A method and system for multi-modal data fusion evaluation and prognosis prediction of septic cardiomyopathy

Abstract

The application provides a multi-modal data fusion evaluation and prognosis prediction method and system for septic cardiomyopathy, which comprises the following steps: obtaining target data and electronic medical records of septic cardiomyopathy patients, combining the target data with a preset first database to construct a correlation matrix; constructing a second database corresponding to the septic cardiomyopathy patients according to the electronic medical records; generating a target interaction path based on the second database; obtaining a first signal according to the causal effect intensity of each node in the target interaction path, and generating a prognosis prediction scheme in combination with a target risk threshold. The application realizes real-time tracking and closed-loop intervention of the pathological process of septic cardiomyopathy, and fundamentally solves the prognosis prediction deviation caused by the time and space disconnection of the traditional database.

Description

Technical Field

[0001] This application relates to the field of multimodal data fusion technology, and in particular to a multimodal data fusion assessment and prognostic prediction method and system for septic cardiomyopathy. Background Technology

[0002] Septic cardiomyopathy, as one of the most critical complications of sepsis, involves a multi-dimensional interaction of systemic inflammatory response, microcirculatory disturbances, and myocardial energy metabolism disorders. Current clinical assessment and prognostic prediction methods heavily rely on static analysis of single-modal data, while the spatiotemporal disconnect of multimodal data has become a core bottleneck restricting precision diagnosis and treatment.

[0003] To address the aforementioned needs, existing technologies propose a multimodal data integration system based on a fixed time window. This system uses a pre-set unified time grid to slice the data of each modality over time, and calculates the mean protein concentration, mean ultrasound parameters, and mean hemodynamic parameters within each window. It then uses statistical correlation analysis to generate association rules.

[0004] Existing solutions of this kind have some drawbacks. For example, the forced alignment of fixed time windows leads to the loss of temporal correlation of key events across windows; static thresholds cannot adapt to the dynamic evolution characteristics of sepsis myocardial injury, which can easily cause false alarms or missed alarms. Summary of the Invention

[0005] This application provides a multimodal data fusion assessment and prognostic prediction method and system for septic cardiomyopathy, which solves the problem of loss of temporal correlation of key events across windows caused by forced alignment of fixed time windows in the prior art.

[0006] In a first aspect, this application provides a multimodal data fusion assessment and prognostic prediction method for septic cardiomyopathy, including: Acquire target data and electronic medical records of patients with septic cardiomyopathy. The target data includes the analysis results corresponding to troponin I and the target complex, respectively. Based on the analysis results and the pre-defined first database, an association matrix is ​​constructed; The labels for the sepsis stage, echocardiographic data, and hemodynamic data in the electronic medical record were aligned to obtain multimodal data; Multimodal data are fused according to preset processing rules to construct a second database corresponding to patients with septic cardiomyopathy; Based on the second database, using a Bayesian inference model, specific biomarkers of sepsis-induced myocardial injury and clinical parameters in multimodal data are analyzed to generate target interaction paths; Based on the causal effect intensity of each node in the target interaction path, the target point is determined. Based on the first signal corresponding to the target point and combined with the target risk threshold, a prognostic prediction scheme is generated. The first signal is an intelligent alarm signal used to provide early risk warning and guide targeted intervention before circulatory failure occurs.

[0007] Optionally, based on the analysis results and a pre-defined first database, an association matrix is ​​constructed, including: The baseline abnormal threshold and dynamic fluctuation range of the second signal within a preset time window are obtained from the preset first database. The second signal is the sequence data of the change of troponin I concentration over time. Based on the analysis results, the signal intensity change rate of the second signal corresponding to the preset time window is calculated. When the signal intensity change rate exceeds the dynamic fluctuation range, the second signal is marked as an abnormal signal. The target feature pattern corresponding to the abnormal signal is extracted from the preset first database. The target feature pattern is a reference fluctuation curve used to dynamically match with the abnormal signal and verify the temporal correlation between inflammation outbreak and myocardial injury. The overlap between the first time window of the abnormal signal and the second time window of the target feature pattern is calculated. When the overlap is greater than the first preset threshold, the target extreme value of the target complex within the overlapping window is obtained. Based on the target extreme value and the benchmark abnormal threshold, a dynamic adjustment coefficient is generated. The target extreme value is used to quantify the maximum rate of change of inflammatory factors within the overlapping window. Based on the dynamic adjustment coefficient, the baseline abnormal threshold is adjusted to generate a target abnormal threshold that matches the analysis results. The target abnormal threshold is the critical value for abnormal judgment after the dynamic adjustment coefficient is corrected, and it adapts to the patient's pathological state in real time. Based on the target anomaly threshold and the target extreme value, the preset time window, the first time window, and the second time window are aligned to generate the correlation matrix between the second signal and the target complex.

[0008] Secondly, this application provides a multimodal data fusion assessment and prognostic prediction system for septic cardiomyopathy, including: The acquisition module is used to acquire target data and electronic medical records of patients with septic cardiomyopathy. The target data includes the analysis results corresponding to troponin I and the target complex, respectively. The module is used to construct an association matrix based on the analysis results and a pre-defined first database; The processing module is used to align the labels for the sepsis stage, echocardiographic data, and hemodynamic data in the electronic medical record to obtain multimodal data. The fusion module is used to perform feature fusion processing on multimodal data according to preset processing rules in order to construct a second database corresponding to patients with septic cardiomyopathy. The inference module is used to analyze specific biomarkers of sepsis-related myocardial injury and clinical parameters in multimodal data based on a second database and using a Bayesian inference model to generate target interaction paths. The early warning module is used to determine the target point based on the causal effect intensity of each node in the target interaction path. Based on the first signal corresponding to the target point and combined with the target risk threshold, a prognostic prediction scheme is generated. The first signal is an intelligent alarm signal used to provide early risk warning and guide targeted intervention before circulatory failure occurs.

[0009] Thirdly, this application provides a computing device including a processor and a memory, wherein the memory stores a computer program, and the processor is configured to run the computer program to execute a multimodal data fusion assessment and prognostic prediction method for septic cardiomyopathy according to any one of the first aspects.

[0010] Fourthly, this application provides a computer storage medium storing computer program instructions thereon, which, when executed by a processor, implement a multimodal data fusion assessment and prognostic prediction method for septic cardiomyopathy, as described in any one of the first aspects.

[0011] In this application, target data and electronic medical records of patients with septic cardiomyopathy are acquired. The target data includes the analysis results corresponding to troponin I and the target complex, respectively. Based on the analysis results and a pre-set first database, an association matrix is ​​constructed. The labels of the sepsis stage, echocardiographic data, and hemodynamic data in the electronic medical records are aligned to obtain multimodal data. The multimodal data is fused according to pre-set processing rules to construct a second database corresponding to patients with septic cardiomyopathy. Based on the second database, a Bayesian inference model is used to analyze the specific biomarkers of septic myocardial injury and the clinical parameters in the multimodal data to generate a target interaction path. According to the causal effect strength of each node in the target interaction path, the target point is determined. Based on the first signal corresponding to the target point and combined with the target risk threshold, a prognostic prediction scheme is generated. The first signal is an intelligent alarm signal used to provide early risk warning and guide targeted intervention before the occurrence of circulatory failure.

[0012] The technical solution provided in this application enables the comprehensive collection of biomarkers and clinical parameters from patients with septic cardiomyopathy, providing fundamental data for multimodal analysis. It overcomes the limitations of traditional fixed-time alignment, resolving the spatiotemporal disconnection problem caused by differences in sampling frequency in multimodal data, generating a time-consistent multidimensional dataset, and breaking through the limitations of traditional correlation analysis. Furthermore, it performs tensor multiplication operations on the association matrix with the first intensity value and key event points in the multimodal data to generate a first sequence; it generates a coupling strength curve by dynamically adjusting coefficients to nonlinearly scale the first intensity value; it extracts extreme points during the inflammatory outbreak period and matches them with key event points based on time deviation, constructing an association network between parent and child nodes, and optimizing conditional probabilities to build a causal inference database.

[0013] This solution addresses the issue of lost real-time correlation of cross-modal events caused by fixed time window averaging in existing solutions. Through dynamic scoring sequences and nonlinear scaling mechanisms, it accurately captures the temporal coupling relationship between myocardial injury, inflammatory outbreaks, and hemodynamic deterioration. Based on the association network constructed by time deviation matching, it optimizes the causal weight allocation of biomarkers and clinical parameters, improves the accuracy of causal inference of sepsis myocardial injury, and provides traceable pathological mechanism support for prognostic early warning.

[0014] These or other aspects of this application will become more apparent in the following description of the embodiments. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 A flowchart illustrating a multimodal data fusion assessment and prognostic prediction method for septic cardiomyopathy provided in this application embodiment; Figure 2 A schematic diagram of the structure of a multimodal data fusion assessment and prognostic prediction system for septic cardiomyopathy provided in an embodiment of this application; Figure 3 This is a schematic diagram of the structure of a computing device provided in an embodiment of this application. Detailed Implementation

[0017] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings.

[0018] In some of the processes described in the specification, claims, and accompanying drawings of this application, multiple operations appearing in a specific order are included. However, it should be clearly understood that these operations may not be executed in the order they appear herein, or may be executed in parallel. The operation numbers, such as 101, 102, etc., are merely used to distinguish different operations and do not themselves represent any execution order. Furthermore, these processes may include more or fewer operations, and these operations may be executed sequentially or in parallel. It should be noted that the descriptions such as "first," "second," etc., in this document are used to distinguish different messages, devices, modules, etc., and do not represent a chronological order, nor do they limit "first" and "second" to different types.

[0019] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0020] Figure 1 This application provides a flowchart of a multimodal data fusion assessment and prognostic prediction method for septic cardiomyopathy, as illustrated in the embodiments of this application. Figure 1 As shown, the method includes: To address the diagnostic and treatment bottlenecks caused by the spatiotemporal disconnection of multimodal data and static analysis in septic cardiomyopathy, this solution takes dynamic cross-modal causal reasoning as its core, breaking through the limitations of existing fixed-time-window integration systems.

[0021] First, based on a database of specific biomarkers for sepsis-induced myocardial injury, a dynamic association matrix is ​​constructed. By matching the rate of change of the second signal intensity with the target extreme value of inflammatory factors, a dynamic adjustment coefficient is generated to correct abnormal thresholds in real time, solving the problem that static thresholds cannot adapt to nonlinear fluctuations during the inflammatory outbreak period. Second, using disease stage markers as the baseline time axis, highly consistent multimodal data is generated, fundamentally overcoming the loss of cross-window event associations caused by traditional mean-based processing. Furthermore, tensor product is used to fuse cross-modal data features, and combined with the dynamic adjustment coefficient to generate the first sequence and coupling strength curve, breaking through the superficial association limitations of traditional statistical correlation analysis. Finally, based on a Bayesian causal inference model, protein and clinical interaction pathways are analyzed. Combined with multimodal collaborative validation and target risk thresholds, personalized prognostic plans are dynamically generated, realizing a closed-loop pathway from molecular mechanisms to clinical decision-making, and completely solving the problems of misdiagnosis and mistreatment caused by data silos, static rules, and causal breaks in existing technologies.

[0022] Based on this, this application provides a multimodal data fusion assessment and prognostic prediction method for septic cardiomyopathy, such as... Figure 1 ,include: Step 101: Obtain the target data and electronic medical records of patients with septic cardiomyopathy. The target data includes the analysis results corresponding to troponin I and the target complex, respectively.

[0023] In this step, the target data refers to the serum protein concentration data detected by mass spectrometry. Troponin I is a specific biomarker for cardiomyocyte injury, and its concentration can be used to reflect the degree of myocardial necrosis. The target complex refers to a complex biomarker composed of multiple pro-inflammatory cytokines, used to quantify the intensity of the systemic inflammatory response. The analytical result refers to the concentration value obtained by converting the protein ion peak intensity data detected by mass spectrometry through a standard curve.

[0024] In this embodiment of the application, the analysis results of troponin I and the target complex are obtained. The target complex includes markers such as interleukin-6 and α-tumor necrosis factor, which are used to characterize the intensity of the systemic inflammatory response. At the same time, the patient's sepsis stage label, echocardiographic data and hemodynamic data are extracted from the hospital's electronic medical record system.

[0025] Step 102: Based on the analysis results and the preset first database, construct the association matrix.

[0026] In this step, the analysis result refers to the concentration value calculated from the protein ion peak intensity data detected by mass spectrometry using a standard curve. The pre-defined first database is a structured database containing the dynamic fluctuation range, target extreme values, and time window association rules of biomarkers related to myocardial injury in sepsis patients. The association matrix is ​​a matrix reflecting the dynamic association strength between troponin I and inflammatory factors within different time windows; the weight values ​​are calculated from the time window overlap and the proportion of target extreme values.

[0027] In this embodiment, the rate of change of troponin I concentration in the analysis results is matched with the dynamic fluctuation range recorded in a preset first database. When the rate of change of troponin I concentration exceeds the upper limit of the corresponding time window fluctuation in the database, it is marked as an abnormal signal, and the target extreme value of the target complex within the time window is extracted. By calculating the overlap between the time window of the abnormal signal and the time window of the characteristic pattern of inflammatory factors, if the overlap ratio exceeds a first preset threshold, a dynamic adjustment coefficient is generated based on the ratio of the target extreme value to the baseline abnormal threshold of troponin I. The dynamic adjustment coefficient is combined with the time window alignment rule to generate the correlation matrix between troponin I and inflammatory factors.

[0028] It should be noted that this embodiment does not specifically limit the size of each preset threshold.

[0029] Step 103: Align the labels for the sepsis stage, echocardiographic data, and hemodynamic data in the electronic medical record to obtain multimodal data.

[0030] In this step, the sepsis stage label refers to the disease stage label defined according to clinical guidelines, used for time alignment of multimodal data. Echocardiographic data refers to the time-series data of cardiac function parameters detected by echocardiography. Hemodynamic data refers to the real-time data stream reflecting circulatory status, such as continuously monitored cardiac output and extravascular pulmonary edema index. Multimodal data refers to the aligned heterogeneous dataset.

[0031] In this embodiment, the sepsis stage in the electronic medical record is used as the time axis reference. The echocardiogram data is interpolated according to the preset time granularity to generate a continuous time series. The hemodynamic data is filtered by moving average to eliminate noise interference and then key event points are marked. The interpolated ultrasound parameter sequence, the filtered hemodynamic data and the disease stage markers are timestamped to form multimodal data with a unified time dimension.

[0032] Step 104: Perform feature fusion processing on the multimodal data according to preset processing rules to construct a second database corresponding to patients with septic cardiomyopathy.

[0033] In this step, the preset processing rules refer to the mathematical operations and time-series matching rules defined for multimodal data fusion. This embodiment does not specifically limit the content or expression of these rules. The second database refers to a structured database that stores the causal relationships between biomarkers and clinical parameters.

[0034] In this embodiment, the row weights of the association matrix, the first intensity value corresponding to the timestamp in the multimodal data, and the density of key event points are multiplied by tensors to generate a first sequence. When the scores in the scoring sequence exceed the target abnormal threshold three times in a row, the nonlinear scaling of the first intensity value by the dynamic adjustment coefficient is triggered to generate a curve reflecting the coupling strength between myocardial injury and inflammation outbreak. The extreme points of the curve during the inflammation outbreak period are extracted and matched with the key event points by time deviation to construct the association network between parent nodes and child nodes. The conditional probability is generated by Bayesian parameter optimization to finally form a second database.

[0035] For example, when the scores in the scoring sequence exceed the target abnormal threshold three times in a row, a curve reflecting the coupling strength between myocardial injury and inflammatory outbreak is generated according to the exponential decay function. If the time difference between the extreme point and the event point is less than 1 hour, a correlation network is constructed between the troponin I threshold, the dynamic adjustment coefficient, the slope of the ultrasound feature decline, and the event point frequency. The conditional probability is generated through Bayesian parameter optimization, and finally a second database is formed.

[0036] Step 105: Based on the second database, use a Bayesian inference model to analyze specific biomarkers of sepsis-related myocardial injury and clinical parameters in multimodal data to generate a target interaction path.

[0037] In this step, the Bayesian inference model refers to a probabilistic inference tool based on Bayesian networks used to calculate the strength of the causal effect between biomarkers and clinical parameters. Specific biomarkers for sepsis-related myocardial injury refer to protein biomarkers specifically used for the diagnosis of sepsis-related myocardial injury. The target interaction path refers to the set of paths describing the causal regulatory relationship between proteins and cardiac function indicators; the path weights reflect the strength of the causal effect.

[0038] In this embodiment, based on the second database, a Bayesian inference model is used to extract the time series of target abnormal threshold triggering events and inflammatory factor concentration gradient adjustment events, and calculate their joint probability distribution with the ultrasound feature decline slope and the frequency of key event points; the causal effect strength from parent node to child node is determined by Markov chain Monte Carlo sampling, and a target interaction path is generated, wherein the path weight is determined by the product of the causal effect strength and the time window length.

[0039] Step 106: Determine the target target based on the causal effect intensity of each node in the target interaction path. Based on the first signal corresponding to the target target and combined with the target risk threshold, generate a prognostic prediction scheme. The first signal is an intelligent alarm signal used to provide early risk warnings and guide targeted interventions before circulatory failure occurs.

[0040] In this step, the causal effect strength refers to a parameter generated by multiplying the Bayesian posterior probability by the time window length, used to quantify the confidence of the causal link. Protein targets refer to key regulatory proteins in the process of sepsis-induced myocardial injury, serving as molecular targets for interventional therapy. The target risk threshold is a clinical standard defining the risk of ventricular structural abnormalities, used for prognostic decision-making. The prognostic prediction plan refers to individualized treatment recommendations generated by combining warning signals and risk thresholds.

[0041] In this embodiment of the application, protein targets are prioritized based on the causal effect intensity of each node in the target interaction path, and the top N targets are selected. The time window of the target is overlapped with the ultrasound feature decline window and the key event point dense window in the multimodal data for verification. If the verification is successful, an individualized prognostic plan including the priority of anti-inflammatory treatment and the timing of mechanical circulatory support is generated in combination with the target risk threshold.

[0042] For example, the time window of the top N target points is overlapped with the ultrasound feature decline window and the dense window of key event points in the multimodal data for verification. If the window overlap ratio exceeds 80%, the verification is successful. If the left ventricular end-diastolic diameter is ≥55mm, the order of interleukin-6 inhibitor administration and the individualized prognostic plan for the initiation of extracorporeal membrane oxygenation triggered by the decrease in cardiac output are generated.

[0043] This application's embodiments generate an association matrix by dynamically matching a biomarker database, solving the problem of spatiotemporal disconnection of multimodal data; utilize a Bayesian causal inference model to analyze the target interaction path, overcoming the limitations of traditional statistical correlation analysis; and generate prognostic plans based on the strength of causal effects and multimodal collaborative verification, shortening clinical decision delays and improving the level of precision diagnosis and treatment of septic cardiomyopathy.

[0044] For example, taking a patient with septic cardiomyopathy as an example, their serum mass spectrometry data and electronic medical records are obtained, the disease course is marked as the inflammatory outbreak period, and left ventricular ejection fraction is measured by echocardiography every 6 hours; when the concentration of troponin I rises beyond the dynamic fluctuation range within 3 hours, abnormal signals are marked and the target extreme value of interleukin-6 within the corresponding window is extracted to generate an association matrix; the disease course marking, the interpolated left ventricular ejection fraction sequence and real-time cardiac output data are aligned to form a multimodal dataset; the association weight matrix is ​​multiplied by the first intensity value to generate the first sequence; when the score exceeds the threshold three times consecutively, the dynamic adjustment coefficient is triggered to scale the ultrasound data to generate a coupling strength curve; the extreme point of the curve in the inflammatory outbreak period is extracted and matched with the event point of sudden drop in cardiac output to construct an association network; the causal path of interleukin-6 regulating the decrease in left ventricular ejection fraction is generated through Bayesian inference, and an early warning signal is generated by combining the left ventricular end-diastolic diameter exceeding the threshold, and finally the prognostic plan of anti-interleukin-6 treatment priority is output.

[0045] This application provides a specific embodiment. Step 102 involves constructing an association matrix based on the analysis results and a preset first database, specifically including the following steps: Step 201: Obtain the baseline abnormal threshold and dynamic fluctuation range of the second signal within a preset time window from the preset first database. The second signal is the sequence data of the change of troponin I concentration over time. In this step, the type of the first database is not specifically limited in this embodiment. The preset time window refers to a pre-defined fixed time interval used to divide the time units for data acquisition and analysis. The second signal refers to the sequence data of troponin I concentration changes over time as detected by mass spectrometry, reflecting the dynamics of myocardial cell damage. The baseline abnormality threshold refers to the static critical value of troponin I concentration obtained based on statistical analysis of historical sepsis patient data, used for preliminary judgment of abnormalities. The dynamic fluctuation range refers to the allowable range of troponin I concentration changes within the preset time window; exceeding this range indicates an abnormality.

[0046] In this embodiment of the application, the baseline abnormal threshold and dynamic fluctuation range of the second signal within a preset time window are extracted from a preset first database. The baseline abnormal threshold is a static critical value obtained based on the statistical analysis of historical sepsis patient data, and the dynamic fluctuation range reflects the normal fluctuation range of troponin I concentration within the time window.

[0047] For example, from a preset first database, the abnormal threshold and ±15% concentration change rate of the second signal at 0.1 ng / mL within 6 hours are extracted.

[0048] Step 202: Based on the analysis results, calculate the rate of change of the signal intensity of the second signal corresponding to the preset time window. When the rate of change of the signal intensity exceeds the dynamic fluctuation range, mark the second signal as an abnormal signal and extract the target feature pattern corresponding to the abnormal signal from the preset first database. The target feature pattern is a reference fluctuation curve used to dynamically match with the abnormal signal and verify the temporal correlation between inflammation outbreak and myocardial injury.

[0049] In this step, the second signal needs to be matched with the baseline parameters in the database. The rate of change of signal intensity refers to the magnitude of change in troponin I concentration within the current time window. An abnormal signal refers to a segment of troponin I data where the rate of change of signal intensity exceeds the dynamic fluctuation range.

[0050] In this embodiment of the application, based on the analysis results, the rate of change of the signal strength of the second signal within a preset time window is calculated. When the rate of change exceeds the upper limit of the dynamic fluctuation range, the second signal within the current time window is marked as an abnormal signal, and the target feature pattern corresponding to the abnormal signal is extracted from the database.

[0051] For example, when the rate of change of signal intensity within a preset time window increases by more than 15%, the second signal within the current time window is marked as an abnormal signal, and the concentration fluctuation curves and gradient ranges of interleukin-6 and α-tumor necrosis factor within the same time window are extracted from the database.

[0052] Step 203: Calculate the overlap between the first time window of the abnormal signal and the second time window of the target feature pattern. When the overlap is greater than the first preset threshold, obtain the target extreme value of the target complex within the overlapping window, and generate a dynamic adjustment coefficient based on the target extreme value and the benchmark abnormal threshold. The target extreme value is used to quantify the maximum rate of change of inflammatory factors within the overlapping window. In this step, the first time window refers to the time interval of the target troponin I abnormal signal. The second time window refers to the time interval of the target feature pattern. Overlap refers to the percentage of the overlap time between the first and second time windows relative to the total duration of the first time window. The first preset threshold is the overlap threshold that triggers subsequent operations. The overlapping window is the time interval at which the first and second time windows intersect. The target extreme value is the difference between the peak and trough values ​​of the target complex concentration within the overlapping window, reflecting the intensity of the inflammatory response. The dynamic adjustment coefficient is the ratio of the target extreme value to the baseline abnormal threshold, used to dynamically adjust the target abnormal threshold.

[0053] In this embodiment of the application, the overlap between the first time window of the abnormal signal and the second time window of the target feature pattern is calculated. When the overlap is greater than a first preset threshold, the target extreme value of the target complex within the overlapping window is extracted, and a dynamic adjustment coefficient is generated based on the ratio of the extreme value to the benchmark abnormal threshold.

[0054] For example, the overlap between the abnormal signal and the target feature pattern between hours 5 and 11 is calculated. When the overlap is greater than 50%, the difference between the peak and trough of interleukin-6 concentration between hours 5 and 9 is taken as the extreme value. The extreme value of 0.8 ng / mL is divided by the baseline threshold of 0.1 ng / mL to generate a dynamic adjustment coefficient of 8.

[0055] Step 204: Adjust the baseline abnormal threshold according to the dynamic adjustment coefficient, and generate a target abnormal threshold that matches the analysis results. The target abnormal threshold is the abnormal judgment threshold after the dynamic adjustment coefficient is corrected, and it adapts to the patient's pathological state in real time. In this embodiment, the dynamic adjustment coefficient is multiplied by the baseline abnormal threshold to generate a target abnormal threshold that matches the current analysis result. This target abnormal threshold dynamically adapts to changes in the concentration gradient of inflammatory factors and reflects the real-time severity of myocardial injury in sepsis.

[0056] Step 205: Based on the target anomaly threshold and the target extreme value, align the preset time window, the first time window, and the second time window to generate the correlation matrix between the second signal and the target complex; In this embodiment, the preset time window, the first time window of the abnormal signal, and the second time window of the inflammatory factor characteristic pattern are aligned as overlapping windows. For each aligned time window, the product of the target abnormal threshold and the target extreme value of the inflammatory factor is calculated as the weight value of the time window. The weight values ​​of all time windows are calculated to generate an association matrix. Each element in the matrix represents the association strength between troponin I and the inflammatory factor within the corresponding time window.

[0057] For example, the preset time window is 6 hours. The abnormal signals from the 3rd to the 9th hour and the characteristic patterns of inflammatory factors from the 5th to the 11th hour are aligned within the 5th to 9th hour. For each aligned time window unit, the product of the target abnormal threshold of 0.8 ng / mL and the target extreme value of inflammatory factors of 0.8 ng / mL is used as the weight value. The weight values ​​of all time windows are calculated to generate an association matrix.

[0058] This application's embodiments generate a correlation matrix by dynamically matching the concentration gradient of inflammatory factors with abnormal troponin I signals, thus solving the problem that traditional static thresholds cannot adapt to the dynamic evolution of myocardial injury in sepsis. Based on the dynamic adjustment mechanism of time window overlap and target extreme value, it accurately captures cross-modal correlations between biomarkers, breaking through the core bottleneck of spatiotemporal disconnection of multimodal data in existing technologies, and improving the sensitivity and specificity of anomaly detection.

[0059] This application provides a specific embodiment. Step 103 involves aligning the sepsis stage tags, echocardiographic data, and hemodynamic data in the electronic medical record to obtain multimodal data. This specifically includes the following steps: Step 301: Based on the echocardiographic data in the electronic medical record, calculate the coefficient of variation of the left ventricular ejection fraction and the instantaneous fluctuation amplitude of the overall longitudinal strain, and generate the first intensity value. The first intensity value is a composite ultrasound parameter used to quantify the degree of dynamic abnormality of cardiac function and provide core quantitative indicators of cardiac function for multimodal data fusion. In this step, the coefficient of variation of left ventricular ejection fraction (LVEF) is the standard deviation of LVEF time-series data divided by the mean, reflecting the fluctuation stability of cardiac pumping function; a larger value indicates greater instability of cardiac function. The instantaneous fluctuation amplitude of global longitudinal strain refers to the absolute difference between measurements of global longitudinal strain at adjacent time points, used to quantify the instantaneous change intensity of myocardial contractile coordination.

[0060] In this embodiment of the application, based on the echocardiogram data in the electronic medical record, the time series data of the left ventricular ejection fraction is extracted and its coefficient of variation is calculated; at the same time, the time series data of the overall longitudinal strain is extracted and its instantaneous fluctuation amplitude is calculated; the coefficient of variation of the left ventricular ejection fraction and the instantaneous fluctuation amplitude of the overall longitudinal strain are added together to generate a first intensity value, which is used to quantify the comprehensive degree of abnormality of cardiac function.

[0061] Step 302: Based on the hemodynamic data in the electronic medical record, calculate the moving average of cardiac output and the rate of change of extravascular lung water index. The time point when the rate of change exceeds the second preset threshold is marked as a critical event point. In this step, the moving average of cardiac output refers to the average of continuously monitored cardiac output data over a sliding time window, used to eliminate transient noise interference. The extravascular lung water index is a quantitative indicator of pulmonary edema obtained through continuous cardiac output monitoring indicated by pulse; a higher value indicates more severe pulmonary edema. The second preset threshold is a pre-set critical value for the rate of change of the extravascular lung water index; exceeding this value is marked as a critical event.

[0062] In this embodiment, the cardiac output in the hemodynamic data is calculated by moving average to smooth out instantaneous noise; the rate of change of the extravascular lung water index is calculated, and when the rate of change of the extravascular lung water index exceeds a second preset threshold, the current time point is marked as a critical event point, indicating a high-risk state of pulmonary edema or circulatory deterioration.

[0063] Step 303: Align the first intensity value and key event points according to the labels of the sepsis stage to generate multimodal data.

[0064] In this step, the critical event point refers to the time point when the rate of change of the extravascular lung water index exceeds a second preset threshold, used to identify high-risk moments of circulatory deterioration. Multimodal data refers to the aligned, unified time-dimension dataset.

[0065] In this embodiment, the sepsis stage is used as the baseline time axis. The first intensity value is converted into a continuous time series through linear interpolation, and the key event points are aligned with the original timestamps. The interpolated first intensity value sequence, key event point markers, and disease stage labels are integrated into a multimodal dataset with a unified time dimension.

[0066] For example, using the start time of the inflammation outbreak as the baseline time axis, the first intensity value calculated every 6 hours is converted into a data point per minute, while the key event points detected every minute are aligned with the original timestamps to generate a multimodal dataset with a unified time dimension.

[0067] This application's embodiments quantify dynamic abnormalities of cardiac function by calculating a first intensity value and combine it with real-time labeling of key hemodynamic event points, overcoming the limitations of traditional single-modal data analysis. Based on the time axis alignment mechanism of disease stage labeling, it solves the problem of spatiotemporal disconnection caused by differences in sampling frequency of multi-source data, generating a highly consistent multimodal dataset, providing a precise temporal correlation foundation for subsequent causal inference.

[0068] This application provides a specific embodiment. Step 104 involves performing feature fusion processing on the multimodal data according to preset processing rules to construct a second database corresponding to patients with septic cardiomyopathy. This specifically includes the following steps: Step 401: Perform tensor product operation on the row dimension weights of the correlation matrix with the first intensity value and key event point of the corresponding timestamp in the multimodal data to generate the first sequence.

[0069] In this step, row dimension weight refers to the weight value corresponding to each row in the association matrix, which represents the association strength between troponin I and inflammatory factors within a specific time window.

[0070] In this embodiment of the application, the row dimension weights of the correlation matrix are multiplied by tensors with the first intensity value of the corresponding timestamp and the key event point in the multimodal data to generate a first sequence. This sequence quantifies the comprehensive correlation strength between myocardial injury and inflammation and hemodynamic deterioration.

[0071] Step 402: When the score value in the first sequence exceeds the target anomaly threshold three times in a row, the first intensity value is non-linearly scaled using a dynamic adjustment coefficient to generate a coupling strength curve; In this step, the coupling strength curve refers to the curve generated after the first strength value is nonlinearly scaled by a dynamic adjustment coefficient, which characterizes the dynamic coupling relationship between myocardial injury and inflammatory outbreak.

[0072] In this embodiment of the application, when the score value in the first sequence exceeds the target abnormal threshold three times in a row, a dynamic adjustment coefficient is triggered to perform nonlinear scaling on the first intensity value to generate a curve reflecting the coupling strength between myocardial injury and inflammatory outbreak. The peaks and troughs of the curve correspond to the key periods of myocardial function deterioration.

[0073] For example, when the concentration in the first sequence exceeds 0.8 ng / mL three times consecutively, an adjustment factor of 8 is triggered, and the ultrasound feature value is multiplied by the natural logarithm of the adjustment factor to generate a curve reflecting the coupling strength between myocardial injury and inflammatory outbreak.

[0074] Step 403: Extract the extreme points of the coupling strength curve during the inflammatory outbreak period in the sepsis stage, match the extreme points with the key event points by time deviation, and generate matching results; In this step, the extreme point during the inflammatory outbreak period refers to the local maximum value of the coupling strength curve during the septic inflammatory outbreak phase, corresponding to the moment when myocardial damage is most severe.

[0075] In this embodiment of the application, the extreme points of the coupling strength curve during the inflammatory outbreak period of sepsis are extracted and matched with the key event points in the multimodal data in terms of time deviation. If the time difference is less than the preset tolerance, it is marked as a successful match, and a matching result containing the matched extreme points and event point pairs is generated.

[0076] Step 404: Based on the matching results, construct an association network and generate the conditional probabilities of the association network to construct a second database corresponding to patients with septic cardiomyopathy. The parent node of the association network is the target abnormality threshold and the dynamic adjustment coefficient, and the child nodes are the descent slope of the first intensity value and the frequency of key event points. In this step, the association network refers to the causal topology composed of parent nodes, child nodes, and their associated edges. Conditional probability refers to the probability of a child node event occurring given a parent node triggering the event, obtained through historical data statistics and Bayesian optimization. Parent nodes refer to biomarker parameters driving the causal relationship, such as target abnormality thresholds and dynamic adjustment coefficients. Child nodes refer to clinical indicators influenced by biomarkers. The rate of decline of the first intensity value refers to the negative rate of change of the first intensity value over time, reflecting the speed of cardiac function deterioration. The frequency of critical event points refers to the number of hemodynamically critical events occurring per unit time, quantifying the degree of circulatory instability.

[0077] In this embodiment, based on the matching results, the parent node is the target anomaly threshold and dynamic adjustment coefficient, and the child node is the descent slope of the first intensity value and the frequency of key event points; the conditional probability from the parent node to the child node is calculated by the Bayesian parameter estimation method, thereby constructing a second database.

[0078] For example, the parent node has a target anomaly threshold of 0.8 ng / mL, a dynamic adjustment coefficient of 8, a first intensity value decreasing slope of 5% per hour, and key event points occurring 1.2 times per hour. The conditional probability parameter is calculated to be 0.75 using the Bayesian parameter estimation method, thereby constructing a second database.

[0079] This application's embodiments generate myocardial injury scores by fusing cross-modal data through tensor product, overcoming the limitations of traditional single-dimensional analysis; based on dynamically adjusted coefficients, a coupling strength curve is generated to accurately capture the temporal correlation between myocardial injury and inflammatory outbreak; through time deviation matching, conditional probability is optimized, fundamentally solving the problem of causal chain breakage caused by static association rules in the prior art, and improving the pathological interpretability of prognostic prediction of septic cardiomyopathy.

[0080] This application provides a specific embodiment, step 404, based on the matching results, constructing an association network and generating conditional probabilities for the association network to construct a second database corresponding to patients with septic cardiomyopathy, wherein the parent node of the association network is the target abnormality threshold and dynamic adjustment coefficient, and the child nodes are the descent slope of the first intensity value and the frequency of key event points, specifically including the following steps: Step 411: Based on the time deviation values ​​of extreme points and key event points in the matching results, when the time deviation value is less than the second preset threshold, establish the association edge from the parent node to the child node and calculate the first weight. Combine the association edge from the parent node to the child node and the first weight to construct the association network. In this step, the time deviation value refers to the absolute value of the time difference between the extreme point of the myocardial injury coupling strength curve and the critical event point, reflecting the tightness of their temporal correlation. The correlation edge refers to the directed edge connecting the parent and child nodes in the correlation network, representing the causal action path. The first weight is the initial causal strength value generated based on the reciprocal of the time deviation value; a larger value indicates a stronger temporal correlation.

[0081] In this embodiment of the application, based on the time deviation value between the extreme point and the key event point in the matching result, when the deviation value is less than the second preset threshold, an associated edge from the parent node to the child node is established, and the first weight is calculated as the reciprocal of the time deviation value, thus completing the initial construction of the associated network.

[0082] For example, the absolute value of the time difference between the peak of the myocardial injury coupling intensity curve and the critical event point is 0.5 hours. This deviation value is less than 1 hour. The association edge is established with the target abnormal threshold, the slope of the dynamic adjustment coefficient to the first intensity value, and the frequency of the critical event point. The weight is calculated as 1 ÷ 0.5 = 2, thereby completing the initial construction of the association network.

[0083] Step 412: Multiply the number of times the target abnormal threshold is triggered during the inflammatory outbreak phase of sepsis with the dynamic adjustment coefficient to generate a second intensity value. Multiply the absolute value of the downward slope of the first intensity value with the frequency of the key event point to generate a third intensity value. In this step, the number of triggers during the inflammatory flare-up period refers to the cumulative number of times the target abnormal threshold is triggered during the inflammatory flare-up period of sepsis, reflecting the abnormal activity of biomarkers.

[0084] In this embodiment of the application, the number of times the target abnormal threshold is triggered during the inflammatory outbreak period of sepsis is counted, and the result is multiplied by a dynamic adjustment coefficient to generate a second intensity value; at the same time, the absolute value of the downward slope of the first intensity value is calculated and multiplied by the frequency of the key event point to generate a third intensity value.

[0085] For example, the target abnormal threshold is triggered 3 times during the inflammatory outbreak phase of sepsis. This is multiplied by the dynamic adjustment coefficient 8 to generate a second intensity value 24. The absolute value of the decreasing slope of the first intensity value is calculated as the product of a decrease of 5% per hour and the frequency of the key event point occurring 1.2 times per hour to generate a third intensity value 6.

[0086] Step 413: Generate the second weight based on the second intensity value, the third intensity value, and the first weight; In this step, the second weight refers to the ratio of the second strength value to the third strength value multiplied by the initial weight, which characterizes the optimization strength of the causal link.

[0087] In this embodiment of the application, the ratio of the second intensity value to the third intensity value is multiplied by the first weight to generate the second weight, which reflects the strength of the causal association between the biomarker and the clinical parameter.

[0088] For example, the second intensity value is 24 and the third intensity value is 6. The ratio of the second intensity value to the third intensity value, 4, is multiplied by the first weight 2 to generate the second weight 8.

[0089] Step 414: Multiply the second weight and the duration of the inflammatory outbreak to generate the conditional probability of the association network, so as to construct a second database corresponding to patients with septic cardiomyopathy.

[0090] In this step, the duration of the inflammatory flare-up refers to the total duration of the inflammatory flare-up during the sepsis stage, and is used to weight the timeliness of causal association.

[0091] In this embodiment of the application, the second weight is multiplied by the duration of the inflammation outbreak to generate a conditional probability, which represents the cumulative strength of the causal association from the parent node to the child node during the inflammation outbreak. This parameter is then written into the second database to complete the database construction.

[0092] For example, the second weight is 8, the duration of the inflammation outbreak is 48 hours, and the multiplication generates a conditional probability of 384. This parameter is written into the second database to complete the database construction.

[0093] This application's embodiments optimize the weights of associated edges by calculating the second and third intensity values, overcoming the limitations of traditional static causal weights; based on the duration of the inflammation outbreak period, conditional probabilities are generated to achieve the timeliness quantification of causal associations, fundamentally solving the problem of missing causal cumulative effects caused by fixed time windows in existing technologies, and improving the clinical applicability of the second database.

[0094] This application provides a specific embodiment. Step 105 involves using a Bayesian inference model based on a second database to analyze specific biomarkers of sepsis-related myocardial injury and clinical parameters in multimodal data to generate a target interaction path. This specifically includes the following steps: Step 501: Extract the peak time point when the dynamic adjustment coefficient reaches its maximum value, and extract the starting time point of the decline of the first intensity value; In this step, the peak time point refers to the time when the dynamic adjustment coefficient reaches its maximum value, reflecting the strongest moment when inflammatory factors regulate myocardial injury. The decline initiation time point refers to the initial time point when the first intensity value begins to decline continuously, characterizing the starting moment of cardiac function deterioration.

[0095] In this embodiment of the application, the time point when the dynamic adjustment coefficient reaches its maximum value is extracted from the second database and defined as the peak time point; at the same time, the starting time point when the first intensity value begins to continuously decrease is extracted.

[0096] Step 502: When the starting time point of the decline lags behind the peak time point, and the overlap between the overlapping window corresponding to the peak time point and the time window corresponding to the starting time point of the decline exceeds a preset ratio, a pathological causal link from the regulation of the target complex to the decline of the first intensity value is established. In this step, the preset ratio refers to the threshold of the overlapping window ratio that triggers the establishment of the pathological causal link, used to verify the temporal correlation of cross-modal events. The pathological causal link refers to the causal relationship pathway that describes the regulation of the target complex leading to a decline in cardiac function.

[0097] In this embodiment, it is determined whether the starting time point of the decline lags behind the peak time point, and the proportion of the overlap between the overlapping window corresponding to the peak time point and the time window corresponding to the starting time point of the decline is calculated; when the overlap proportion exceeds the preset proportion, a pathological causal link from the regulation of the target complex to the decline of the first intensity value is established.

[0098] For example, the starting time point of the decline is the 14th hour, which lags behind the peak time point of the 12th hour. The overlap ratio between the 10th hour to the 14th hour and the 14th hour to the 18th hour is calculated to be 25%. When the overlap ratio exceeds 20%, a pathological causal link from the peak time point event to the starting time point event is established.

[0099] Step 503: Multiply the average rate of change of the weight value of the second signal and the target complex in the correlation matrix with the first intensity value within the time window corresponding to the descent start time to generate the third weight; In this step, the weight value of the second signal and the target complex refers to the numerical value in the correlation matrix that reflects the dynamic correlation strength between troponin I and inflammatory factors.

[0100] In this embodiment, the weight value of the second signal and the target complex is extracted from the correlation matrix, and the average rate of change of the first intensity value within the decreasing start time window is calculated; the weight value is multiplied by the average rate of change to generate a third weight, which quantifies the causal strength of the link; for example, the weight value of the second signal and the target complex is 0.64, and the average rate of change from the 14th to the 18th hour is calculated to be 3% per hour; the weight value is multiplied by the average rate of change to 1.92 to generate a third weight, which quantifies the causal strength of the link.

[0101] Step 504: Integrate all pathological causal links and third weights that meet the preset conditions to generate the target interaction path.

[0102] In this step, the third weight refers to the parameter generated by multiplying the weight value by the rate of change of the ultrasound feature, which quantifies the confidence of the causal link.

[0103] In this embodiment of the application, all pathological causal links that meet the conditions are screened and their weight values ​​are sorted from high to low to generate target interaction paths, wherein the higher the weight value of the link, the stronger the causal association between inflammatory factors and the deterioration of cardiac function.

[0104] This application's embodiments establish a pathological causal link by verifying the temporal lag between the peak of inflammatory factors and the decline in cardiac function, overcoming the limitations of traditional schemes that rely solely on statistical correlation; and generate path weights based on the dynamic product of weight values ​​and change rates to generate target interaction paths.

[0105] This application provides a specific embodiment. Step 106 involves determining the target target based on the causal effect intensity of each node in the target interaction path, and generating a prognostic prediction scheme based on the first signal corresponding to the target target and the target risk threshold. The specific steps include the following: Step 601: Generate the causal effect strength based on the product of the third weight in the target interaction path and the length of the preset time window of the corresponding pathological causal link; In this step, the preset time window of the pathological causal link refers to the pre-set time period of the pathological causal link, which is used to calculate the cumulative intensity of the causal effect over time.

[0106] In this embodiment of the application, each third weight is extracted from the target interaction path and multiplied by the preset time window length of the corresponding pathological causal link to generate the causal effect strength. This parameter reflects the cumulative influence strength of the causal link in the time dimension. For example, the third weight is 1.92, the time window length is 4 hours, and the multiplication results in a time window length of 7.68. This parameter reflects the cumulative influence strength of the causal link in the time dimension.

[0107] Step 602: Select protein targets whose causal effect intensity exceeds the third preset threshold, prioritize the protein targets, and generate the second sequence; In this step, the third preset threshold refers to the critical value of the causal effect intensity that triggers target screening, reflecting the level of the causal link.

[0108] In this embodiment, a third preset threshold is set to screen protein targets whose causal effect intensity exceeds the threshold, and second sequences are generated by sorting them from high to low intensity values ​​to guide the priority selection of therapeutic targets.

[0109] Step 603: When the overlap ratio between the time window corresponding to the protein target in the second sequence and the time window corresponding to the descent start time point or the dense window corresponding to the key event point exceeds the fourth preset threshold, it is marked as the target. In this step, the overlap ratio refers to the percentage of the overlap time between two time windows relative to the total duration of the target time window, used to verify the temporal consistency of multimodal data. The fourth preset threshold is the threshold for recording the overlap ratio of the target points, ensuring strong correlation between cross-modal events.

[0110] In this embodiment of the application, the overlap ratio between the time window corresponding to the protein target in the second sequence and the first intensity value decrease start time window or key event point dense window is calculated. When the overlap ratio exceeds the fourth preset threshold, the target is marked as the target target.

[0111] Step 604: If the time deviation between the peak time point and the key event point corresponding to the target point is less than the fifth preset threshold, generate the first signal; In this step, the fifth preset threshold refers to the critical value of the time deviation between the peak time point and the key event point, which is used to trigger the early warning signal.

[0112] In this embodiment of the application, the time deviation between the peak time point of the target point and the key event point is extracted. When the time deviation value is less than a fifth preset threshold, a first signal is generated.

[0113] Step 605: When the left ventricular end-diastolic diameter in the multimodal data exceeds the target risk threshold for three consecutive measurements, a prognostic prediction scheme is generated by combining the first signal.

[0114] In this step, the three consecutive measurements of the left ventricular end-diastolic diameter refer to the left ventricular end-diastolic diameter measured by three consecutive echocardiographic examinations, reflecting the degree of ventricular structural dilation. Anti-inflammatory treatment priority refers to the sequence of drug interventions determined based on target priority. Mechanical circulatory support timing refers to the time point at which external circulatory support devices are activated, determined based on warning signals and cardiac function indicators. Ventricular protective dose refers to the drug dosage regimen adjusted according to the risk of ventricular remodeling.

[0115] In this embodiment of the application, when the left ventricular end-diastolic diameter in the multimodal data exceeds the target risk threshold for three consecutive measurements, an individualized prognostic treatment plan is generated by combining the first signal, which includes the priority of anti-inflammatory treatment, the timing of mechanical circulatory support, and the ventricular protection dose.

[0116] This application's embodiments accurately screen high-confidence treatment targets by quantifying the intensity of causal effects and multimodal collaborative verification; combined with hemodynamic early warning and target risk thresholds, it generates individualized prognostic plans, breaking through the limitations of traditional plans that rely on a single indicator for treatment decisions, and improving the timeliness and accuracy of intervention for septic cardiomyopathy.

[0117] Figure 2 This application provides a schematic diagram of the structure of a multimodal data fusion assessment and prognostic prediction system for septic cardiomyopathy, as shown in the embodiments of this application. Figure 2 As shown, the system includes: The acquisition module 21 is used to acquire target data and electronic medical records of patients with septic cardiomyopathy. The target data includes the analysis results corresponding to troponin I and the target complex, respectively. Module 22 is used to construct an association matrix based on the analysis results and a pre-defined first database; Processing module 23 is used to align the labels of the sepsis stage, echocardiographic data and hemodynamic data in the electronic medical record to obtain multimodal data; The fusion module 24 is used to perform feature fusion processing on multimodal data according to preset processing rules in order to construct a second database corresponding to patients with septic cardiomyopathy. The reasoning module 25 is used to analyze specific biomarkers of sepsis-induced myocardial injury and clinical parameters in multimodal data based on the second database and using a Bayesian reasoning model to generate a target interaction path. The early warning module 26 is used to determine the target target point based on the causal effect intensity of each node in the target interaction path, and generate a prognostic prediction scheme based on the first signal corresponding to the target target point and the target risk threshold. The first signal is an intelligent alarm signal used to provide early risk warning and guide targeted intervention before the occurrence of circulatory failure.

[0118] Figure 2 A multimodal data fusion assessment and prognostic prediction system for septic cardiomyopathy can perform... Figure 1 The implementation principle and technical effects of the multimodal data fusion assessment and prognostic prediction method for septic cardiomyopathy shown in the embodiment will not be elaborated further. The specific operation methods of each module and unit in the multimodal data fusion assessment and prognostic prediction system for septic cardiomyopathy in the above embodiment have been described in detail in the embodiments related to this method, and will not be elaborated further here.

[0119] In one possible design, Figure 2 The multimodal data fusion assessment and prognostic prediction system for septic cardiomyopathy shown in the embodiment can be implemented as a computing device, such as... Figure 3As shown, the computing device may include a storage component 31 and a processing component 32; Storage component 31 stores one or more computer instructions, wherein one or more computer instructions are invoked and executed by processing component 32.

[0120] Processing component 32 is used to: acquire target data and electronic medical records of patients with septic cardiomyopathy, including the analysis results corresponding to troponin I and the target complex; construct an association matrix based on the analysis results and a preset first database; align the labels of the sepsis stage, echocardiographic data, and hemodynamic data in the electronic medical records to obtain multimodal data; perform feature fusion processing on the multimodal data according to preset processing rules to construct a second database corresponding to patients with septic cardiomyopathy; based on the second database, use a Bayesian inference model to analyze the specific biomarkers of septic myocardial injury and the clinical parameters in the multimodal data to generate a target interaction path; determine the target point based on the causal effect strength of each node in the target interaction path, and generate a prognostic prediction scheme based on the first signal corresponding to the target point and the target risk threshold.

[0121] The processing component 32 may include one or more processors to execute computer instructions to complete all or part of the steps in the above-described method. Alternatively, the processing component may be implemented as one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field-programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, or other electronic components to perform the above-described method.

[0122] Storage component 31 is configured to store various types of data to support operations at the terminal. The storage component can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0123] Of course, computing devices may also include other components, such as input / output interfaces, display components, communication components, etc.

[0124] Input / output interfaces provide interfaces between processing components and peripheral interface modules, which can be output devices, input devices, etc.

[0125] The communication components are configured to facilitate wired or wireless communication between computing devices and other devices.

[0126] The computing device can be a physical device or an elastic computing host provided by a cloud computing platform. In this case, the computing device can refer to a cloud server, and the aforementioned processing components, storage components, etc., can be basic server resources rented or purchased from the cloud computing platform.

[0127] This application also provides a computer storage medium storing a computer program, which, when executed by a computer, can perform the above-described functions. Figure 1 The embodiment shown illustrates a multimodal data fusion assessment and prognostic prediction method for septic cardiomyopathy.

[0128] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0129] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0130] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of various embodiments or some parts of embodiments.

[0131] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A multimodal data fusion assessment and prognostic prediction method for septic cardiomyopathy, characterized in that, include: Acquire target data and electronic medical records of patients with septic cardiomyopathy, wherein the target data includes the analysis results corresponding to troponin I and the target complex, respectively; Based on the analysis results and the preset first database, an association matrix is ​​constructed; The labels for the sepsis stage, echocardiographic data, and hemodynamic data in the electronic medical record are aligned to obtain multimodal data. The multimodal data is subjected to feature fusion processing according to preset processing rules to construct a second database corresponding to the patients with septic cardiomyopathy; Based on the second database, using a Bayesian inference model, specific biomarkers of sepsis-induced myocardial injury and clinical parameters in the multimodal data are analyzed to generate a target interaction path; Based on the causal effect intensity of each node in the target interaction path, the target target is determined. Based on the first signal corresponding to the target target and combined with the target risk threshold, a prognostic prediction scheme is generated. The first signal is an intelligent alarm signal used to provide early risk warning and guide targeted intervention before circulatory failure occurs.

2. The method according to claim 1, characterized in that, Based on the analysis results and the pre-defined first database, an association matrix is ​​constructed, including: The baseline abnormal threshold and dynamic fluctuation range of the second signal within a preset time window are obtained from the preset first database. The second signal is the sequence data of the change of troponin I concentration over time. Based on the analysis results, the signal intensity change rate of the second signal corresponding to the preset time window is calculated. When the signal intensity change rate exceeds the dynamic fluctuation range, the second signal is marked as an abnormal signal, and a target feature pattern corresponding to the abnormal signal is extracted from the preset first database. The target feature pattern is a reference fluctuation curve used to dynamically match with the abnormal signal and verify the temporal correlation between inflammation outbreak and myocardial injury. The overlap between the first time window of the abnormal signal and the second time window of the target feature pattern is calculated. When the overlap is greater than a first preset threshold, the target extreme value of the target complex within the overlapping window is obtained. Based on the target extreme value and the benchmark abnormal threshold, a dynamic adjustment coefficient is generated. The target extreme value is used to quantify the maximum rate of change of inflammatory factors within the overlapping window. Based on the dynamic adjustment coefficient, the baseline abnormal threshold is adjusted to generate a target abnormal threshold that matches the analysis result. The target abnormal threshold is an abnormal judgment threshold used after the dynamic adjustment coefficient is corrected, and it adapts to the patient's pathological state in real time. Based on the target anomaly threshold and the target extreme value, the preset time window, the first time window, and the second time window are aligned to generate an association matrix.

3. The method according to claim 1, characterized in that, The sepsis stage tags, echocardiographic data, and hemodynamic data in the electronic medical record are aligned to obtain multimodal data, including: Based on the echocardiographic data in the electronic medical record, the coefficient of variation of the left ventricular ejection fraction and the instantaneous fluctuation amplitude of the overall longitudinal strain are calculated to generate a first intensity value. The first intensity value is a composite ultrasound parameter used to quantify the degree of dynamic abnormality of cardiac function and to provide core quantitative indicators of cardiac function for multimodal data fusion. Based on the hemodynamic data in the electronic medical record, the moving average of cardiac output and the rate of change of extravascular lung water index are calculated. The time point when the rate of change exceeds a second preset threshold is marked as a critical event point. Based on the labels of the sepsis stage, the first intensity value and the key event points are aligned to generate multimodal data.

4. The method according to claim 1, characterized in that, The multimodal data is subjected to feature fusion processing according to preset processing rules to construct a second database corresponding to the patients with septic cardiomyopathy, including: The row dimension weights of the correlation matrix are multiplied by tensors with the first intensity value and key event point of the corresponding timestamp in the multimodal data to generate the first sequence; When the score value in the first sequence exceeds the target anomaly threshold three times in a row, the first intensity value is non-linearly scaled using a dynamic adjustment coefficient to generate a coupling intensity curve. Extract the extreme points of the coupling strength curve during the inflammatory outbreak period in the sepsis stage, and match the extreme points with the key event points by time deviation to generate matching results; Based on the matching results, an association network is constructed, and the conditional probabilities of the association network are generated to construct a second database corresponding to the patients with septic cardiomyopathy. The parent node of the association network is the target abnormality threshold and the dynamic adjustment coefficient, and the child nodes are the descent slope of the first intensity value and the frequency of key event points.

5. The method according to claim 4, characterized in that, Based on the matching results, an association network is constructed, and conditional probabilities of the association network are generated to build a second database corresponding to the patients with septic cardiomyopathy. The parent nodes of the association network are the target abnormality threshold and dynamic adjustment coefficient, and the child nodes are the descent slope of the first intensity value and the frequency of key event points, including: Based on the time deviation values ​​of the extreme points and key event points in the matching results, when the time deviation value is less than the second preset threshold, an association edge from the parent node to the child node is established, and a first weight is calculated. Combining the association edge from the parent node to the child node and the first weight, an association network is constructed. The second intensity value is generated by multiplying the number of times the target abnormal threshold is triggered during the inflammatory outbreak phase of the sepsis stage with the dynamic adjustment coefficient. The third intensity value is generated by multiplying the absolute value of the downward slope of the first intensity value with the frequency of the key event point. A second weight is generated based on the second intensity value, the third intensity value, and the first weight; The second weight and the duration of the inflammatory outbreak are multiplied to generate the conditional probabilities of the association network, thereby constructing a second database corresponding to the patients with septic cardiomyopathy.

6. The method according to claim 1, characterized in that, Based on the second database, using a Bayesian inference model, specific biomarkers of sepsis-related myocardial injury and clinical parameters in the multimodal data are analyzed to generate a target interaction path, including: Extract the peak time point when the dynamic adjustment coefficient reaches its maximum value, and extract the starting time point of the decline of the first intensity value; When the starting point of the decline lags behind the peak time point, and the overlap between the overlapping window corresponding to the peak time point and the time window corresponding to the starting point of the decline exceeds a preset ratio, a pathological causal link from the regulation of the target complex to the decline of the first intensity value is established. The third weight is generated by multiplying the weight values ​​of the second signal and the target complex in the correlation matrix with the average rate of change of the first intensity value within the time window corresponding to the descent start time. Integrate all pathological causal links and third weights that meet the preset conditions to generate the target interaction path.

7. The method according to claim 1, characterized in that, Based on the causal effect strength of each node in the target interaction path, target points are determined. Based on the first signal corresponding to the target point and combined with the target risk threshold, a prognostic prediction scheme is generated, including: The causal effect strength is generated by multiplying the third weight in the target interaction path by the length of the preset time window of the corresponding pathological causal link. Protein targets with causal effect strength exceeding a third preset threshold are selected, and the protein targets are prioritized to generate a second sequence; When the overlap ratio between the time window corresponding to the protein target in the second sequence and the time window corresponding to the descent start time point or the dense window corresponding to the key event point exceeds the fourth preset threshold, it is marked as a target. If the time deviation between the peak time point and the key event point corresponding to the target point is less than the fifth preset threshold, a first signal is generated. When the left ventricular end-diastolic diameter in the multimodal data exceeds the target risk threshold for three consecutive measurements, a prognostic prediction scheme is generated in conjunction with the first signal.

8. A multimodal data fusion assessment and prognostic prediction system for septic cardiomyopathy, characterized in that, include: The acquisition module is used to acquire target data and electronic medical records of patients with septic cardiomyopathy. The target data includes the analysis results corresponding to troponin I and the target complex, respectively. The construction module is used to construct an association matrix based on the analysis results and a preset first database; The processing module is used to align the labels for the sepsis stage, echocardiographic data, and hemodynamic data in the electronic medical record to obtain multimodal data. The fusion module is used to perform feature fusion processing on the multimodal data according to preset processing rules to construct a second database corresponding to the patients with septic cardiomyopathy. The inference module is used to analyze specific biomarkers of sepsis-induced myocardial injury and clinical parameters in the multimodal data based on the second database using a Bayesian inference model, and generate a target interaction path. The early warning module is used to determine the target target point based on the causal effect intensity of each node in the target interaction path, and generate a prognostic prediction scheme based on the first signal corresponding to the target target point and in combination with the target risk threshold. The first signal is an intelligent alarm signal used to provide early risk warning and guide targeted intervention before circulatory failure occurs.

9. A computing device, characterized in that, It includes a processing component and a storage component; the storage component stores one or more computer instructions; the one or more computer instructions are invoked and executed by the processing component to implement a multimodal data fusion assessment and prognostic prediction method for septic cardiomyopathy as described in any one of claims 1 to 7.

10. A computer storage medium, characterized in that, The device contains a computer program that, when executed by a computer, implements a multimodal data fusion assessment and prognostic prediction method for septic cardiomyopathy as described in any one of claims 1 to 7.

Images