A severe patient multi-organ failure evolution path prediction system

Abstract

The application relates to the technical field of medical data processing, and discloses a critical patient multi-organ function failure evolution path prediction system, which comprises a collection and preprocessing module, an adaptive dynamic characteristic extraction module, a physical dissipation constraint causal topology analysis module, a closed-loop feedback controller and a cascade failure path deduction module. The system extracts dynamic characteristics and calculates a physical effectiveness coefficient based on multi-modal physiological signals, uses the coefficient to correct transfer entropy to construct a multi-organ coupling network; the closed-loop feedback controller dynamically adjusts the embedding dimension parameter of the front-end characteristic extraction according to the total in-degree coupling strength of the network, forming a bidirectional constraint closed loop between the physical layer and the information layer. The application can effectively identify a pathological driving source, deduce the cascade propagation sequence of organ function failure, eliminate false causal connections through physical mechanism constraint and feedback regulation, and improve the accuracy of evolution path prediction.

Description

Technical Field

[0001] This invention relates to the field of medical data processing technology, specifically to a system for predicting the evolution path of multiple organ failure in critically ill patients. Background Technology

[0002] Currently, multiple organ dysfunction syndrome (MODS) is one of the leading causes of death among patients in intensive care units. The physiological mechanisms of critically ill patients are complex; the failure of a single organ is often not an isolated event, but rather triggers a chain reaction through complex physiological coupling mechanisms, rapidly affecting other organ systems. The evolution of this pathological state exhibits highly nonlinear and dynamic time-varying characteristics. The key to clinical treatment lies in accurately predicting the evolutionary path before the cascading spread of organ failure occurs, thereby enabling targeted interventions.

[0003] For data analysis in the aforementioned intensive care settings, current technologies primarily rely on bedside monitors to collect multimodal physiological signals such as heart rate, blood pressure, and respiratory rate. Conventional approaches typically employ time-domain or frequency-domain analysis methods to process these signals individually, or use clinical scales such as the SOFA score to periodically assess the functional status of individual organs. Some cutting-edge applications are beginning to utilize multivariate time-series analysis or machine learning algorithms to calculate correlation coefficients or Granger causality relationships between different physiological indicators, attempting to construct static organ correlation networks to assist physicians in observing whether abnormal synchronous changes exist among various organs.

[0004] However, existing analytical methods still have shortcomings in practical applications. First, traditional dynamic feature extraction often presupposes fixed parameters such as embedding dimensions, ignoring the dramatic fluctuations in the system coupling complexity during the pathological evolution of critically ill patients, leading to biased or overfitting feature representations in non-stationary states. Second, organ network construction often relies solely on statistical transfer entropy calculations, lacking dissipation constraints based on physiological and physical mechanisms, making it susceptible to numerous spurious causal connections due to data noise or common driving factors, resulting in network topology distortion. Furthermore, existing technologies mostly focus on the static assessment of the current state of each organ, making it difficult to accurately identify the driving sources of pathological cascade failure by quantifying the net flow of information, and unable to deduce the specific failure propagation sequence and path.

[0005] Therefore, the present invention provides a system for predicting the evolution path of multiple organ failure in critically ill patients, in order to address the shortcomings of the prior art. Summary of the Invention

[0006] To address the shortcomings of existing technologies, this invention provides a prediction system for the evolution path of multiple organ dysfunction syndrome (MODS) in critically ill patients. This system solves the problem that existing MODS prediction methods fail to accurately locate the pathological source and deduce the path of failure due to the incompatibility of feature extraction parameters with dynamic pathological changes and the lack of physical constraints in network construction leading to false connections.

[0007] To achieve the above objectives, the present invention provides the following technical solution: a system for predicting the evolution path of multiple organ failure in critically ill patients, comprising:

[0008] A multimodal physiological signal acquisition and preprocessing module is used to output a standardized multidimensional synchronous time series matrix;

[0009] An adaptive dynamics feature extraction module is used to perform phase space reconstruction on the multidimensional synchronous time series matrix to extract physical layer dynamics features and output physical effectiveness coefficients;

[0010] A causal topology analysis module under physical dissipation constraints is used to construct a multi-organ coupling network representing the interaction relationship of the information layer using the physical effectiveness coefficient;

[0011] The cascading failure path deduction module is used to deduce the cascading propagation sequence of organ failure based on the multi-organ coupling network and generate evolution path prediction results.

[0012] A closed-loop feedback controller is used to extract the total in-degree coupling strength from the multi-organ coupling network and provide feedback.

[0013] The adaptive dynamic feature extraction module responds to the feedback of the total in-degree coupling strength and dynamically adjusts the embedding dimension parameters of the phase space reconstruction to form a bidirectional constraint closed loop between the physical layer dynamic feature extraction and the information layer interaction relationship construction.

[0014] By adopting the above technical solution, the system can dynamically optimize feature extraction parameters based on the global coupling state of the organ network by establishing a bidirectional constraint closed-loop structure between physical layer dynamic feature extraction and information layer network topology construction. Specifically, the closed-loop feedback controller feeds back the total in-degree coupling strength, which characterizes the degree of network coupling, to the front end, enabling the adaptive dynamic feature extraction module to perceive the current complexity level of the system and adjust the embedding dimension accordingly. This mechanism overcomes the shortcomings of traditional methods where physical feature extraction and network analysis are separated, ensuring that even in pathological states with high coupling of multiple organs, the extracted dynamic features can still accurately reflect the physical essence of the system, thereby improving the accuracy and robustness of causal network construction and failure path deduction.

[0015] Preferably, the multimodal physiological signal acquisition and preprocessing module uses wavelet transform filters to remove baseline drift, electromyographic interference, and power frequency interference from the acquired high-frequency waveform data; for the acquired low-frequency discrete data, a cubic spline interpolation algorithm is used to map the low-frequency discrete data onto a time axis that is consistent with the high-frequency waveform data, filling the sampling gaps and forming a continuous time series; and all channel data are standardized using the Z-score standardization method.

[0016] By adopting the above technical solution, the problems of inconsistent sampling frequencies and dimensional differences of physiological signals of different modalities were solved, providing a time-domain aligned and scale-uniform data foundation for subsequent multivariate coupling analysis, and effectively eliminating the impact of noise interference and missing data on model accuracy.

[0017] Preferably, the adaptive dynamics feature extraction module calculates the physical effectiveness coefficient by: estimating the local Jacobian matrix of the trajectory points in the reconstructed phase space using the k-nearest neighbor algorithm, calculating the trace of the local Jacobian matrix to quantify the divergence of the phase volume, and using the sigmoid activation function to map the divergence into a normalized value as the physical effectiveness coefficient.

[0018] By employing the above technical solution, the local dissipative characteristics of the system are characterized using the local Jacobian matrix trace of the phase space trajectory, enabling the quantification of the current functional state of the organ from a physical mechanism perspective. Mapping the divergence to a physical validity coefficient provides a criterion based on a physical mechanism for subsequent elimination of spurious causal connections; that is, when an organ is in an unstable or highly dissipative state, its ability to transmit or receive information is correspondingly constrained.

[0019] Preferably, the strategy for the adaptive dynamic feature extraction module to dynamically adjust the embedding dimension parameter is as follows: read the total in-degree coupling strength fed back from the previous moment; set a preset baseline embedding dimension; calculate the product of the natural logarithm of the total in-degree coupling strength and the adjustment coefficient; subtract the product from the baseline embedding dimension to obtain a calculated value; perform a floor operation on the calculated value; compare the calculation result with a preset minimum embedding dimension threshold; and select the larger value between the calculation result and the minimum embedding dimension threshold as the embedding dimension parameter at the current moment.

[0020] By adopting the above technical solution, a dimension adaptive adjustment mechanism based on information flow intensity was constructed. When the total in-degree coupling strength increases, it indicates that the information redundancy between systems is improved or the synchronization is enhanced. By appropriately reducing the embedding dimension through logarithmic decay, both computational disasters caused by excessive dimensionality and overfitting can be avoided. At the same time, setting a minimum threshold ensures that the basic topological structure of phase space reconstruction is not destroyed, thus achieving a balance between computational efficiency and feature representation accuracy.

[0021] Preferably, the causal topology analysis module under physical dissipation constraints constructs the multi-organ coupling network by: using kernel density estimation to calculate the original transfer entropy between monitoring organ node pairs, multiplying the original transfer entropy value from the source organ node to the target organ node by the physical effectiveness coefficient of the target organ node to obtain the physical constraint transfer entropy, and determining the weight of the directed edge connecting the nodes based on the physical constraint transfer entropy.

[0022] By adopting the above technical solution, a physical validity coefficient is introduced to correct the propagation entropy calculated purely statistically. Essentially, this involves weighting the incoming information flow using the physical receiving capability of the target node. This physical-information fusion method can effectively suppress spurious causal connections caused by data noise or co-driving forces, enabling the constructed multi-organ coupling network to more realistically reflect the physiological interaction mechanisms under pathological conditions.

[0023] Preferably, the closed-loop feedback controller extracts the total in-degree coupling strength by: for each organ node in the multi-organ coupling network, accumulating the physical constraint transfer entropy of all source organ nodes pointing to the organ node to obtain the total in-degree coupling strength of the organ node at the current moment.

[0024] By adopting the above technical solution, the strength of a specific organ as an information aggregation point in the network is quantified. This indicator macroscopically reflects the degree to which the organ is affected by other organs, providing a global feedback variable for the parameter adjustment of the front-end feature extraction module.

[0025] Preferably, the closed-loop feedback controller performs a time-step delay processing on the calculated total in-degree coupling strength value, and sends the delayed total in-degree coupling strength as a feedback parameter to the adaptive dynamic feature extraction module.

[0026] By adopting the above technical solution, the temporal causal logic of feedback control is ensured, so that the parameter adjustment at the current moment is based on the network state at the previous moment, avoiding algebraic loop problems and ensuring the stability of system operation.

[0027] Preferably, the cascaded failure path deduction module identifies the pathological driving source node by calculating the net outflow entropy of each organ node in the multi-organ coupling network. The cascaded failure path deduction module calculates the net outflow entropy as follows: for any organ node in the multi-organ coupling network, the sum of the physical constraint transmission entropy from the organ node to all other nodes in the network is calculated as the total outflow information, and the sum of the physical constraint transmission entropy from all other nodes in the network to the organ node is calculated as the total inflow information. The net outflow entropy of the organ node is obtained by subtracting the total inflow information from the total outflow information. The cascaded failure path deduction module selects the node with the largest positive net outflow entropy value as the pathological driving source node.

[0028] By employing the above technical solution, the dominant direction of information flow in the network is clarified through the calculation of net outflow entropy. The node with the largest positive net outflow entropy indicates that its outward pathological impact far outweighs the regulatory information it receives, thereby accurately pinpointing the primary lesion organ leading to multiple organ dysfunction syndrome.

[0029] Preferably, the cascading failure path deduction module employs a maximum probability path search algorithm, using the identified pathological driving source node as the starting point, to search for a propagation path in the network topology starting from the starting point. The propagation path consists of a chain of node sequences with the highest cumulative transfer probability, used to characterize the cascading transmission order of the pathological state.

[0030] By adopting the above technical solutions, it is possible to predict the most likely propagation trajectory of pathological states in organ networks based on the current coupled network topology, thereby providing clinicians with intuitive predictions of evolution paths and helping to carry out targeted early interventions before cascade failure occurs.

[0031] Preferably, an implementation architecture combining edge computing and cloud computing is adopted; the multimodal physiological signal acquisition and preprocessing module and the adaptive dynamic feature extraction module are integrated at the edge computing end for performing data processing and calculating the physical validity coefficient locally; the causal topology analysis module under physical dissipation constraints, the closed-loop feedback controller, and the cascaded failure path inference module are deployed at the cloud processing center; the edge computing end and the cloud processing center are connected through an encrypted communication link for transmitting the physical validity coefficient, the dimensionality-reduced feature vector, and the feedback total in-degree coupling strength.

[0032] By adopting the above technical solution, the computing load is rationally allocated. The edge is responsible for signal processing and feature extraction with high real-time requirements, reducing the bandwidth pressure of uploading raw big data; the cloud is responsible for network topology analysis and global path inference with high computing power requirements. At the same time, the existence of a feedback mechanism enables the cloud to remotely control the computing parameters of the edge, realizing distributed intelligent collaborative prediction.

[0033] This invention provides a system for predicting the progression of multiple organ failure in critically ill patients. It has the following beneficial effects:

[0034] 1. This invention establishes a bidirectional constraint closed-loop structure between physical layer feature extraction and information layer topology construction. A closed-loop feedback controller feeds back the total in-degree coupling strength, representing the network coupling state, to the front end, dynamically adjusting the embedding dimension parameters for phase space reconstruction. This adaptive adjustment mechanism enables the system to optimize the feature extraction strategy in real time based on the complexity of interactions between multiple organs. This avoids the overfitting or underfitting problems of traditional static parameter settings when processing non-stationary physiological signals, ensuring the accuracy of dynamic feature extraction and the robustness of the system under highly coupled pathological conditions involving multiple organs.

[0035] 2. This invention introduces a topological constraint mechanism based on physical dissipation characteristics. By calculating the trace-quantized phase volume divergence of the local Jacobian matrix and generating a physical validity coefficient, this coefficient is used to weight and correct the transfer entropy calculated statistically. This method, combining physical mechanisms with information theory, utilizes the physical receiving capabilities of target organs to constrain information flow, effectively eliminating spurious causal connections caused by data noise or co-driving forces. This improves the realism of multi-organ coupled network construction, enabling it to more accurately reflect the inherent interactive logic of physiological systems.

[0036] 3. This invention identifies pathological driving source nodes by calculating net outflow entropy and deduces the cascading failure sequence by combining a maximum probability path search algorithm. This scheme can accurately locate the initial lesion organ causing system collapse from complex network topology and quantitatively predict the propagation trajectory of pathological states in the organ network, providing clinicians with visualized prediction of functional failure evolution paths, which helps to achieve early warning and precise intervention before the occurrence of multiple organ dysfunction syndrome. Attached Figure Description

[0037] Figure 1 This is an architecture diagram of a system for predicting the evolution path of multiple organ failure in critically ill patients according to an embodiment of the present invention.

[0038] Figure 2 This is a flowchart of the method for predicting the evolution path of multiple organ failure in critically ill patients according to an embodiment of the present invention.

[0039] Among them, 100 is the multimodal physiological signal acquisition and preprocessing module; 200 is the adaptive dynamic feature extraction module; 300 is the causal topology analysis module under physical dissipation constraints; 400 is the closed-loop feedback controller; and 500 is the cascade failure path inference module. Detailed Implementation

[0040] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0041] See attached document Figure 1 This invention provides a system for predicting the evolution path of multiple organ failure in critically ill patients. The system includes a multimodal physiological signal acquisition and preprocessing module 100, an adaptive dynamic feature extraction module 200, a causal topology analysis module under physical dissipation constraints 300, a closed-loop feedback controller 400, and a cascade failure path deduction module 500.

[0042] The multimodal physiological signal acquisition and preprocessing module 100 is configured to acquire and standardize multi-channel time-series data from critically ill patients. This module connects to bedside monitoring equipment and the hospital information system via a communication protocol to acquire data sets in real time. These data sets contain physiological signals from monitored organ nodes, specifically divided into high-frequency waveform data and low-frequency discrete data. The high-frequency waveform data includes electrocardiogram (ECG) data, invasive arterial pressure data, and blood oxygen saturation waveform data. The low-frequency discrete data includes blood gas analysis lactate levels, creatinine data, and bilirubin data.

[0043] For the acquired high-frequency waveform data, the multimodal physiological signal acquisition and preprocessing module 100 integrates a wavelet transform filter. This wavelet transform filter is configured to denoise the high-frequency waveform data. Specific denoising operations include removing baseline drift, electromyographic interference, and power frequency interference to preserve the nonlinear dynamic characteristics of the signal.

[0044] For the acquired low-frequency discrete data, the multimodal physiological signal acquisition and preprocessing module 100 uses a cubic spline interpolation algorithm. This cubic spline interpolation algorithm maps the low-frequency discrete data onto a time axis that is consistent with the high-frequency waveform data, thereby filling the sampling gaps and forming a continuous time series with a uniform sampling rate.

[0045] The multimodal physiological signal acquisition and preprocessing module 100 further standardizes all channel data processed above. In this embodiment, the Z-score normalization method is used. After standardization, the multimodal physiological signal acquisition and preprocessing module 100 generates an aligned multidimensional synchronous time series matrix and outputs this multidimensional synchronous time series matrix to the adaptive dynamic feature extraction module 200.

[0046] See attached document Figure 1 The system for predicting the evolution path of multiple organ failure in critically ill patients further includes an adaptive dynamic feature extraction module 200. The adaptive dynamic feature extraction module 200 is connected to the output of the multimodal physiological signal acquisition and preprocessing module 100 and is configured to receive an aligned multidimensional synchronous time series matrix. The main function of the adaptive dynamic feature extraction module 200 is to map the one-dimensional time series to a high-dimensional phase space to recover the attractor structure of the system. This process includes two main computational steps: dynamic phase space reconstruction and local dissipation rate calculation in phase space.

[0047] The adaptive dynamics feature extraction module 200 first performs a dynamic phase space reconstruction operation. During this operation, the adaptive dynamics feature extraction module 200 does not use pre-fixed embedding dimension parameters, but instead employs a dynamic adjustment strategy. This module is configured to receive parameters from the closed-loop feedback controller 400, namely the total in-degree coupling strength. This strategy is based on the following physical logic: when a specific organ, as a subsystem, is strongly driven by other external organs, its own dynamic degrees of freedom are compressed, thus manifesting as a reduction in the system's dimensionality.

[0048] Based on the above logic, the adaptive dynamics feature extraction module 200 reads the total in-degree coupling strength calculated for the current organ at the previous time step. If the total in-degree coupling strength increases, the module reduces the embedding dimension at the current time step using a logarithmic decay function. The adaptive dynamics feature extraction module 200 calculates the embedding dimension at the current time step according to the following formula. Embedding dimension of each organ node

[0049] ;

[0050] in, express Time of the first Embedding dimension of each organ node; This represents the preset minimum embedding dimension threshold, used to prevent information loss due to excessively low dimensions. This represents the preset baseline embedding dimension, corresponding to the system degrees of freedom when there is no external coupling; This represents the adjustment coefficient, used to control the rate of decay; Represents the natural logarithm operation; express The feedback is always coming back. Total in-degree coupling strength of each organ node; This indicates a floor operation. Through this mechanism, the adaptive dynamic feature extraction module 200 can adjust the dimensional parameters in the signal reconstruction process in real time to match the current controlled state of the organ.

[0051] After completing the phase space reconstruction, the adaptive dynamics feature extraction module 200 further performs local dissipation rate calculation in the phase space to evaluate the degree of physical response of the organ to external drives. In the reconstructed high-dimensional phase space, this module uses the k-nearest neighbor algorithm to track the local evolution of trajectory points and estimates the local Jacobian matrix of the trajectory points accordingly.

[0052] The adaptive dynamics feature extraction module 200 quantifies the divergence of phase volume by calculating the trace of the local Jacobian matrix. This divergence, as a physical quantity, reflects the volume change trend of the system in phase space. Specifically, if the calculated divergence is negative, it indicates that the phase volume converges, corresponding to the system being in a stable dissipative response state; if the calculated divergence is positive, it indicates that the phase volume expands, corresponding to the system being in a divergent or unstable state.

[0053] Based on the calculated divergence value, the adaptive dynamics feature extraction module 200 uses a sigmoid activation function to map the divergence to a normalized numerical value, which is defined as the physical effectiveness coefficient. Physical efficiency coefficient A higher value indicates that the current signal change in the organ has a clear physical attractor characteristic, rather than random noise. The adaptive dynamic feature extraction module 200 will calculate the physical effectiveness coefficient. The output is sent to the subsequent causal topology analysis module 300 under physical dissipation constraints.

[0054] See attached document Figure 1 The system for predicting the evolution path of multiple organ failure in critically ill patients further includes a causal topology analysis module 300 under physical dissipation constraints. This module 300 is connected to the output of the adaptive dynamics feature extraction module 200 and is configured to construct a directed weighted network between organ nodes. This module uses the physical validity coefficients calculated by the adaptive dynamics feature extraction module 200 as constraints to correct the information flow at the statistical level.

[0055] The causal topology analysis module 300 under physical dissipation constraints first performs information flow calculation operations. This module uses kernel density estimation to calculate the raw transfer entropy between all monitored organ node pairs. The raw transfer entropy is used to quantify the information gain value between the source organ node and the target organ node.

[0056] Subsequently, the causal topology analysis module 300 under physical dissipation constraints performs a physical constraint correction operation. This operation verifies the effectiveness of information transmission by introducing the physical response state of the target organ node. Specifically, the causal topology analysis module 300 under physical dissipation constraints forcibly multiplies the original transmission entropy value from the source organ node to the target organ node by the physical effectiveness coefficient of the target organ node. The causal topology analysis module 300 under physical dissipation constraints calculates the physical constraint transmission entropy according to the following formula:

[0057] ;

[0058] in, Indicates in The source organ node after being corrected by physical constraints at every moment Pointing to the target organ node The propagation entropy value; Indicates in The source organ node calculated at each time step Pointing to the target organ node The original transfer entropy value Indicates in target organ node at any time The physical effectiveness coefficient.

[0059] Through the aforementioned computational logic, the causal topology analysis module 300 under physical dissipation constraints can identify and suppress non-physically driven associations. When the physical validity coefficient of the target organ node is low, i.e., when the target organ node does not exhibit dissipative convergence response at the physical level, even if statistical calculations show a high original transfer entropy between the two organs, the calculated physical constraint transfer entropy will be forcibly decayed. This mechanism eliminates statistically spurious connections, thereby outputting a physical constraint transfer entropy matrix filtered by physical laws.

[0060] Based on the generated physical constraint transfer entropy matrix, the causal topology analysis module 300 under physical dissipation constraints constructs a multi-organ coupling network graph in real time. In this multi-organ coupling network graph, nodes represent various monitored organs, and the weights of the directed edges connecting nodes are determined by the corresponding physical constraint transfer entropy values. The causal topology analysis module 300 under physical dissipation constraints transmits the constructed network graph data to the closed-loop feedback controller 400 and the cascaded failure path inference module 500.

[0061] Continue to refer to the appendix Figure 1The system for predicting the evolution path of multiple organ failure in critically ill patients also includes a closed-loop feedback controller 400. The closed-loop feedback controller 400 is connected to the output of the causal topology analysis module 300 under physical dissipation constraints and the control input of the adaptive dynamic feature extraction module 200, and is configured to realize a feedback adjustment loop from the information layer to the physical layer.

[0062] The closed-loop feedback controller 400 first performs the total in-degree coupling strength calculation operation. This module receives multi-organ coupling network data, including the physical constraint transfer entropy between organ nodes, generated by the causal topology analysis module 300 under physical dissipation constraints. For each organ node in the network, the closed-loop feedback controller 400 calculates the sum of the weights of all directed edges pointing to that node. The closed-loop feedback controller 400 calculates the weight of the first in-degree coupling strength using the following formula. Individual organ nodes Total in-degree coupling strength at time:

[0063] ;

[0064] in, Indicates the first Individual organ nodes Total in-degree coupling strength at time 1; Indicates in Constant time from source organ node Pointing to the target organ node The physical constraints transmit entropy values; summation symbol This means for all items not equal to source organ node Perform cumulative calculations.

[0065] Subsequently, the closed-loop feedback controller 400 executes feedback control operations. This module delays the calculated total in-degree coupling strength value by one time step. The closed-loop feedback controller 400 feeds this delayed parameter back to the dynamic phase space reconstruction unit in the adaptive dynamics feature extraction module 200. This feedback parameter is configured as the basis for the adaptive dynamics feature extraction module 200 to determine the embedding dimension parameter in the next time step. Through this feedback mechanism, the system establishes a bidirectional constraint structure, where network topology data is used to define the dimension parameters of physical layer signal reconstruction, and the reconstruction features of the physical layer are, in turn, used to verify the effectiveness of the network topology. This closed-loop iterative processing method enables the system to dynamically adjust the analysis parameters according to the coupling state between organs when processing complex data in an intensive care environment.

[0066] Continue to refer to the appendix Figure 1The system for predicting the evolution path of multiple organ failure in critically ill patients further includes a cascade failure path deduction module 500. The cascade failure path deduction module 500 is connected to the output of the causal topology analysis module 300 under physical dissipation constraints and is configured to perform pathological evolution prediction based on the generated physical constraint transfer entropy matrix and multi-organ coupling network graph.

[0067] The cascaded failure path deduction module 500 first performs a driver source identification operation. This module aims to identify the dominant pathological driver node from the multi-organ coupling network. The cascaded failure path deduction module 500 achieves this identification process by calculating the net outflow entropy of each organ node in the network. The net outflow entropy is used to quantify the difference between the external output information flow and the external input information flow received by an organ node. The cascaded failure path deduction module 500 calculates the first... Individual organ nodes Net outflow entropy at time:

[0068] ;

[0069] in, Indicates the first Individual organ nodes Net outflow entropy at any given moment; Indicates the first Each organ node points to all other nodes in the network. The sum of the physical constraints and the entropy, that is, the total outflow of information; Represents all other nodes in the network. Pointing to the The sum of the physical constraint transmission entropy of each organ node is the total inflow of information. The cascade failure path inference module 500 compares the net outflow entropy values ​​of all organ nodes and determines the node with the largest positive net outflow entropy value as the current pathological driving source node.

[0070] After identifying the pathological driving source node, the cascade failure path deduction module 500 performs a maximum probability path search operation. This module treats the multi-organ coupled network as a directed weighted graph, where the weights of edges are determined by the physical constraint propagation entropy. The cascade failure path deduction module 500 employs a maximum probability path search algorithm, starting from the identified pathological driving source node, and searches the network topology for propagation paths originating from that starting point. This search process prioritizes directed edges with higher connection weights to find the node sequence chain with the highest cumulative transition probability. This chain represents the cascade transmission order of pathological states in the organ network.

[0071] The cascading failure path deduction module 500 finally performs the output presentation operation. This module converts the calculated cascading failure path data into a visual signal. This visual signal is configured to drive the display terminal to present an evolution path diagram on the interactive interface. The diagram clearly indicates the transmission sequence of organ failure and the estimated time window calculated based on the path weights. Through this output, the system shows medical personnel the complete pathological evolution process from the pathological driving source, propagating through high-weighted coupling edges to downstream damaged organs.

[0072] See attached document Figure 2 This invention provides a method for predicting the evolution path of multiple organ failure in critically ill patients, which may include the following steps:

[0073] Step S1, System Initialization. In the initial stage of the system's monitoring program startup, for each organ node in the monitored network, the system sets the embedding dimension parameter used for phase space reconstruction at the initial time to a pre-configured baseline value. Simultaneously, the system initializes the total in-degree coupling strength parameter of all organ nodes to zero at the initial time. This initialization operation establishes the starting state of the system's bidirectional constraint closed-loop iterative operation, ensuring a unified baseline for subsequent dynamic analysis.

[0074] Step S2, sliding window data processing. The system moves the data capture window along the time axis according to preset time step parameters to achieve segmented processing of continuous physiological signals. Within each updated time window, the system continuously acquires multimodal physiological data sequences through the data interface. The system standardizes the captured high-frequency waveform data and the interpolated low-frequency discrete data to generate a time-aligned multidimensional synchronous time series matrix. This matrix is ​​transmitted as real-time input data to the next-level physical layer analysis module.

[0075] Step S3, First-level Constraint: Adaptive Feature Reconstruction. This step is located in the physical layer analysis stage of the system and aims to dynamically adjust signal processing parameters based on feedback signals. The system reads the total in-degree coupling strength parameter generated by the feedback at the previous time step. The system adaptively adjusts the embedding dimension at the current time step based on this parameter. Specifically, for organ nodes affected by high total in-degree coupling strength, the system determines that they are in a state of forced vibration, and the dynamic degrees of freedom are compressed. Therefore, the system reduces the value of its embedding dimension through logarithmic decay logic. Based on the adjusted embedding dimension, the system reconstructs the phase space of the time series. In the reconstructed high-dimensional phase space, the system tracks the local evolution of trajectory points and calculates the divergence of the phase volume. The system uses an activation function to map the calculated divergence into a physical validity coefficient, which is used to quantify the significance of the physical attractor features exhibited by the current organ signal, i.e., to distinguish whether the signal has a deterministic physical response or random noise.

[0076] Step S4, Second Layer Constraint: Physical Verification of Causality. This step is located in the information layer analysis phase of the system and aims to filter statistically spurious causal connections using the evaluation results of the physical layer. The system first uses kernel density estimation to calculate the original transfer entropy values ​​between all organ node pairs, which characterize the information gain strength of the source organ to the target organ. Subsequently, the system executes physical constraint correction logic, weighting the original transfer entropy using the physical validity coefficient generated in step S3. Specifically, the system multiplies the original transfer entropy value from the source organ to the target organ by the physical validity coefficient of the target organ. If the physical validity coefficient of the target organ is low, it indicates that the organ does not exhibit a clear physical dissipation response, and the system will forcibly attenuate the connection strength of this path. Finally, the system generates a physical constraint transfer entropy matrix based on the corrected values, thereby eliminating spurious connections that, although statistically significant, lack physical response support.

[0077] Step S5: Closed-loop iteration and result output. Based on the physical constraint transfer entropy matrix generated in step S4, the system updates and calculates the total in-degree coupling strength of each organ node. This calculation is achieved by accumulating the physical constraint transfer entropy values ​​pointing to that node. The system delays the updated total in-degree coupling strength value by one time step and feeds it back to step S3 as the control parameter for determining the embedding dimension in the next time step, thus forming a bidirectional constraint closed-loop control loop between the physical layer and the information layer. Simultaneously, the system constructs a multi-organ coupling network topology using the current physical constraint transfer entropy matrix. In this topology, the system identifies the node with the largest net outflow entropy as the pathological driving source and searches for the highest probability propagation path, finally generating an evolution path prediction result on the interactive interface, including the cascading failure sequence and the expected time window.

[0078] See attached document Figure 1 and attached Figure 2 One embodiment of the present invention illustrates the specific operation of the system for predicting the evolution path of multiple organ dysfunction syndrome in critically ill patients in a specific clinical scenario. This embodiment selects the monitoring scenario of severe pneumonia complicated by early septic shock. At this time, the patient is in the physiological compensation period, and routine vital sign monitoring data show that blood pressure and heart rate parameters are within the standard reference range. However, the system identifies abnormal coupling characteristics within the organ network through microdynamic analysis.

[0079] The multimodal physiological signal acquisition and preprocessing module 100 first acquires the patient's electrocardiogram waveform, invasive arterial pressure waveform, and blood oxygen saturation data. In the physical layer sensing stage, the adaptive dynamic feature extraction module 200 processes the heart rate variability signal. This module detects a critical slowing characteristic in the cardiac signal trajectory within the reconstructed phase space, and the calculated phase volume divergence value shows a clear convergence trend, indicating a high dissipation state. Based on this physical characteristic, the adaptive dynamic feature extraction module 200 generates a cardiac physical effectiveness coefficient close to one, indicating that the cardiac subsystem is currently in a controlled state driven by strong external forces, rather than a random fluctuation state caused by its own regulatory mechanisms.

[0080] The system then proceeds to the information layer analysis phase, where the causal topology analysis module 300, under physical dissipation constraints, performs transfer entropy calculation. The calculation results show that the information flow intensity from the lung node to the heart node is significantly higher than the baseline level. This module uses the physical validity coefficient generated by the adaptive dynamics feature extraction module 200 to weight and verify this transfer entropy value. Since the physical validity coefficient of the target node, i.e., the heart node, is high, the system confirms that the connection has clear physical dissipation response support, thus determining that the drive from the lung node to the heart node is a genuine pathological drive, rather than statistical noise. Based on this, the causal topology analysis module 300, under physical dissipation constraints, identifies the lung node as the pathological driving source node in the current organ network.

[0081] Based on this, the system performs closed-loop feedback adjustment. The closed-loop feedback controller 400 reads the network topology data and confirms that the heart node is subjected to high-in-degree coupling from the lung node. In the next calculation cycle, the closed-loop feedback controller 400 sends a control command to the adaptive dynamic feature extraction module 200 to automatically reduce the embedding dimension parameter used for heart signal reconstruction. The technical effect of this parameter adjustment operation is that by reducing the degrees of freedom to match the physical state of the heart under forced vibration, background noise generated by the heart's own regulation is filtered out, allowing the subsequent analysis process to focus on the pathological signal components driven by the lungs.

[0082] Finally, the cascade failure path inference module 500 generates predicted paths based on a defined coupled network. The evolution path output by this module shows a cascade sequence in which lung inflammation leads to increased right ventricular afterload, subsequently causing obstruction of hepatic venous return. The system marks the failure propagation chain from the lung node through the heart node to the liver node and the expected time window on the interactive interface. This output suggests to clinicians that although current liver function biochemical indicators are normal, right ventricular dysfunction already constitutes a definite pre-existing risk of future liver failure, thus supporting medical personnel to implement fluid management and anti-inflammatory interventions targeting the right ventricular component before liver function indicators deteriorate.

[0083] See attached documentFigure 1 and attached Figure 2 An embodiment of the present invention further illustrates the hardware deployment architecture of the system for predicting the evolution path of multiple organ failure in critically ill patients. This architecture employs a physical implementation method that combines edge computing and cloud collaboration, aiming to adapt to different levels of medical resource environments and optimize data processing efficiency.

[0084] In this hardware architecture, the edge computing terminal is deployed within the expansion module of the bedside monitor. This edge computing terminal physically integrates a multimodal physiological signal acquisition and preprocessing module 100 and an adaptive dynamic feature extraction module 200. The edge computing terminal is configured to directly execute the sliding window data processing step S2 and the first-level constraint, i.e., the adaptive feature reconstruction step S3, in the method flow. Specifically, the edge computing terminal performs wavelet transform filtering and denoising on the acquired high-frequency physiological waveform data locally, and performs phase space reconstruction operations and local divergence calculations. Deploying these processing steps involving dense matrix operations of high-frequency waveforms at the local edge avoids the transmission of massive amounts of raw waveform data over the network, thereby reducing the pressure on network bandwidth while ensuring patient data privacy.

[0085] The cloud server or central department workstation constitutes the cloud processing center of this architecture. This cloud processing center physically deploys a causal topology analysis module 300 under physical dissipation constraints, a closed-loop feedback controller 400, and a cascaded failure path inference module 500. The cloud processing center is configured to receive low-dimensional feature parameters uploaded from the edge computing terminal. These parameters include physical validity coefficients and dimensionality-reduced feature vectors, rather than the original waveform data. The cloud processing center centrally executes the second layer of constraints in the method flow, namely the physical verification causal step S4 and the closed-loop iteration and result output step S5. Utilizing the parallel computing capabilities of the cloud server, the cloud processing center can quickly process the transfer entropy calculation of the fully connected network matrix, realizing real-time construction and path inference of multi-organ coupled networks.

[0086] A bidirectional communication link is established between the edge computing terminal and the cloud processing center via an encrypted medical IoT protocol. This communication link is used not only to upload low-dimensional feature parameters but also to transmit closed-loop control commands. Specifically, the total in-degree coupling strength parameter calculated by the cloud processing center is transmitted back to the edge computing terminal with low latency through this communication link. The adaptive dynamic feature extraction module 200 of the edge computing terminal receives this feedback parameter and uses it as the basis for adjusting the embedding dimension in the next moment, thereby realizing bidirectional constraint closed-loop control between the physical layer and the information layer at the hardware level.

Claims

1. A system for predicting the progression of multiple organ failure in critically ill patients, characterized in that, include: A multimodal physiological signal acquisition and preprocessing module is used to output a standardized multidimensional synchronous time series matrix; An adaptive dynamics feature extraction module is used to perform phase space reconstruction on the multidimensional synchronous time series matrix to extract physical layer dynamics features and output physical effectiveness coefficients; A causal topology analysis module under physical dissipation constraints is used to construct a multi-organ coupling network representing the interaction relationship of the information layer using the physical effectiveness coefficient; The cascading failure path deduction module is used to deduce the cascading propagation sequence of organ failure based on the multi-organ coupling network and generate evolution path prediction results. A closed-loop feedback controller is used to extract the total in-degree coupling strength from the multi-organ coupling network and provide feedback. The adaptive dynamic feature extraction module responds to the feedback of the total in-degree coupling strength and dynamically adjusts the embedding dimension parameters of the phase space reconstruction to form a bidirectional constraint closed loop between the physical layer dynamic feature extraction and the information layer interaction relationship construction.

2. The system for predicting the evolution path of multiple organ failure in critically ill patients according to claim 1, characterized in that, The multimodal physiological signal acquisition and preprocessing module uses wavelet transform filters to remove baseline drift, electromyographic interference, and power frequency interference from the acquired high-frequency waveform data. For the collected low-frequency discrete data, a cubic spline interpolation algorithm is used to map the low-frequency discrete data onto a time axis that is consistent with the high-frequency waveform data, filling the sampling gaps and forming a continuous time series; All channel data were standardized using the Z-score standardization method.

3. The system for predicting the evolution path of multiple organ failure in critically ill patients according to claim 1, characterized in that, The adaptive dynamics feature extraction module calculates the physical effectiveness coefficient in the following way: In the reconstructed phase space, the local Jacobian matrix of the trajectory points is estimated using the k-nearest neighbor algorithm. The trace of the local Jacobian matrix is ​​calculated to quantify the divergence of the phase volume. The divergence is then mapped to a normalized value as the physical effectiveness coefficient using the sigmoid activation function.

4. The system for predicting the evolution path of multiple organ failure in critically ill patients according to claim 1, characterized in that, The strategy for the adaptive dynamic feature extraction module to dynamically adjust the embedding dimension parameters is as follows: Read the total in-degree coupling strength fed back from the previous moment; set a preset baseline embedding dimension; calculate the product of the natural logarithm of the total in-degree coupling strength and the adjustment coefficient; subtract the product from the baseline embedding dimension to obtain the calculated value; perform a floor operation on the calculated value; compare the operation result with a preset minimum embedding dimension threshold; and select the larger value between the operation result and the minimum embedding dimension threshold as the embedding dimension parameter at the current moment.

5. A system for predicting the evolution path of multiple organ failure in critically ill patients according to claim 1, characterized in that, The causal topology analysis module under the physical dissipation constraint constructs the multi-organ coupled network in the following way: The kernel density estimation method is used to calculate the original transfer entropy between monitoring organ node pairs. The original transfer entropy value from the source organ node to the target organ node is multiplied by the physical effectiveness coefficient of the target organ node to obtain the physical constraint transfer entropy. The weight of the directed edge connecting the nodes is determined based on the physical constraint transfer entropy.

6. A system for predicting the evolution path of multiple organ failure in critically ill patients according to claim 5, characterized in that, The closed-loop feedback controller extracts the total in-degree coupling strength in the following way: For each organ node in the multi-organ coupling network, the physical constraint propagation entropy of all source organ nodes pointing to the organ node is accumulated to obtain the total in-degree coupling strength of the organ node at the current time.

7. A system for predicting the evolution path of multiple organ failure in critically ill patients according to claim 1, characterized in that, The closed-loop feedback controller delays the calculated total in-degree coupling strength value by one time step, and sends the delayed total in-degree coupling strength as a feedback parameter to the adaptive dynamic feature extraction module.

8. A system for predicting the evolution path of multiple organ failure in critically ill patients according to claim 5, characterized in that, The cascaded failure path deduction module identifies pathological driving source nodes by calculating the net outflow entropy of each organ node in the multi-organ coupling network. The cascade failure path deduction module calculates the net outflow entropy as follows: for any organ node in the multi-organ coupling network, the sum of the physical constraint transmission entropies from the organ node to all other nodes in the network is calculated as the total outflow information, the sum of the physical constraint transmission entropies from all other nodes in the network to the organ node is calculated as the total inflow information, and the net outflow entropy of the organ node is obtained by subtracting the total inflow information from the total outflow information. The cascade failure path deduction module selects the node with the largest positive net outflow entropy value as the pathological driving source node.

9. A prediction system for the evolution path of multiple organ failure in critically ill patients according to claim 8, characterized in that, The cascading failure path deduction module employs a maximum probability path search algorithm. Starting from the identified pathological driving source node, it searches for a propagation path in the network topology that originates from the starting point. The propagation path consists of a chain of node sequences with the highest cumulative transfer probability, which is used to characterize the cascading transmission order of the pathological state.

10. A prediction system for the evolution path of multiple organ failure in critically ill patients according to claim 1, characterized in that, An implementation architecture that combines edge computing and cloud collaboration is adopted; The multimodal physiological signal acquisition and preprocessing module and the adaptive dynamic feature extraction module are integrated into the edge computing terminal for performing data processing and calculating the physical effectiveness coefficient locally. The causal topology analysis module under physical dissipation constraints, the closed-loop feedback controller, and the cascaded failure path inference module are deployed in a cloud processing center. The edge computing terminal is connected to the cloud processing center via an encrypted communication link, which is used to transmit the physical validity coefficient, the dimensionality-reduced feature vector, and the feedback of the total in-degree coupling strength.

Images