A method and system for intelligent monitoring of organ transplant patients post-surgery

Abstract

The application provides a kind of organ transplant patient postoperative intelligent monitoring method and system, the method comprises: based on the physiological parameter baseline model of the organ transplant type and pathological characteristics of patient is constructed, for automatically adjusting baseline threshold range according to the recovery stage of patient after operation, and the individualized calibration of fusing patient preoperative physiological parameter baseline with the statistical baseline of same type organ transplant recipient population;Multi-dimensional physiological parameters are input into the model, and the deviation index of each dimension physiological parameter is calculated, when the deviation index exceeds the early warning threshold of corresponding stage, the time sequence correlation characteristics of physiological parameters between different dimensions are identified by time sequence correlation algorithm, to detect the biomarker combination characteristics of potential rejection reaction, generate risk level assessment results and corresponding clinical intervention suggestions, to improve the accuracy and clinical usability of organ transplant patient postoperative physiological monitoring, ensure the accuracy and stability of deviation index calculation, and realize accurate, personalized early warning ability of rejection reaction.

Description

Technical Field

[0001] This invention relates to the field of physiological monitoring technology, and in particular to an intelligent monitoring method and system for postoperative organ transplant patients. Background Technology

[0002] In recent years, with the development of wearable devices and telemedicine, remote monitoring technology after organ transplantation has gradually attracted attention. It uses wearable devices to collect physiological parameters of patients in their home environment, realizes remote data monitoring through wireless transmission, and attempts to establish a baseline of physiological parameters. It provides early warning of abnormalities by comparing the deviation index of the current physiological parameters from the baseline.

[0003] However, current remote monitoring after organ transplantation lacks precise and personalized early warning capabilities for rejection. Furthermore, it often relies on fixed thresholds or static baselines for anomaly assessment, leading to frequent false alarms in the early postoperative period due to physiological fluctuations and missed alarms in the later postoperative period due to baseline drift. This results in insufficient monitoring accuracy and clinical usability. In addition, individual differences exist in the preoperative physiological state of organ transplant patients. Some patients have underlying diseases causing abnormal physiological parameters that deviate significantly from the population statistical baseline. Alternatively, in the early postoperative period, the patient's physiological state is not yet stable, and individual preoperative baselines may conflict with the population statistical baseline. Current technology lacks the ability to adaptively handle baseline conflicts, affecting the accuracy and stability of deviation index calculations. Summary of the Invention

[0004] This invention provides a method and system for intelligent monitoring of organ transplant patients after surgery, aiming to improve the accuracy and clinical usability of postoperative physiological monitoring of organ transplant patients, ensure the accuracy and stability of deviation index calculation, and achieve precise and personalized early warning capability for rejection.

[0005] In a first aspect, the present invention provides a method for intelligent monitoring of organ transplant patients after surgery, comprising: Collect multidimensional physiological parameters of patients in their home environment; A physiological parameter baseline model is constructed based on the patient's organ transplant type and pathological characteristics. The physiological parameter baseline model adopts a sliding time window statistical method to automatically adjust the baseline threshold range according to the patient's postoperative recovery stage, and integrates the patient's preoperative physiological parameter baseline with the statistical baseline of the same type of organ transplant recipient group for personalized calibration. The multidimensional physiological parameters are input into the physiological parameter baseline model, and the deviation index of each physiological parameter is calculated. When the deviation index of any dimension exceeds the warning threshold of the corresponding postoperative recovery stage, the temporal correlation characteristics of physiological parameters in different dimensions are identified by the temporal correlation algorithm. Based on the temporal correlation characteristics, the combination characteristics of biomarkers of potential rejection reaction are detected, and the risk level assessment result is generated. Based on the risk level assessment results, corresponding clinical intervention recommendations are matched and sent to the patient's mobile terminal.

[0006] Preferably, the multidimensional physiological parameters include body temperature, heart rate, blood pressure, blood oxygen saturation, respiratory rate, organ function-specific parameters, tissue oxygenation index, and records of immunosuppressant administration time.

[0007] Preferably, the construction of a baseline model of physiological parameters based on the patient's organ transplant type and pathological characteristics includes: The postoperative recovery phase of patients is divided into multiple phases, and each phase is configured with an independent baseline calculation strategy. The initial physiological parameter baseline model is determined based on the patient's organ transplant type. Individual data of the patient is constructed based on the patient's preoperative physiological parameters and pathological characteristics. The similarity between the individual data and the group data of the same type of organ transplant recipients is calculated at each stage. When the similarity is higher than a preset similarity threshold, the individual data weight of the baseline calculation strategy for the corresponding stage in the initial physiological parameter baseline model is increased to above a first preset value, and the group data weight is reduced accordingly. When the similarity is lower than a preset similarity threshold, the individual data weight of the baseline calculation strategy for the corresponding stage in the initial physiological parameter baseline model is reduced to below a second preset value, and the group data weight is increased accordingly to generate a physiological parameter baseline model that integrates the baseline calculation strategies of multiple stages.

[0008] Furthermore, before calculating the similarity between the individual data and the group data of organ transplant recipients of the same type at each stage, the method further includes: The individual data is cleaned, and missing values ​​are filled using an interpolation algorithm.

[0009] Preferably, the postoperative recovery phase of the patient is divided into a hyperacute phase, an acute phase, a subacute phase, and a chronic phase. The hyperacute phase is within three days after surgery, the acute phase is from three days to three months after surgery, the subacute phase is from three months to twelve months after surgery, and the chronic phase is more than twelve months after surgery.

[0010] Preferably, the calculation of the deviation index of each dimension of physiological parameters includes: The distribution similarity between the physiological parameters of each dimension within the current time window and the statistical baseline of the corresponding postoperative recovery stage is measured, and the information entropy difference between the probability distribution of individual physiological parameters and the probability distribution of the statistical baseline of the group is compared to quantify the degree of individual deviation from the group as the distribution deviation component. Fit the trend lines of physiological parameters in each dimension within the current time window, calculate the angle between the trend lines and the population statistical baseline, obtain the deviation angle between the current trend direction and the population statistical baseline, and calculate the duration for which the trend lines continuously exceed the fluctuation range of the population statistical baseline to obtain the trend maintenance duration. Based on the deviation angle and the trend maintenance duration, evaluate the temporal stability deviation component. A physiological correlation topology is established among physiological parameters. When a single-dimensional physiological parameter deviates, it is detected whether other-dimensional physiological parameters that have a physiological coupling relationship with the deviated physiological parameter synchronously show a cooperative change pattern. If so, the cross-dimensional coupling deviation component is calculated. The physiological correlation topology is used to reflect the functional coupling relationship between different physiological parameters. The cooperative change pattern is identified by comparing the temporal relationship of the deviations between the correlated dimensions and the consistency of the change direction. Based on the current postoperative recovery stage, different fusion weights are assigned to the distribution deviation component, the temporal stability deviation component, and the cross-dimensional coupling deviation component, and the weighted average value is calculated as the deviation index.

[0011] Preferably, the step of detecting the biomarker combination features of potential rejection based on the time-series correlation features and generating a risk level assessment result includes: The time-series correlation features are input into the early rejection prediction model, and the output is the combination features and probability values ​​of biomarkers for future acute rejection. Based on the probability values, a risk level assessment result is generated. The early rejection prediction model adopts an encoder-decoder architecture, where the encoder is a multilayer long short-term memory network and the decoder is a fully connected neural network.

[0012] Preferably, the step of inputting the time-series correlation features into the early prediction model of rejection reaction and outputting the combination features and probability values ​​of biomarkers for future acute rejection reaction includes: The temporal correlation features are transformed into a three-dimensional spatiotemporal tensor. The correlation strength values ​​between each pair of physiological parameters at different times are calculated to fill the three-dimensional spatiotemporal tensor, forming a temporal correlation strength field. The first dimension of the three-dimensional spatiotemporal tensor represents the physiological parameter dimension, the second dimension represents the time step, and the third dimension represents the correlation strength level. The temporal correlation intensity field is fed into the first layer of the encoder's long short-term memory network to extract short-term local correlation patterns, resulting in a short-term correlation feature map. The short-term correlation feature map is then fed into the second layer of the encoder's long short-term memory network to extract medium-term trend correlation patterns, resulting in a medium-term correlation feature map. The medium-term correlation feature map is then fed into the third layer of the encoder's long short-term memory network to extract long-term evolution correlation patterns, resulting in a long-term correlation feature map. The short-term correlation feature map, medium-term correlation feature map, and long-term correlation feature map are then spliced ​​and fused along the feature dimension to obtain a multi-scale spatiotemporal correlation feature body. A biomarker combination decoupling generation channel is established in the decoder. Parallel decoupling channels with the same number of biomarker types are set up, with each channel corresponding to one biomarker. The multi-scale spatiotemporal correlation feature volume is sent to each parallel decoupling channel. Each channel calculates the occurrence confidence of the corresponding biomarker through a fully connected neural network, and gating connections between channels are set according to the physiological correlation of each biomarker. When a target biomarker channel has a confidence level exceeding the activation threshold, the response intensity of other biomarker channels that have a physiological synergistic relationship with the target biomarker channel is enhanced through gating connections, while the response of channels that have a physiological antagonistic relationship with the target biomarker channel is suppressed. The combination of the confidence levels of the biomarkers output by each channel forms a biomarker combination feature. The biomarker combination features are fed into the probability estimation layer, which calculates the similarity between the biomarker combination features and historically diagnosed rejection cases, generating the probability value of acute rejection occurring on future days.

[0013] Preferably, the matching of corresponding clinical intervention recommendations based on the risk level assessment results includes: Based on the risk level assessment results, matching clinical intervention recommendations are retrieved from a pre-built knowledge base. These recommendations include clinical treatment plans, examination recommendations, and lifestyle adjustment guidelines corresponding to different risk levels.

[0014] Secondly, the present invention also provides an intelligent monitoring system for postoperative organ transplant patients, comprising: The data acquisition module is used to collect multi-dimensional physiological parameters of patients in their home environment; The module is used to construct a baseline model of physiological parameters based on the patient's organ transplant type and pathological characteristics. The baseline model of physiological parameters adopts a sliding time window statistical method to automatically adjust the baseline threshold range according to the patient's postoperative recovery stage, and integrates the patient's preoperative physiological parameter baseline with the statistical baseline of the same type of organ transplant recipient group for personalized calibration. The calculation module is used to input the multi-dimensional physiological parameters into the physiological parameter baseline model, calculate the deviation index of each dimension of physiological parameters, and when the deviation index of any dimension exceeds the warning threshold of the corresponding postoperative recovery stage, the temporal correlation algorithm is used to identify the temporal correlation characteristics of physiological parameters in different dimensions, and the biomarker combination characteristics of potential rejection reaction are detected based on the temporal correlation characteristics to generate risk level assessment results. The matching module is used to match corresponding clinical intervention recommendations based on the risk level assessment results and send the clinical intervention recommendations to the patient's mobile terminal.

[0015] Compared with the prior art, the technical solution of the present invention has at least the following advantages: This invention provides an intelligent monitoring method and system for postoperative organ transplant patients. Based on the patient's organ transplant type and pathological characteristics, a physiological parameter baseline model is constructed. Combined with a sliding time window statistical method, the baseline threshold range is automatically adjusted according to the postoperative recovery stage, allowing the baseline threshold to dynamically evolve with the recovery stage. This reduces the false alarm rate caused by physiological fluctuations in the early postoperative period and the false negative rate caused by baseline drift in the later postoperative period, improving the accuracy and clinical usability of monitoring. Furthermore, by fusing the patient's preoperative physiological parameter baseline with the statistical baseline of a group of organ transplant recipients of the same type for personalized calibration, the deviation index of each dimension of physiological parameters is calculated based on the physiological parameter baseline model. This solves the problem of calculating the deviation index when individual baselines conflict with group statistical baselines in the early postoperative period, improving the stability and reliability of the deviation index calculation. In addition, when the deviation index exceeds the warning threshold of the corresponding postoperative recovery stage, a time-series correlation algorithm can capture the coordinated change patterns of different physiological parameters over time, revealing systemic pathological signals that cannot be detected by single-parameter analysis. Furthermore, through time-series correlation feature analysis, specific marker combinations of rejection reactions are identified, significantly reducing the false positive rate and improving the accuracy of early rejection reaction identification, thereby achieving precise and personalized early warning capabilities for rejection reactions. Attached Figure Description

[0016] Figure 1 This is a flowchart of an embodiment of an intelligent monitoring method for postoperative organ transplant patients according to the present invention; Figure 2 This is a flowchart of another embodiment of the intelligent monitoring method for postoperative organ transplant patients according to the present invention; Figure 3 This is a structural block diagram of one embodiment of an intelligent monitoring system for postoperative organ transplant patients according to the present invention. Detailed Implementation

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

[0018] Please refer to Figure 1 As shown, the present invention provides a method for intelligent monitoring of organ transplant patients after surgery, comprising the following steps: S11. Collect multi-dimensional physiological parameters of patients in their home environment; S12. Construct a physiological parameter baseline model based on the patient's organ transplant type and pathological characteristics. The physiological parameter baseline model adopts a sliding time window statistical method to automatically adjust the baseline threshold range according to the patient's postoperative recovery stage, and integrates the patient's preoperative physiological parameter baseline with the statistical baseline of the same type of organ transplant recipient group for personalized calibration. S13. Input the multi-dimensional physiological parameters into the physiological parameter baseline model, calculate the deviation index of each dimension of physiological parameters, and when the deviation index of any dimension exceeds the warning threshold of the corresponding postoperative recovery stage, identify the temporal correlation characteristics of physiological parameters in different dimensions through the temporal correlation algorithm, detect the biomarker combination characteristics of potential rejection based on the temporal correlation characteristics, and generate risk level assessment results. S14. Based on the risk level assessment results, match the corresponding clinical intervention recommendations and send the clinical intervention recommendations to the patient's mobile terminal.

[0019] This embodiment utilizes intelligent monitoring devices deployed in the patient's home environment to continuously collect various physiological data related to their post-organ transplant condition, extending the monitoring scenario from the hospital to the home and filling the monitoring gap after discharge. The data collection dimensions cover vital signs, biochemical indicators, immune-related indicators, and patient-reported data. Collection methods include automatic monitoring via wearable devices, periodic measurements with home testing instruments, and manual data entry via mobile terminals.

[0020] Preferably, the multidimensional physiological parameters include body temperature, heart rate, blood pressure, blood oxygen saturation, respiratory rate, organ function-specific parameters, tissue oxygenation index, and records of immunosuppressant administration time.

[0021] Secondly, a baseline model of physiological parameters is constructed that adapts to the individual characteristics of patients and dynamically evolves with the postoperative recovery process. This baseline model is a reference benchmark model established by statistical analysis methods based on the patient's preoperative physiological parameters and data from similar patient groups. It is used to assess the degree of deviation of the current physiological state, including baseline values ​​(central tendency) and baseline range (dispersion). The sliding time window method is used to dynamically update the model with the postoperative recovery stage.

[0022] Based on the type of organ transplant received by the patient (e.g., liver, kidney, heart, lung, etc.) and the pathological characteristics of the primary disease (e.g., autoimmune liver disease, diabetes, kidney disease, etc.), a baseline model of physiological parameters was constructed. A sliding time window statistical method was used to set baseline threshold ranges for different postoperative recovery stages. The patient's stable preoperative physiological parameter baseline (reflecting individual physiological characteristics) was weighted and fused with the statistical baseline of the same type of organ transplant recipient group (reflecting organ-specific recovery patterns). The weights were dynamically adjusted according to postoperative time; for example, the weight of the group baseline was higher in the early postoperative period, and the individual baseline weight gradually increased as the recovery progressed, achieving personalized calibration.

[0023] Preferably, the patient's postoperative recovery phase is divided into a hyperacute phase, an acute phase, a subacute phase, and a chronic phase. The hyperacute phase is within three days after surgery, the acute phase is from three days to three months after surgery, the subacute phase is from three to twelve months after surgery, and the chronic phase is more than twelve months after surgery.

[0024] Then, the collected multidimensional physiological parameters are input into the baseline model, and the deviation index of each parameter is calculated. When the deviation index of any dimension parameter exceeds the preset warning threshold for the recovery stage (e.g., the deviation index is greater than 2), the correlation pattern of different physiological parameters in the time dimension is analyzed by the time-series correlation algorithm to identify the regularity of the change of specific parameters leading or lagging behind other parameters, and obtain the time-series correlation features. The time-series correlation features are compared with the combination patterns of biomarkers of known rejection reactions to detect whether there are specific combination features that predict acute rejection reactions, i.e., biomarker combination features, and a risk level assessment result is generated based on the degree of matching.

[0025] The system has a built-in knowledge base of graded intervention recommendations, which can map different risk levels to specific clinical intervention recommendations. These clinical intervention recommendations include clinical treatment plans, examination recommendations, and lifestyle adjustment guidelines corresponding to different risk levels.

[0026] Finally, the instructions are sent to the patient and their family via mobile devices in the form of push notifications, illustrated guides, or video tutorials, and the attending physician is copied as needed.

[0027] This invention provides an intelligent monitoring method for postoperative organ transplant patients. Based on the patient's organ transplant type and pathological characteristics, a physiological parameter baseline model is constructed. Combined with a sliding time window statistical method, the baseline threshold range is automatically adjusted according to the postoperative recovery stage, allowing the baseline threshold to dynamically evolve with the recovery stage. This reduces the false alarm rate caused by physiological fluctuations in the early postoperative period and the false negative rate caused by baseline drift in the later postoperative period, improving the accuracy and clinical usability of monitoring. Furthermore, by fusing the patient's preoperative physiological parameter baseline with the statistical baseline of a group of organ transplant recipients of the same type for personalized calibration, the deviation index of each dimension of physiological parameters is calculated based on the physiological parameter baseline model. This solves the problem of calculating the deviation index when individual baselines conflict with group statistical baselines in the early postoperative period, improving the stability and reliability of the deviation index calculation. In addition, when the deviation index exceeds the warning threshold of the corresponding postoperative recovery stage, a time-series correlation algorithm can capture the coordinated change patterns of different physiological parameters over time, revealing systemic pathological signals that cannot be detected by single-parameter analysis. Furthermore, through time-series correlation feature analysis, specific marker combinations of rejection reactions are identified, significantly reducing the false positive rate and improving the accuracy of early rejection reaction identification, thereby achieving precise and personalized early warning capabilities for rejection reactions.

[0028] In one embodiment, the construction of a baseline model of physiological parameters based on the patient's organ transplant type and pathological characteristics includes: The postoperative recovery phase of patients is divided into multiple phases, and each phase is configured with an independent baseline calculation strategy. The initial physiological parameter baseline model is determined based on the patient's organ transplant type. Individual data of the patient is constructed based on the patient's preoperative physiological parameters and pathological characteristics. The similarity between the individual data and the group data of the same type of organ transplant recipients is calculated at each stage. When the similarity is higher than a preset similarity threshold, the individual data weight of the baseline calculation strategy for the corresponding stage in the initial physiological parameter baseline model is increased to above a first preset value, and the group data weight is reduced accordingly. When the similarity is lower than a preset similarity threshold, the individual data weight of the baseline calculation strategy for the corresponding stage in the initial physiological parameter baseline model is reduced to below a second preset value, and the group data weight is increased accordingly to generate a physiological parameter baseline model that integrates the baseline calculation strategies of multiple stages.

[0029] This embodiment discretizes the continuous postoperative timeline into several key stages with physiological characteristics based on the objective laws of physiological recovery after organ transplantation. The stage division is based on factors including organ type, degree of surgical trauma, and immunosuppression regimen. Each stage is configured with an independent baseline calculation strategy, including the applicable set of physiological parameters, data acquisition frequency, sliding window duration, individual data weight, and population data weight, to adapt to the rate and characteristics of physiological changes at different stages.

[0030] Based on the organ transplant type, a pre-defined universal baseline model for that category is used as the initial physiological parameter baseline model. This model includes the typical physiological parameter ranges and variation curves for that organ transplant type at each stage. Furthermore, physiological parameters and pathological characteristics from the patient's preoperative stable period are extracted to construct individual data characterizing the patient's individual physiological traits. Finally, at each recovery stage, the similarity between the individual data and the distribution of population data at that stage (such as the mean physiological parameters of the same type of organ transplant recipient at that stage) is calculated to quantify the degree of deviation between the patient's recovery and the typical recovery pattern of the population.

[0031] When the similarity exceeds a preset similarity threshold (e.g., 0.7), it indicates that the patient's physiological characteristics, primary disease background, and preoperative condition highly match the typical pattern of similar recipients, and the population data has strong reference and predictive value for this patient. At this point, in the baseline calculation strategy at this stage, the fusion weight of individual data is increased to a first preset value (e.g., greater than 60%), and the weight of population data is correspondingly reduced (e.g., less than 40%), so that the baseline model more fully reflects the patient's own physiological history and enhances the degree of individualization.

[0032] When the similarity score is below a preset similarity threshold, it indicates that the patient has a unique pathological background, comorbidities, or physiological characteristics, significantly differing from the typical recovery pattern of similar recipients. Relying solely on individual historical data or group data may lead to baseline bias. In this case, the baseline calculation strategy at this stage reduces the fusion weight of individual preoperative data to a second preset value (e.g., less than 30%), and correspondingly increases the weight of group statistical data (e.g., greater than 70%), prioritizing the common recovery patterns of this organ transplant type. This further avoids misjudging preoperative pathological conditions as postoperative abnormalities, ensuring that the baseline reflects the true postoperative recovery expectations. Finally, the baseline calculation strategies at all stages are integrated to form a dynamic physiological parameter baseline model covering the entire recovery cycle.

[0033] In this embodiment, by dividing the postoperative recovery phase into multiple stages and configuring independent baseline calculation strategies, the baseline model achieves phased and refined adaptation to postoperative physiological dynamic changes. This ensures that the baseline calculation for each stage aligns with the specific physiological recovery focus and rate of change for that period, improving the stage adaptability of the baseline setting. Secondly, by increasing the weight of individual data and decreasing the weight of group data when the similarity is above a threshold, sufficient individualized calibration for patients with typical recovery patterns is achieved, making the baseline more faithfully reflect the patient's own physiological history and enhancing the individual specificity and predictive accuracy of the baseline model. Furthermore, by decreasing the weight of individual data and increasing the weight of group data when the similarity is below a threshold, priority protection is given to the group patterns of patients with special pathological backgrounds, preventing baseline distortion caused by interference from individual preoperative pathological states, and ensuring that the baseline model still has a reliable physiological reference benchmark under special circumstances. Finally, by integrating the baseline calculation strategies of multiple stages to generate a complete baseline model, the consistency of the baseline strategy and the unity of stage specificity throughout the entire recovery cycle are achieved, allowing the physiological parameter baseline to evolve smoothly with the recovery process, ensuring both accurate assessment within each stage and natural transitions between stages.

[0034] In one embodiment, before calculating the similarity between the individual data and the group data of similar organ transplant recipients at each stage, the method further includes: The individual data is cleaned, and missing values ​​are filled using an interpolation algorithm.

[0035] In this embodiment, data cleaning involves quality checks and anomaly handling of individual data, including identifying and correcting outliers (such as extreme values ​​that significantly exceed the physiological range), removing duplicate records, standardizing units of measurement, and correcting time errors, thereby eliminating noise and errors in individual data and improving data quality.

[0036] Missing value imputation involves using interpolation algorithms and combining the changing trends of adjacent valid data points to make reasonable estimates of missing data points in individual data, thereby completing the data sequence, ensuring the integrity of individual data, and avoiding deviations or inability to perform similarity calculations due to missing data.

[0037] In one embodiment, please refer to Figure 2 As shown, the calculation of the deviation index of each dimension of physiological parameters includes: S131. Measure the distribution similarity between the physiological parameters of each dimension within the current time window and the statistical baseline of the corresponding postoperative recovery stage, and compare the information entropy difference between the probability distribution of individual physiological parameters and the probability distribution of the statistical baseline of the group, and quantify the degree of individual deviation from the group as the distribution deviation component. S132. Fit the trend lines of physiological parameters in each dimension within the current time window, calculate the angle between the trend lines and the population statistical baseline, obtain the deviation angle between the current trend direction and the population statistical baseline, and calculate the duration for which the trend lines continuously exceed the fluctuation range of the population statistical baseline to obtain the trend maintenance duration. Based on the deviation angle and the trend maintenance duration, evaluate the temporal stability deviation component. S133. Establish a physiological correlation topology between physiological parameters. When a single-dimensional physiological parameter deviates, detect whether other-dimensional physiological parameters that have a physiological coupling relationship with the deviated physiological parameter synchronously show a cooperative change pattern. If so, calculate the cross-dimensional coupling deviation component. The physiological correlation topology is used to reflect the functional coupling relationship between different physiological parameters. The cooperative change pattern is identified by comparing the temporal order of deviations between the correlated dimensions and the consistency of the change direction. S134. Based on the current postoperative recovery stage, assign different fusion weights to the distribution deviation component, temporal stability deviation component, and cross-dimensional coupling deviation component, and calculate the weighted average as the deviation index. First, construct individual experience probability distributions for the physiological parameters collected within the current time window, calculate the distribution similarity measure between this individual distribution and the corresponding statistical baseline probability distribution of the group in the postoperative recovery stage, compare the information entropy difference between the two distributions. The larger the information entropy difference, the greater the randomness or uncertainty of the individual's physiological state deviates from the typical pattern of the group. This quantitative value is used as the distribution deviation component.

[0038] Then, the least squares method is used to fit the trend line of physiological parameters within the current time window. The angle between this trend line and the population statistical baseline is calculated as the deviation angle. The larger the angle, the more significant the deviation of the trend direction from the expected recovery direction of the population. At the same time, the continuous duration for which the statistical trend line exceeds the preset fluctuation range of the population statistical baseline is defined as the trend maintenance duration. The temporal stability deviation component is obtained by combining the deviation angle and the trend maintenance duration, which is used to reflect the strength and persistence of the abnormal trend.

[0039] Furthermore, a physiological correlation topology is pre-established. This topology is a graph structure where nodes represent physiological parameters, edges represent functional coupling relationships between parameters (e.g., renal function and electrolytes, inflammation and body temperature), and edge weights reflect the coupling strength. When a physiological parameter in a certain dimension is identified as deviating, its topological neighbor nodes (e.g., blood urea nitrogen, urine volume, blood potassium) are traversed to detect whether these correlated parameters exhibit synchronous and coordinated change patterns. That is, the phenomenon of multiple physiological parameters changing synchronously over time or according to a specific temporal sequence is an important indicator for identifying systemic pathological states.

[0040] The identification of synergistic change patterns requires that the temporal relationship of the deviations in the associated parameters conforms to a known physiological causal chain (e.g., an increase in serum creatinine precedes an increase in blood urea nitrogen), and that the direction of change is consistent (either increasing or decreasing in the same direction). If synergistic changes are detected, cross-dimensional coupling deviation components are calculated based on the topological distance, coupling strength, and magnitude of the synergistic changes to reflect the severity of systemic pathological changes.

[0041] Finally, the deviation assessment results from the three dimensions are combined to generate the final deviation index. The sensitivity of the three deviation components varies at different postoperative recovery stages: in the hyperacute phase, physiological fluctuations are drastic, so distribution deviation is of greater concern; in the acute phase, the risk of rejection is high, so temporal stability is emphasized; and in the subacute phase and beyond, when systemic lesions appear, cross-dimensional coupling is emphasized. Therefore, based on the current recovery stage, the fusion weights of the three components are dynamically adjusted (e.g., hyperacute phase: distribution deviation component 0.5, temporal stability deviation component 0.3, cross-dimensional coupling deviation component 0.2; acute phase: distribution deviation component 0.3, temporal stability deviation component 0.4, cross-dimensional coupling deviation component 0.3), and their weighted average is calculated as the final deviation index for that dimension of physiological parameter.

[0042] This embodiment measures the similarity of individual physiological parameters within the current time window to the population statistical baseline and compares the information entropy difference as a distribution deviation component to capture the discreteness and uncertainty of individual physiological states, avoiding the loss of distributional morphology information caused by relying solely on mean comparisons. Secondly, by fitting a trend line to calculate the deviation angle from the population baseline and statistically analyzing the trend duration as a temporal stability deviation component, it achieves an assessment of the directionality and persistence of deviation from a dynamic evolutionary perspective, distinguishing between transient fluctuations and persistent pathological trends, and improving the early identification capability of progressive deterioration. Furthermore, by establishing a physiological correlation topology and detecting synergistic change patterns to calculate cross-dimensional coupled deviation components, it achieves the identification of the pathological significance of multi-parameter synergistic deviations at the system level, revealing systemic functional disorders behind single-parameter abnormalities, and improving the pathological specificity and clinical relevance of early warnings. Finally, by dynamically adjusting the fusion weights of the three components according to the postoperative recovery stage and calculating the deviation index, it achieves stage adaptation of the deviation assessment strategy to the physiological recovery pattern, enabling the early warning mechanism to focus on the most relevant risk characteristics at different stages, improving the accuracy and timeliness of early warnings. In one embodiment, the step of detecting the biomarker combination features of potential rejection based on the time-series correlation features and generating a risk level assessment result includes: The time-series correlation features are input into the early rejection prediction model, and the output is the combination features and probability values ​​of biomarkers for future acute rejection. Based on the probability values, a risk level assessment result is generated. The early rejection prediction model adopts an encoder-decoder architecture, where the encoder is a multilayer long short-term memory network and the decoder is a fully connected neural network.

[0043] In this embodiment, the identified temporal correlation features are used as input data and fed into a trained early rejection response prediction model. The input data is in the form of a time series tensor, including multi-parameter correlation features at multiple time steps. The model extracts temporal dependencies and high-order abstract features layer by layer through an encoder structure.

[0044] The early prediction model for rejection employs an encoder-decoder architecture to achieve sequence-to-sequence predictive mapping. The encoder consists of multiple stacked long short-term memory networks, each layer containing memory units and gating mechanisms (input gate, forget gate, and output gate). By processing the temporal correlation features of the input layer by layer, it captures long-distance temporal dependencies, extracts high-order temporal pattern representations, and outputs a fixed-dimensional context vector. The decoder is a fully connected neural network that receives the context vector output by the encoder. Through multiple nonlinear transformations, it maps the vector to the output space, outputting the probability value of an acute rejection reaction occurring within a specific future time window (e.g., the next 72 hours), as well as the biomarker combinations that predict this rejection reaction, i.e., which parameter combinations and change patterns are most predictive.

[0045] Finally, the probability values ​​are mapped to preset risk level classification criteria. Probability values ​​below the first threshold (e.g., 0.3) are classified as low-risk, probability values ​​between the first and second thresholds (e.g., 0.6) are classified as medium-risk, and probability values ​​above the second threshold are classified as high-risk. Different risk levels correspond to different clinical response priorities and clinical intervention recommendations.

[0046] This embodiment can feed temporal correlation features into an early rejection reaction prediction model. This model uses a multi-layer long short-term memory network as an encoder to effectively capture long-distance dependencies in temporal correlation features and enhance the ability to extract early weak pathological signals. Secondly, by using a fully connected neural network as a decoder, a nonlinear mapping is achieved from the high-level temporal representation extracted by the encoder to specific risk probabilities and biomarker combination features. The output results have both risk quantification values ​​and interpretable prediction basis.

[0047] In one embodiment, inputting the time-series correlation features into an early rejection prediction model and outputting a combination of biomarker features and probability values ​​for future acute rejection includes: The temporal correlation features are transformed into a three-dimensional spatiotemporal tensor. The correlation strength values ​​between each pair of physiological parameters at different times are calculated to fill the three-dimensional spatiotemporal tensor, forming a temporal correlation strength field. The first dimension of the three-dimensional spatiotemporal tensor represents the physiological parameter dimension, the second dimension represents the time step, and the third dimension represents the correlation strength level. The temporal correlation intensity field is fed into the first layer of the encoder's long short-term memory network to extract short-term local correlation patterns, resulting in a short-term correlation feature map. The short-term correlation feature map is then fed into the second layer of the encoder's long short-term memory network to extract medium-term trend correlation patterns, resulting in a medium-term correlation feature map. The medium-term correlation feature map is then fed into the third layer of the encoder's long short-term memory network to extract long-term evolution correlation patterns, resulting in a long-term correlation feature map. The short-term correlation feature map, medium-term correlation feature map, and long-term correlation feature map are then spliced ​​and fused along the feature dimension to obtain a multi-scale spatiotemporal correlation feature body. A biomarker combination decoupling generation channel is established in the decoder. Parallel decoupling channels with the same number of biomarker types are set up, with each channel corresponding to one biomarker. The multi-scale spatiotemporal correlation feature volume is sent to each parallel decoupling channel. Each channel calculates the occurrence confidence of the corresponding biomarker through a fully connected neural network, and gating connections between channels are set according to the physiological correlation of each biomarker. When a target biomarker channel has a confidence level exceeding the activation threshold, the response intensity of other biomarker channels that have a physiological synergistic relationship with the target biomarker channel is enhanced through gating connections, while the response of channels that have a physiological antagonistic relationship with the target biomarker channel is suppressed. The combination of the confidence levels of the biomarkers output by each channel forms a biomarker combination feature. The biomarker combination features are fed into the probability estimation layer, which calculates the similarity between the biomarker combination features and historically diagnosed rejection cases, generating the probability value of acute rejection occurring on future days.

[0048] This embodiment can calculate the correlation strength value (such as Pearson correlation coefficient) between each pair of physiological parameters at each time step and fill it into a three-dimensional spatiotemporal tensor to form a temporal correlation strength field with spatiotemporal structure.

[0049] The first layer, Long Short-Term Memory (LSTM), takes the temporal association strength field as input, focuses on short-term local time windows (e.g., 6-12 hours), captures instantaneous fluctuations in parameter associations, and outputs a short-term association feature map. The second layer receives the output from the first layer, expands the receptive field to a medium-term time span (e.g., 24-48 hours), extracts the persistent trend changes in parameter associations, and outputs a medium-term association feature map. The third layer further expands to a long-term time span (e.g., 48-72 hours), captures the slow evolution and periodic patterns of association patterns, and outputs a long-term association feature map. Finally, the feature maps output from the three layers are concatenated along the feature dimensions, fusing short, medium, and long-term multi-scale information to form a multi-scale spatiotemporal association feature body.

[0050] Based on clinically known biomarkers associated with rejection (such as interleukin-2 receptor, tumor necrosis factor-α, interferon-γ, and C-reactive protein), an equal number of parallel decoupled channels are set up, with each channel independently responsible for predicting the confidence level of the presence of one biomarker. Each channel contains a fully connected neural network that receives multi-scale spatiotemporal correlation feature volumes and outputs the confidence level of the presence of that biomarker (a probability value between 0 and 1). Simultaneously, based on the known physiological correlations (synergistic or antagonistic relationships) between the biomarkers, a gated connection network is established between the channels to form a structured channel interaction mechanism.

[0051] An activation threshold (e.g., 0.5) is set to filter out target channels with confidence levels exceeding the threshold. Based on preset gating connection weights, the responses of the target channels are propagated: for channels with a physiological synergistic relationship with the target channel, their response intensity is enhanced (e.g., multiplied by an enhancement coefficient greater than 1); for channels with a physiological antagonistic relationship, their response intensity is suppressed (e.g., multiplied by a decay coefficient less than 1). The confidence levels of each channel after gating are combined to form a structured biomarker ensemble feature. This feature not only includes the independent confidence levels of each biomarker but also implicitly encodes the physiological association constraints between biomarkers.

[0052] The probability estimation layer has a built-in historical database of confirmed rejection cases. Each case is labeled with a combination of biomarkers and the time of diagnosis. The generated combination of biomarkers is compared with the case database to calculate the similarity. The most similar historical cases are retrieved, and the probability of acute rejection occurring on future days is calculated by weighting the similarity. The higher the similarity and the more recent the case diagnosis time, the higher the probability weight of the corresponding date.

[0053] In this embodiment, a three-layer long short-term memory network is used to extract and fuse short-term local, medium-term trend, and long-term evolutionary correlation patterns, achieving collaborative capture of multi-scale spatiotemporal features. This enables the model to simultaneously perceive transient anomalies, persistent trends, and slow evolutions, improving the comprehensiveness of pathological pattern recognition across different time dimensions. Secondly, by setting parallel decoupled channels corresponding to each biomarker and establishing gating connections between channels, the dynamic regulation of synergistic channels and antagonistic channels is enhanced using gating connections. This achieves automatic reasoning about the interaction relationships between biomarkers, avoiding physiologically contradictory combinations caused by isolated predictions, ensuring the output conforms to immunological principles, and improving the biological rationality and accuracy of predictions. Furthermore, a probability estimation layer calculates the similarity to historical confirmed cases to generate risk probabilities, achieving case-based risk quantification. This transforms abstract feature matching into intuitive probability values ​​and improves the accuracy of risk warnings.

[0054] Please refer to Figure 3 As shown, embodiments of the present invention also provide an intelligent monitoring system for postoperative organ transplant patients, which may specifically include: The data acquisition module 11 is used to collect multi-dimensional physiological parameters of patients in their home environment; Module 12 is used to construct a physiological parameter baseline model based on the patient's organ transplant type and pathological characteristics. The physiological parameter baseline model adopts a sliding time window statistical method to automatically adjust the baseline threshold range according to the patient's postoperative recovery stage, and integrates the patient's preoperative physiological parameter baseline with the statistical baseline of the same type of organ transplant recipient group for personalized calibration. The calculation module 13 is used to input the multi-dimensional physiological parameters into the physiological parameter baseline model, calculate the deviation index of each dimension of physiological parameters, and when the deviation index of any dimension exceeds the warning threshold of the corresponding postoperative recovery stage, the temporal correlation characteristics of physiological parameters in different dimensions are identified by the temporal correlation algorithm, and the combination characteristics of biomarkers of potential rejection reaction are detected based on the temporal correlation characteristics to generate risk level assessment results. The matching module 14 is used to match corresponding clinical intervention recommendations based on the risk level assessment results and send the clinical intervention recommendations to the patient's mobile terminal.

[0055] Regarding the system in the above embodiments, the specific ways in which each module performs operations have been described in detail in the embodiments related to the method, and will not be elaborated here.

[0056] In one embodiment, the present invention also provides a storage medium storing computer-readable instructions, which, when executed by one or more processors, cause the one or more processors to perform the aforementioned intelligent monitoring method for postoperative organ transplant patients. The storage medium may be a ROM, random access memory (RAM), CD-ROM, magnetic tape, floppy disk, or optical data storage device, etc.

[0057] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. This computer program can be stored in a storage medium, and when executed, it can include the processes of the embodiments of the methods described above. The aforementioned storage medium can be a non-volatile storage medium such as a magnetic disk, optical disk, or read-only memory (ROM), or random access memory (RAM).

[0058] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0059] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of this patent should be determined by the appended claims.

Claims

1. A method for intelligent monitoring of organ transplant patients after surgery, characterized in that, include: Collect multidimensional physiological parameters of patients in their home environment; A physiological parameter baseline model is constructed based on the patient's organ transplant type and pathological characteristics. The physiological parameter baseline model adopts a sliding time window statistical method to automatically adjust the baseline threshold range according to the patient's postoperative recovery stage, and integrates the patient's preoperative physiological parameter baseline with the statistical baseline of the same type of organ transplant recipient group for personalized calibration. The multidimensional physiological parameters are input into the physiological parameter baseline model, and the deviation index of each physiological parameter is calculated. When the deviation index of any dimension exceeds the warning threshold of the corresponding postoperative recovery stage, the temporal correlation characteristics of physiological parameters in different dimensions are identified by the temporal correlation algorithm. Based on the temporal correlation characteristics, the combination characteristics of biomarkers of potential rejection reaction are detected, and the risk level assessment result is generated. Based on the risk level assessment results, corresponding clinical intervention recommendations are matched and sent to the patient's mobile terminal.

2. The method according to claim 1, characterized in that, The multidimensional physiological parameters include body temperature, heart rate, blood pressure, blood oxygen saturation, respiratory rate, organ function-specific parameters, tissue oxygenation index, and records of immunosuppressant administration time.

3. The method according to claim 1, characterized in that, The baseline model of physiological parameters constructed based on the patient's organ transplant type and pathological characteristics includes: The postoperative recovery phase of patients is divided into multiple phases, and each phase is configured with an independent baseline calculation strategy. The initial physiological parameter baseline model is determined based on the patient's organ transplant type. Individual data of the patient is constructed based on the patient's preoperative physiological parameters and pathological characteristics. The similarity between the individual data and the group data of the same type of organ transplant recipients is calculated at each stage. When the similarity is higher than a preset similarity threshold, the individual data weight of the baseline calculation strategy for the corresponding stage in the initial physiological parameter baseline model is increased to above a first preset value, and the group data weight is reduced accordingly. When the similarity is lower than a preset similarity threshold, the individual data weight of the baseline calculation strategy for the corresponding stage in the initial physiological parameter baseline model is reduced to below a second preset value, and the group data weight is increased accordingly to generate a physiological parameter baseline model that integrates the baseline calculation strategies of multiple stages.

4. The method according to claim 3, characterized in that, Before calculating the similarity between the individual data and the group data of similar organ transplant recipients at each stage, the method further includes: The individual data is cleaned, and missing values ​​are filled using an interpolation algorithm.

5. The method according to claim 1, characterized in that, The postoperative recovery stages of the patients are divided into the hyperacute phase, the acute phase, the subacute phase, and the chronic phase. The hyperacute phase is within three days after surgery, the acute phase is from three days to three months after surgery, the subacute phase is from three months to twelve months after surgery, and the chronic phase is more than twelve months after surgery.

6. The method according to claim 1, characterized in that, The calculation of the deviation index of each dimension of physiological parameters includes: The distribution similarity between the physiological parameters of each dimension within the current time window and the statistical baseline of the corresponding postoperative recovery stage is measured, and the information entropy difference between the probability distribution of individual physiological parameters and the probability distribution of the statistical baseline of the group is compared to quantify the degree of individual deviation from the group as the distribution deviation component. Fit the trend lines of physiological parameters in each dimension within the current time window, calculate the angle between the trend lines and the population statistical baseline, obtain the deviation angle between the current trend direction and the population statistical baseline, and calculate the duration for which the trend lines continuously exceed the fluctuation range of the population statistical baseline to obtain the trend maintenance duration. Based on the deviation angle and the trend maintenance duration, evaluate the temporal stability deviation component. A physiological correlation topology is established among physiological parameters. When a single-dimensional physiological parameter deviates, it is detected whether other-dimensional physiological parameters that have a physiological coupling relationship with the deviated physiological parameter synchronously show a cooperative change pattern. If so, the cross-dimensional coupling deviation component is calculated. The physiological correlation topology is used to reflect the functional coupling relationship between different physiological parameters. The cooperative change pattern is identified by comparing the temporal relationship of the deviations between the correlated dimensions and the consistency of the change direction. Based on the current postoperative recovery stage, different fusion weights are assigned to the distribution deviation component, the temporal stability deviation component, and the cross-dimensional coupling deviation component, and the weighted average value is calculated as the deviation index.

7. The method according to claim 1, characterized in that, The process of detecting potential rejection reactions based on the time-series correlation features and generating risk level assessment results includes: The time-series correlation features are input into the early rejection prediction model, and the output is the combination features and probability values ​​of biomarkers for future acute rejection. Based on the probability values, a risk level assessment result is generated. The early rejection prediction model adopts an encoder-decoder architecture, where the encoder is a multilayer long short-term memory network and the decoder is a fully connected neural network.

8. The method according to claim 7, characterized in that, The step of inputting the time-series correlation features into the early prediction model for rejection reactions and outputting the combination features and probability values ​​of biomarkers for future acute rejection reactions includes: The temporal correlation features are transformed into a three-dimensional spatiotemporal tensor. The correlation strength values ​​between each pair of physiological parameters at different times are calculated to fill the three-dimensional spatiotemporal tensor, forming a temporal correlation strength field. The first dimension of the three-dimensional spatiotemporal tensor represents the physiological parameter dimension, the second dimension represents the time step, and the third dimension represents the correlation strength level. The temporal correlation intensity field is fed into the first layer of the encoder's long short-term memory network to extract short-term local correlation patterns, resulting in a short-term correlation feature map. The short-term correlation feature map is then fed into the second layer of the encoder's long short-term memory network to extract medium-term trend correlation patterns, resulting in a medium-term correlation feature map. The medium-term correlation feature map is then fed into the third layer of the encoder's long short-term memory network to extract long-term evolution correlation patterns, resulting in a long-term correlation feature map. The short-term correlation feature map, medium-term correlation feature map, and long-term correlation feature map are then spliced ​​and fused along the feature dimension to obtain a multi-scale spatiotemporal correlation feature body. A biomarker combination decoupling generation channel is established in the decoder. Parallel decoupling channels with the same number of biomarker types are set up, with each channel corresponding to one biomarker. The multi-scale spatiotemporal correlation feature volume is sent to each parallel decoupling channel. Each channel calculates the occurrence confidence of the corresponding biomarker through a fully connected neural network, and gating connections between channels are set according to the physiological correlation of each biomarker. When a target biomarker channel has a confidence level exceeding the activation threshold, the response intensity of other biomarker channels that have a physiological synergistic relationship with the target biomarker channel is enhanced through gating connections, while the response of channels that have a physiological antagonistic relationship with the target biomarker channel is suppressed. The combination of the confidence levels of the biomarkers output by each channel forms a biomarker combination feature. The biomarker combination features are fed into the probability estimation layer, which calculates the similarity between the biomarker combination features and historically diagnosed rejection cases, generating the probability value of acute rejection occurring on future days.

9. The method according to claim 1, characterized in that, The clinical intervention recommendations matched based on the risk level assessment results include: Based on the risk level assessment results, matching clinical intervention recommendations are retrieved from a pre-built knowledge base. These recommendations include clinical treatment plans, examination recommendations, and lifestyle adjustment guidelines corresponding to different risk levels.

10. A postoperative intelligent monitoring system for organ transplant patients, characterized in that, include: The data acquisition module is used to collect multi-dimensional physiological parameters of patients in their home environment; The module is used to construct a baseline model of physiological parameters based on the patient's organ transplant type and pathological characteristics. The baseline model of physiological parameters adopts a sliding time window statistical method to automatically adjust the baseline threshold range according to the patient's postoperative recovery stage, and integrates the patient's preoperative physiological parameter baseline with the statistical baseline of the same type of organ transplant recipient group for personalized calibration. The calculation module is used to input the multi-dimensional physiological parameters into the physiological parameter baseline model, calculate the deviation index of each dimension of physiological parameters, and when the deviation index of any dimension exceeds the warning threshold of the corresponding postoperative recovery stage, the temporal correlation algorithm is used to identify the temporal correlation characteristics of physiological parameters in different dimensions, and the biomarker combination characteristics of potential rejection reaction are detected based on the temporal correlation characteristics to generate risk level assessment results. The matching module is used to match corresponding clinical intervention recommendations based on the risk level assessment results and send the clinical intervention recommendations to the patient's mobile terminal.

Images