transition probability and cumulative time cost of each path are calculated based on the connection weights and time delays of each edge on the path. In a specific implementation, the transition probability of a path is calculated from the connection weights of each directed edge on the path; the transition probability of a path... The calculation formula is:

[0095]

[0096] in: This represents the transition probability of this path; This indicates the number of directed edges contained in the path; Indicates the first path The connection weights of the directed edges are given, with a range of (0,1). In the example, the weights of the three edges of path 1 are 0.7, 0.8, and 0.6, respectively. The transition probability of path 1 is... In practice, the cumulative time cost of a path is obtained by summing the time delays of each directed edge on the path. ,in Indicates the first path The time delay is recorded for each directed edge. In the example, the time delays for the three edges of path 1 are 3 days, 5 days, and 7 days, respectively. The cumulative time cost of path 1 is... sky.

[0097] Optionally, the transition probabilities and cumulative time costs of all paths can be integrated to generate a discrete probability distribution of arrival at the map termination state at different future time points. In specific implementations, the calculated cumulative time cost for each path will be used... Round to the nearest integer time point and set the transition probability of the path. The cumulative time cost is added to the probability corresponding to that point in time. In the example, the cumulative time cost of path 1 is 15 days, and the transition probability is 0.336; the cumulative time cost of path 2 is 10 days, and the transition probability is 0.45; the cumulative time cost of path 3 is 12 days, and the transition probability is 0.25. After integration, a discrete probability distribution is generated. For example, the probability of reaching the termination state on day 10 is 0.45, on day 12 it is 0.25, and on day 15 it is 0.336. This discrete probability distribution constitutes the risk quantification trajectory.

[0098] In some embodiments, the metabolically sensitive and circulatory sensitive waveform segments that contribute the most to the trajectory direction in the physiological waveform sequence are screened in reverse according to the risk quantification trajectory. In a specific implementation, key time points that cause a sharp increase in the probability distribution are identified in the risk quantification trajectory. This is achieved by analyzing the first difference of the discrete probability distribution curve. In the example probability distribution, the probability value on day 10 (0.45) shows a significant jump compared to day 9 (assumed to be 0.05), thus day 10 is identified as a key time point. In a specific implementation, the physiological state evolution map is traced back to locate key network edges whose connection weights change significantly before the key time point. The system then backtracks and checks the edges with the largest rate of change in connection weights among all network edges within a time window before the key time point.

[0099] It is understandable that the corresponding time window in the original physiological waveform sequence is located based on the metabolic pathway inflection point or circulatory process critical point associated with the key network edge. In the example, the starting node N4 of the key network edge "N4->Termination State" is associated with a historical metabolic pathway inflection point, which is recorded as occurring at "2023-11-01 08:15:00" in the original physiological waveform sequence. In specific implementation, a time window centered on this time point and extending 5 minutes before and after it is extracted, and the original energy metabolism sub-waveform data within this time window is extracted; this segment of data is the metabolically sensitive waveform segment. Optionally, if the key network edge is associated with a circulatory process critical point, then the original blood circulation sub-waveform data within the corresponding time window is extracted as the circulatory sensitive waveform segment. It is understandable that through this reverse screening process, specific waveform segments that contribute highly to the evolution of risk trends can be accurately located from a massive amount of original physiological waveform sequences.

[0100] In one embodiment of the present invention, after examining the Granger causal effects of metabolic pathway inflection points on subsequent cyclic process critical points, and after examining the Granger causal effects of cyclic process critical points on subsequent metabolic pathway inflection points, before constructing inflection points and critical points with significant bidirectional or unidirectional Granger causal effects as network nodes, a step of refined evaluation of causal effects is included. When Granger causal effects are detected, the information transfer efficiency in the direction from metabolic pathway inflection points to cyclic process critical points is further analyzed. This efficiency can be calculated by transfer entropy or similar quantitative indicators. Simultaneously, the information transfer efficiency in the opposite direction from cyclic process critical points to metabolic pathway inflection points is analyzed. Subsequently, inflection points and critical points with significant bidirectional or unidirectional Granger causal effects are constructed as network nodes. Specifically, metabolic pathway inflection points and cyclic process critical points whose information transfer efficiency calculated in the above steps is higher than a preset transfer threshold are identified as physiologically significant network nodes, and only these nodes are included in the network construction.

[0101] After identifying metabolic pathway inflection points and cyclic process critical points with information transmission efficiency exceeding a preset threshold as physiologically significant network nodes, the process further includes a step of optimizing the network node structure. Time-based clustering analysis is performed on all physiologically significant network nodes to identify node clusters that aggregate in the time domain. The average or comprehensive index of information transmission efficiency among nodes within each node cluster is calculated as the cluster's cohesion coefficient. The node cluster with the highest cohesion coefficient is identified as the core physiological regulatory cluster. When constructing the initial structure of the physiological state association network, this core cluster can be used as a basis for simplification and expansion.

[0102] In practice, when a Granger causal influence is detected between a metabolic pathway inflection point and the critical point of the subsequent cyclic process, the information transfer efficiency from the metabolic pathway inflection point to the critical point of the cyclic process is further analyzed. This information transfer efficiency is quantified by calculating the transfer entropy from the time series of the metabolic pathway inflection point to the time series of the cyclic process critical point. In one example, for a metabolic pathway inflection point... A critical point of a cyclic process Calculate from arrive The propagation entropy value is calculated based on local segment data of the sub-waveforms where the two event points are located. arrive Information transmission efficiency value In practical implementation, the information transfer efficiency from the critical point of the cyclic process to the turning point of the metabolic pathway is analyzed simultaneously. In the same example, the efficiency from the critical point of the cyclic process is calculated. To the metabolic pathway inflection point The information transmission efficiency value obtained is the transmission entropy. .

[0103] In some embodiments, metabolic pathway inflection points and cyclic process critical points with information transmission efficiency exceeding a preset transmission threshold are identified as physiologically significant network nodes. The preset transmission threshold is set as follows: In the example, due to the metabolic pathway inflection point... To the critical point of the cyclic process Information transmission efficiency value Higher than the preset transmission threshold Therefore, metabolic pathway inflection point These were identified as network nodes with physiological significance. And from the critical point of the cyclic process... To the metabolic pathway inflection point Information transmission efficiency value Below the preset transmission threshold Therefore, the critical point of the cyclic process This node was not identified as a significant network node in this direction; however, if If, in the examination of other event points, its information transmission efficiency value as a cause is higher than the threshold, it may still be identified as a network node due to other causal relationships.

[0104] In practice, cluster analysis is performed on network nodes with physiological significance to identify node clusters that aggregate in the time domain. The K-means clustering algorithm based on network node timestamps is employed. In one example, 12 network nodes were identified as having physiological significance, with their timestamps distributed within a certain range. The number of clusters was set accordingly. After clustering, three node clusters are formed: Cluster 1 contains 4 nodes with adjacent timestamps, Cluster 2 contains 5 nodes, and Cluster 3 contains 3 nodes. In specific implementation, the cluster cohesion coefficient, which measures the information transmission efficiency between nodes within each cluster, is calculated. The cluster cohesion coefficient is defined as the average of the information transmission efficiency values ​​between all node pairs within the cluster, and its calculation follows the formula:

[0105]

[0106] in: This represents the community cohesion coefficient of a given node community. This indicates the number of network nodes contained within the community; This represents the number of all possible node pairs within the community, and its value is... ; Represents nodes within a community With nodes The maximum value of the bidirectional information transmission efficiency between nodes. In the example, for community 1 containing 4 nodes, there are 6 pairs of nodes within it. The maximum transmission efficiency value between each pair of nodes is calculated and averaged to obtain the cohesion coefficient of community 1. The cohesion coefficient of community 2 was calculated using the same method. The cohesion coefficient of community 3 is Optionally, the node community with the highest community cohesion coefficient can be identified as the core physiological regulatory community. In the example, the community cohesion coefficient of community 3 is... Therefore, node community 3 was identified as the core physiological regulatory community, as it has the highest probability of occurrence. This core physiological regulatory community reflects a set of key physiological events that occur concurrently in the time domain and have close internal information transmission. In some embodiments, the core physiological regulatory community is used to simplify the initial structure of the physiological state association network. When constructing the initial version of the network, the core physiological regulatory community can be treated as a highly cohesive submodule or supernode, thereby reducing the initial complexity of the network structure.

[0107] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A method for predicting the risk of chronic diseases, characterized in that, Includes the following steps: Receives user physiological waveform sequences continuously acquired by a sensor network; In the physiological waveform sequence, energy metabolism sub-waveforms and blood circulation sub-waveforms with rhythmic differences are synchronously divided; Cross-causal tests were performed on the morphological evolution path of the energy metabolism sub-waveform and the dynamic fluctuation process of the blood circulation sub-waveform to generate a physiological state correlation network. Based on the physiological state association network, non-stationary disturbance events detected in physiological waveform sequences are traced and labeled. Non-stationary perturbation events with traceability tags are injected into the physiological state association network, and the topological connection strength of the energy metabolism sub-waveform and the blood circulation sub-waveform is iteratively updated to form a user-specific physiological state evolution map. Based on the physiological state evolution map, the trend of users developing target chronic pathological states within a preset time span is inferred, and a risk quantification trajectory is obtained; Based on the risk quantification trajectory, the metabolically sensitive waveform segment and the circulatory sensitive waveform segment that contribute the most to the trajectory direction in the physiological waveform sequence are screened in reverse.

2. The method for predicting chronic disease risk according to claim 1, characterized in that, In the physiological waveform sequence, energy metabolism sub-waveforms and blood circulation sub-waveforms with rhythmic differences are synchronously divided, including the following steps: Multi-scale fluctuation pattern recognition was performed on physiological waveform sequences to separate the metabolic dominant pattern component with a fluctuation period of minutes. In the physiological waveform sequence, the cyclic dominant mode component with a fluctuation period of seconds is simultaneously separated; Construct a sequence of instantaneous phase differences between the metabolic-dominant mode component and the circulatory-dominant mode component, and calculate the phase-locking index; When the phase-locked index remains below the dynamic separation threshold, the metabolic-dominant mode component is identified as the energy metabolism sub-waveform, and the circulatory-dominant mode component is identified as the blood circulation sub-waveform.

3. The method for predicting chronic disease risk according to claim 2, characterized in that, The process of performing a cross-causal test on the morphological evolution path of the energy metabolism sub-waveform and the dynamic fluctuation process of the blood circulation sub-waveform to generate a physiological state correlation network includes the following steps: Locate the metabolic pathway inflection point where the waveform slope abruptly changes along the morphological evolution path of the energy metabolism sub-waveform. During the dynamic fluctuations of the blood circulation sub-waveform, locate the critical point of the cyclic process where the amplitude jumps. To examine the Granger causal effects of metabolic pathway inflection points on critical points of subsequent cyclic processes, and to examine the Granger causal effects of critical points of cyclic processes on Granger causal effects of metabolic pathway inflection points. Inflection points and critical points with bidirectional or unidirectional Granger causal effects are constructed as network nodes; A physiological state association network is generated by using the time delay and intensity of causal influence between nodes as weighted edges.

4. The method for predicting chronic disease risk according to claim 1, characterized in that, The method of tracing and labeling non-stationary disturbances detected in physiological waveform sequences based on physiological state association networks includes the following steps: Real-time monitoring of the local complexity of physiological waveform sequences; when the complexity index exceeds the adaptive warning line, a non-stationary perturbation event is triggered. Extract the local morphological vector of the energy metabolism sub-waveform and the local dynamic vector of the blood circulation sub-waveform when the non-stationary disturbance event occurs; In the physiological state association network, search for the historical network node with the highest similarity to the local morphological vector and local dynamic vector; The causal influence types of the historical network nodes found are marked as the source type of non-stationary disturbance events.

5. The method for predicting chronic disease risk according to claim 4, characterized in that, The process of injecting non-stationary perturbation events with traceable tags into the physiological state association network, iteratively updating the topological connection strength between the energy metabolism sub-waveform and the blood circulation sub-waveform, and forming a user-specific physiological state evolution map includes the following steps: Transform a non-stationary disturbance event with a source-tracing type into a virtual network node with initial weights; The virtual network node is connected to the physiological state association network, and the time delay of the causal influence between it and the original network node is calculated. Based on the time lag of causal influence, dynamically adjust the connection weights between virtual network nodes and nodes related to energy metabolism sub-waveforms and blood circulation sub-waveforms; After multiple iterations, until the network connection weights converge, a physiological state evolution map containing the dynamics of metabolic-circulatory interactions is formed.

6. The method for predicting chronic disease risk according to claim 1, characterized in that, The process of extrapolating the trend of a user developing a target chronic pathological state within a preset time span based on a physiological state evolution map, and obtaining a risk quantification trajectory, includes the following steps: In the physiological state evolution map, a map termination state corresponding to the target chronic pathological state is defined; Starting from the current state of the graph, enumerate all paths that reach the final state of the graph through the graph edges; Based on the connection weights and time delays of each edge on the path, calculate the transition probability and cumulative time cost of each path; By integrating the transition probabilities and cumulative time costs of all paths, a discrete probability distribution of the arrival at the map termination state at different future time points is generated, and the discrete probability distribution constitutes the risk quantification trajectory.

7. The method for predicting chronic disease risk according to claim 6, characterized in that, The process of reverse-selecting the metabolically sensitive and circulatory sensitive waveform segments that contribute the most to the trajectory direction in the physiological waveform sequence based on the risk quantification trajectory includes the following steps: In the risk quantification trajectory, identify the key time points that lead to a sharp increase in the probability distribution; By tracing back the evolutionary map of physiological states, we can locate key network edges whose connection weights have changed significantly before critical time points. Based on the metabolic pathway inflection points or cyclic process critical points associated with the key network edges, the corresponding time windows are located in the original physiological waveform sequence. The original waveform data within the time window are extracted and used as the metabolism-sensitive waveform segment and the circulation-sensitive waveform segment, respectively.

8. The method for predicting chronic disease risk according to claim 3, characterized in that, After examining the Granger causal effects of metabolic pathway inflection points on subsequent cyclic process critical points, and after examining the Granger causal effects of cyclic process critical points on subsequent metabolic pathway inflection points, before constructing inflection points and critical points with significant bidirectional or unidirectional Granger causal effects as network nodes, the following steps are also included: When Granger causality is examined, the efficiency of information transfer from the metabolic pathway inflection point to the critical point of the cyclic process is further analyzed. Simultaneously analyze the information transfer efficiency from the critical point of the cyclic process to the turning point of the metabolic pathway; The process of constructing inflection points and critical points with significant bidirectional or unidirectional Granger causal effects as network nodes includes the following steps: Metabolic pathway inflection points and cyclic process critical points with information transmission efficiency exceeding a preset transmission threshold are identified as network nodes with physiological significance.

9. The method for predicting chronic disease risk according to claim 8, characterized in that, After identifying metabolic pathway inflection points and cyclic process critical points with information transmission efficiency exceeding a preset transmission threshold as physiologically significant network nodes, the method further includes the following steps: Cluster analysis was performed on the network nodes with physiological significance to identify node clusters that aggregate in the time domain; Calculate the community cohesion coefficient for the information transmission efficiency between nodes within each node community; The node community with the highest cohesion coefficient is identified as the core physiological regulatory community, which is used to simplify the initial structure of the physiological state association network.

10. A chronic disease risk prediction system, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the chronic disease risk prediction method according to any one of claims 1 to 9.

Images