Method and system for dynamic assessment of occult hemorrhage risk after renal biopsy based on multi-source data fusion

Abstract

The application relates to the technical field of medical information processing, and discloses a method and system for dynamically evaluating post-renal biopsy occult bleeding risk based on multi-source data fusion, wherein the method comprises the following steps: constructing a standardized uniform observation vector sequence; constructing a short-time instant risk model; constructing a mechanism to establish a long-time lag risk transfer model; generating a hierarchical treatment decision and a review scheduling plan; gain correcting the risk probability; and outputting a final occult bleeding risk grade. Compared with the scheme for postoperative bleeding risk in the prior art, which only performs static risk evaluation based on a single time section index, especially in the case that postoperative vital signs are temporarily stable but there is delayed hemoglobin decline, the technical problem that the continuous prediction of the risk transfer process of'short-time stability-long-time lag deterioration' cannot be realized is solved. Since the time-varying state transfer probability matrix is introduced in the application, and the risk transfer acceleration is determined, the sensitivity of postoperative risk identification is improved.

Description

Technical Field

[0001] This invention relates to the field of medical information processing technology, and in particular to a method and system for dynamic assessment of the risk of occult bleeding after renal biopsy based on multi-source data fusion. Background Technology

[0002] Currently, the assessment of bleeding risk after renal biopsy mainly relies on a comprehensive analysis of postoperative vital sign changes, single hemoglobin test results, and physician's experience. However, this type of assessment is usually based on static judgments using single-time-section data, lacking the ability to continuously model the evolution of postoperative risk and failing to characterize the dynamic changes in occult bleeding at different time stages.

[0003] For example, in the early postoperative period (0–6 hours), some patients may have relatively stable vital signs such as systolic blood pressure, heart rate, and mean arterial pressure, but may experience a significant decrease in hemoglobin or imaging hematoma expansion at 12 hours or even later. Current technologies often use fixed thresholds or single logistic regression models for risk prediction, failing to effectively capture the risk transfer process of "short-term stability – long-term delayed deterioration," leading to delayed bleeding warnings. Furthermore, during follow-up examinations, traditional methods typically correct risks by recalculating risk scores or manually adjusting risk levels, lacking a systematic probability calibration mechanism. This makes them susceptible to occasional abnormal fluctuations or single-test errors, reducing the stability and consistency of risk assessment. In addition, existing methods generally do not perform unified timeline alignment and structured fusion processing on multi-source data (vital signs, laboratory test data, imaging features, and puncture baseline parameters), making it impossible to establish a continuous risk evolution model across time scales. In the long-term risk prediction stage, there is also a lack of probabilistic modeling methods for risk state transition patterns, making it difficult to generate recursive risk state distribution results.

[0004] Therefore, there is an urgent need for a dynamic assessment method for the risk of occult bleeding after renal biopsy based on multi-source data fusion, in order to improve the sensitivity, continuity and predictive stability of occult bleeding risk identification, reduce the risk of delayed diagnosis and optimize the postoperative triage and treatment decision-making process. Summary of the Invention

[0005] To address the aforementioned technical shortcomings, the purpose of this invention is to propose a dynamic assessment method for the risk of occult bleeding after renal biopsy based on multi-source data fusion. This method aims to solve the technical problem that existing schemes that rely solely on a single time-section indicator for static risk assessment of postoperative bleeding risk are unable to continuously predict the risk transfer process of "short-term stability to long-term delayed deterioration" when postoperative vital signs are temporarily stable but there is a delayed decrease in hemoglobin.

[0006] To solve the above-mentioned technical problems, the present invention adopts the following technical solution: The present invention provides a method for dynamic assessment of the risk of occult bleeding after renal biopsy based on multi-source data fusion.

[0007] The method for dynamic assessment of the risk of occult bleeding after renal biopsy based on multi-source data fusion includes:

[0008] Step S10: Obtain postoperative multi-source observation data, and perform multi-source feature construction task based on postoperative multi-source observation data using a unified time axis alignment and baseline difference fusion mechanism, and output a standardized unified observation vector sequence;

[0009] Step S20: Based on the standardized unified observation vector sequence, a short-time exponential decay weighted fusion mechanism is used to perform a short-time instantaneous risk modeling task and output the short-time endpoint risk status;

[0010] Step S30: Based on the short-term endpoint risk state, a time-varying state transition probability construction mechanism is used to perform the long-term lag risk transition modeling task, and output the long-term risk state distribution results;

[0011] Step S40: Based on the long-term risk status distribution results, a risk transfer acceleration judgment mechanism is used to execute the hierarchical disposal decision-making construction task and output the review and scheduling plan;

[0012] Step S50: Based on the review scheduling plan, use the consistency calibration feedback mechanism to perform the risk probability correction task and output the final occult bleeding risk level.

[0013] Preferably, step S10, which involves acquiring postoperative multi-source observation data, performing a multi-source feature construction task based on the postoperative multi-source observation data using a unified time axis alignment and baseline difference fusion mechanism, and outputting a standardized unified observation vector sequence, specifically includes:

[0014] Step S101: Acquire postoperative multi-source observation data, including vital sign data collected within 0 to 6 hours postoperatively with a sampling step of 5 minutes, vital sign data collected within 6 to 72 hours postoperatively with a sampling step of 30 minutes, laboratory hemoglobin test data, bedside ultrasound imaging feature data, and puncture baseline parameters; among them, vital sign data include systolic blood pressure (SBP), diastolic blood pressure (DBP), heart rate (HR), and mean arterial pressure (MAP); laboratory hemoglobin test data includes hemoglobin concentration (Hb); bedside ultrasound imaging feature data includes the maximum diameter of the hematoma; and puncture baseline parameters include the number of punctures, needle diameter, and puncture depth.

[0015] Step S102: First, perform time axis alignment processing on the postoperative multi-source observation data, output the aligned postoperative multi-source observation data, and then extract the hemoglobin change based on the laboratory hemoglobin detection data. Based on the change in hemoglobin Hemoglobin production trend characteristic variables;

[0016] Step S103: Obtain laboratory testing delay information, encode the laboratory testing delay information as missing label variables, and perform feature fusion and splicing of missing label variables, hemoglobin trend feature variables and post-alignment multi-source observation data to output a standardized unified observation vector sequence.

[0017] Preferably, step S20, which involves performing a short-term immediate risk modeling task based on a standardized unified observation vector sequence using a short-term exponential decay weighted fusion mechanism and outputting the short-term endpoint risk state, specifically includes:

[0018] Step S201: Preset a short window and exponentially decaying weights Based on exponential decay weights For short window The standardized unified observation vector sequence within the sequence undergoes exponential decay weighted fusion to generate a weighted short-time feature vector. ;

[0019] Step S202: Introduce the logistic probability mapping function, and apply it to the weighted short-time feature vector. Perform risk probability calculations and output short-term real-time risk probabilities;

[0020] Step S203: Based on the preset multi-threshold gating judgment rules, the short-term immediate risk probability is discretized into four levels of short-term risk status; all four levels of short-term risk status within the short-term window are merged, and the short-term endpoint risk status is finally output.

[0021] Preferably, step S203, the step of discretizing the short-term immediate risk probability into four levels of short-term risk states based on a preset multi-threshold gating judgment rule, specifically includes:

[0022] Step S2031: Preset a multi-threshold set, the multi-threshold set including a first risk threshold. Second risk threshold and the third risk threshold Among them, the first risk threshold Used to distinguish between stable and low-risk states; second risk threshold Used to distinguish between low-risk and medium-risk states; third risk threshold Used to distinguish between medium-risk and high-risk states; and meets the following conditions: ;

[0023] Step S2032: Based on a multi-threshold set, and combined with a nonlinear mapping model with time decay weights, perform gated discretization processing on the short-term instantaneous risk probability, and output the discrete risk probability p;

[0024] when At that time, the determination of the short-term immediate risk probability corresponds to the first short-term risk state. ;

[0025] when At that time, the determination of the short-term immediate risk probability corresponds to the second short-term risk state. ;

[0026] when At that time, the determination of the short-term immediate risk probability corresponds to the third short-term risk state. ;

[0027] when At that time, the determination of the short-term immediate risk probability corresponds to the fourth short-term risk state. ;

[0028] Step S2033: Output a Level 4 short-term risk status based on the above judgment results.

[0029] Preferably, step S30, which involves performing a long-term lag risk transition modeling task based on a time-varying state transition probability construction mechanism using a short-term endpoint risk state, and outputting the long-term risk state distribution results, specifically includes:

[0030] Step S301: Encode the short-term endpoint risk state into the long-term initial state and generate the initial state distribution;

[0031] Step S302: Obtain the standardized unified observation vector sequence of the past 6 hours, and construct a long-term weighted feature vector based on the standardized unified observation vector sequence of the past 6 hours. Construct a time-varying state transition probability matrix based on the long-term weighted feature vector using the exponential mapping Markov chain transition rate modeling method.

[0032] Step S303: Based on the initial state distribution and the time-varying state transition probability matrix, the future state distribution is calculated recursively using matrix multiplication, and the long-term risk state distribution result is output.

[0033] Preferably, step S40, which involves using a risk transfer acceleration determination mechanism based on long-term risk state distribution results to construct a tiered disposal decision-making task and output a review scheduling plan, specifically includes:

[0034] Step S401 extracts the short-term instantaneous risk probability difference between adjacent time points in the long-term risk state distribution results as a risk transfer acceleration index;

[0035] Step S402: Based on the threshold range of the risk transfer acceleration index, perform a classification judgment and generate the corresponding disposal level;

[0036] Step S403: Generate a treatment decision vector and a review scheduling plan according to the corresponding treatment level. The review scheduling plan includes at least a vital signs monitoring frequency adjustment plan, a hemoglobin retesting time plan, and an image review plan.

[0037] Preferably, step S50, which involves executing a risk probability correction task based on a consistency calibration feedback mechanism according to the review scheduling plan and outputting the final occult bleeding risk level, specifically includes:

[0038] Step S501: After executing the review scheduling plan, obtain the review hemoglobin change value, review image hematoma change information, and review hematoma maximum diameter mapping value;

[0039] Step S502: Based on the changes in hemoglobin, changes in hematoma in the re-examination images, and the maximum diameter mapping of the re-examination hematoma, a re-examination calibration factor is constructed using the proportional gain calibration method. Based on the re-examination calibration factor, the long-term risk status distribution results are adjusted by gain, and the final occult bleeding risk level is output.

[0040] This invention also provides a dynamic assessment system for the risk of occult bleeding after renal biopsy based on multi-source data fusion, including:

[0041] The multi-source data structuring module is used to acquire postoperative multi-source observation data. Based on the postoperative multi-source observation data, a unified time axis alignment and baseline difference fusion mechanism is used to perform multi-source feature construction tasks and output a standardized unified observation vector sequence.

[0042] The short-term risk modeling module is used to perform short-term real-time risk modeling tasks based on a standardized unified observation vector sequence and a short-term exponential decay weighted fusion mechanism, and output the short-term endpoint risk status.

[0043] The long-term risk transfer modeling module is used to perform long-term lag risk transfer modeling tasks based on short-term endpoint risk states using a time-varying state transition probability construction mechanism, and outputs long-term risk state distribution results.

[0044] The risk acceleration decision module is used to construct a graded disposal decision based on the risk transfer acceleration judgment mechanism according to the long-term risk status distribution results, and outputs a review and scheduling plan.

[0045] The risk consistency calibration module is used to perform risk probability correction tasks according to the review scheduling plan using a consistency calibration feedback mechanism, and output the final occult bleeding risk level.

[0046] The present invention also provides a device for dynamic assessment of the risk of occult bleeding after renal biopsy based on multi-source data fusion, comprising: a memory, a processor, and a program for dynamic assessment of the risk of occult bleeding after renal biopsy based on multi-source data fusion stored in the memory and executable on the processor. When the program for dynamic assessment of the risk of occult bleeding after renal biopsy based on multi-source data fusion is executed by the processor, a method for dynamic assessment of the risk of occult bleeding after renal biopsy based on multi-source data fusion is realized.

[0047] The present invention also provides a computer program product, including a dynamic assessment program for the risk of occult bleeding after renal biopsy based on multi-source data fusion. When the program is executed by a processor, it implements the method for dynamic assessment of the risk of occult bleeding after renal biopsy based on multi-source data fusion.

[0048] The beneficial effects of this invention are as follows: By constructing a two-stage risk modeling structure with short-term window and long-term lag, employing an exponential decay weighted fusion mechanism to extract vital signs and hemoglobin trend features, and using a time-varying state transition probability matrix to characterize the risk state evolution process, this invention achieves continuous modeling of the risk transition process from "short-term stability" to "long-term deterioration." Compared to traditional static assessment methods based on single-time-section data or fixed-time-point threshold judgments, this invention can capture the lag characteristic change trend of occult hemorrhage, improve the early warning capability for the risk of delayed hemoglobin drop, and enhance the temporal continuity and stability of risk prediction.

[0049] This invention introduces a risk transfer acceleration index based on long-term risk prediction, and performs first-order difference analysis on the trend of risk probability changes to avoid misjudgments caused by single-point abnormal fluctuations. Simultaneously, a proportional gain calibration method is used in the review stage to convert changes in hemoglobin and hematoma information from imaging into probability gain factors, dynamically correcting the predicted risk. Compared to existing technologies that recalculate or manually adjust risk levels after review, this invention achieves continuous modulation of the probability space while maintaining the stability of the original state transition model structure, improving the consistency of risk assessment. Attached Figure Description

[0050] To more clearly illustrate the technical solutions in the embodiments of the present invention 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 only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0051] Figure 1This is a flowchart illustrating the first embodiment of the method for dynamic assessment of the risk of occult bleeding after renal biopsy based on multi-source data fusion according to the present invention.

[0052] Figure 2 This is a schematic diagram of postoperative multi-source observation data and early weak trend signals from the first embodiment of the dynamic assessment method for the risk of occult bleeding after renal biopsy based on multi-source data fusion of the present invention.

[0053] Figure 3 This diagram illustrates the short-term risk response advantages of the exponential decay weighting mechanism in the first embodiment of the dynamic assessment method for the risk of occult bleeding after renal biopsy based on multi-source data fusion, according to the present invention.

[0054] Figure 4 This is a schematic diagram of the short-term risk state trajectory of the multi-threshold gating judgment output of the first embodiment of the dynamic assessment method for the risk of occult bleeding after renal biopsy based on multi-source data fusion of the present invention.

[0055] Figure 5 This is a schematic diagram of the device for the dynamic assessment of the risk of occult bleeding after renal biopsy based on multi-source data fusion, as described in this invention. Detailed Implementation

[0056] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0057] Example 1: As Figure 1 The diagram shown is a flowchart of the first embodiment of the method for dynamic assessment of the risk of occult bleeding after renal biopsy based on multi-source data fusion according to the present invention. The first embodiment of the method for dynamic assessment of the risk of occult bleeding after renal biopsy based on multi-source data fusion according to the present invention is presented.

[0058] In the first embodiment, the method for dynamic assessment of the risk of occult bleeding after renal biopsy based on multi-source data fusion includes:

[0059] Step S10: Obtain postoperative multi-source observation data, and perform multi-source feature construction task based on postoperative multi-source observation data using a unified time axis alignment and baseline difference fusion mechanism, and output a standardized unified observation vector sequence;

[0060] It should be noted that the "unified time axis alignment and baseline differential fusion mechanism" refers to the mechanism used when there are differences in sampling frequency, timestamp accuracy, and data delay output among multi-source heterogeneous data. This is achieved by establishing a unified master time axis, reconstructing various types of observation data in a synchronized manner, and then introducing baseline differential processing and structured feature stitching after synchronization. This results in a unified observation vector sequence with temporal consistency and trend expression capabilities. Unified time axis alignment includes resampling of high-frequency vital sign data, time mapping of low-frequency laboratory test data, and time matching of image feature data. Baseline differential fusion includes constructing change features for hemoglobin concentration, performing standardization processing based on individual baselines for vital signs, and constructing relative change features for image diameters. This includes not only numerical alignment but also structured encoding of detection delay states to distinguish between "no change" and "not yet detected" semantic states, ensuring that subsequent model input data possesses complete semantic expressive capabilities.

[0061] Understandably, by unifying the timeline alignment, data from different devices and systems become comparable at the same point in time, thus avoiding risk assessment bias caused by inconsistent sampling times. Baseline difference fusion allows the model to focus on the trend of indicator changes rather than absolute numerical differences, enhancing comparability between different patients. Missing variable embedding enables the model to identify uncertainties caused by detection delays, thereby improving the overall stability and robustness of risk modeling. Technically, this step achieves structural unification, time synchronization, and trend enhancement of multi-source heterogeneous data, providing a continuous temporal input foundation for subsequent short-term risk assessment and long-term risk transfer modeling.

[0062] It should be understood that, compared to traditional methods that directly use raw vital sign values ​​or single hemoglobin test results for risk scoring, this step no longer relies on absolute values ​​at isolated time points. Instead, it strengthens the expression of temporal evolution characteristics through a unified time structure and trend differencing. Traditional methods often ignore the objective fact that laboratory testing has a time delay, assuming risk stability in the untested stage, which can easily lead to early risk underestimation. This step, however, explicitly encodes the detection delay state, enabling the model to distinguish between "value not decreasing" and "value not yet obtained," thereby reducing the risk of misjudgment. Furthermore, traditional methods do not systematically reconstruct data from different sampling frequencies, which can easily lead to breaks in trend judgment. This step, however, reconstructs a continuous time series structure through a unified time axis, effectively improving the continuity and accuracy of risk evolution modeling. For example, if a patient's vital signs are sampled every 5 minutes within 2 hours post-surgery, but hemoglobin test results are only output once after 4 hours, without time alignment, the model might mistakenly assume hemoglobin remains stable during the 2-4 hour post-surgery period. By using a unified timeline alignment mechanism, the hemoglobin data status can be marked at each unified time point, and the detection delay information can be structured and encoded, allowing the model to identify that the time period is in a "detection pending update state." As another example, if two patients have pre-operative hemoglobin levels of A and B (A being greater than B), and both decreased by 2 g / L post-surgery, directly using absolute values ​​for modeling would still show the former as normal while the latter might be misjudged as having higher risk. By constructing hemoglobin change characteristics through a baseline differential fusion mechanism, the model can uniformly identify the "decreasing trend" as a risk signal, thereby improving the consistency of risk identification.

[0063] Step S20: Based on the standardized unified observation vector sequence, a short-time exponential decay weighted fusion mechanism is used to perform a short-time instantaneous risk modeling task and output the short-time endpoint risk status;

[0064] It should be noted that the "short-term exponential decay weighted fusion mechanism" refers to assigning different weights to historical observations in a standardized unified observation vector sequence according to their distance from the current time within a preset short-term window. Observations closer to the current time have higher weights, and those farther away have lower weights, thus constructing a time-sensitive weighted short-term feature vector. The exponential decay weighting method uses an exponential decay function as the weight allocation rule, controlling the influence of historical data on the current risk assessment through a preset decay coefficient. The short-term window is used to depict the dynamic fluctuation trend of vital signs and hemoglobin changes within a short period. Based on the weighted short-term feature vector, a logistic probability mapping function is introduced to construct the short-term instantaneous risk probability, and a four-level short-term risk state is output through multi-threshold gating rules. Finally, the risk states at each time point within the short-term window are merged to form the short-term endpoint risk state.

[0065] Understandably, this step strengthens the influence of "recent changes" through a time-weighted mechanism, making the model more sensitive to fluctuations in vital signs, trends in hemoglobin decline, and changes in imaging signals. When a patient experiences a slight decrease in blood pressure or a gradual increase in heart rate within a short period, the exponentially decaying weighted mechanism amplifies these changes in the feature vector, thereby improving the response speed of immediate risk prediction. Simultaneously, by fusing risk states from multiple time points within a short window, the interference of a single abnormal fluctuation on the final judgment can be reduced, improving the stability and reliability of short-term risk state output.

[0066] For example, such as Figure 2 The figure shows the continuous trends of heart rate (HR), systolic blood pressure (SBP), and mean arterial pressure (MAP) in patients from 0 to 6 hours post-surgery, overlaid with discrete observation points of delayed hemoglobin (Hb) and the maximum diameter of the hematoma (D). An early, subtle trend phase from 2.0 to 3.5 hours is also marked. The figure reveals that during this phase, the patient's vital signs did not show drastic abnormalities, and the values ​​remained within clinically acceptable ranges. However, the heart rate showed a sustained, slow upward trend, while systolic blood pressure and mean arterial pressure showed a gradual downward trend, reflecting a "trend shift" rather than a "transient abnormality." Traditional methods typically rely on single-time-point thresholds for judgment, such as triggering alarms only when systolic blood pressure falls below a certain fixed value or heart rate rises above a certain fixed value. Therefore, it is difficult to identify such gradual changes in a timely manner. This invention, through a unified time axis alignment mechanism, structurally integrates high-frequency vital sign data, low-frequency laboratory hemoglobin detection data, and intermittent imaging data, forming a unified observation vector sequence across all data types on the same time dimension, thereby continuously depicting the "dynamic trend." Therefore, even if hemoglobin levels have not yet decreased significantly, potential bleeding signals can still be detected in advance through the trend evolution of vital signs, providing a continuous dynamic basis for subsequent short-term risk modeling. For example... Figure 3The figure shows a comparison of two methods for calculating short-term, real-time risk probability: one curve represents the risk probability calculated using an exponentially decaying weighted fusion mechanism, and the other uses a simple moving average. The threshold lines in multi-threshold gating are also marked. It is clear from the figure that when a weak trend begins to appear after 2.0 hours, the exponentially decaying weighted curve rises significantly faster than the simple mean curve and crosses the medium-risk threshold earlier. This is because the exponentially decaying weighted mechanism uses a time-distance-related weighting rule, giving greater weight to data closer to the current moment and gradually decreasing weight to historical data. Therefore, as heart rate gradually increases and blood pressure gradually decreases, the contribution of recent data to the risk probability is amplified, thus accelerating the response speed of the risk curve. In contrast, the simple moving average assigns the same weight to all data within the window, averaging data from the early stable phase with recent abnormal trends, resulting in a smoothed risk curve and a delayed response. Figure 4 The figure illustrates the evolution trajectory after discretizing continuous risk probabilities into four short-term risk states using multi-threshold gating rules. As can be seen from the figure, in the early, weak trend stage, the risk state gradually rises from a lower level to a higher level, and then gradually declines after the trend eases. The entire state transition process exhibits continuous evolution characteristics, rather than frequent jumps. This invention uses a three-threshold structure to form a four-level risk hierarchy, mapping continuous probabilities to four states: "stable," "low risk," "medium risk," and "high risk." It also combines gating rules to constrain state elevation, avoiding frequent state switching caused by single abnormal fluctuations. Compared to traditional binary classification or single-threshold methods, this invention can not only determine the existence of risk but also reflect the stage degree of risk and its evolution path.

[0067] Step S30: Based on the short-term endpoint risk state, a time-varying state transition probability construction mechanism is used to perform the long-term lag risk transition modeling task, and output the long-term risk state distribution results;

[0068] It should be noted that the "time-varying state transition probability construction mechanism" refers to the following: Based on the short-term endpoint risk state as the initial state for the long-term stage, instead of using a fixed state transition matrix, a long-term weighted feature vector is constructed by combining a standardized unified observation vector sequence over a recent period. Based on this feature vector, a state transition probability matrix that changes over time is dynamically generated. The state transition probability matrix is ​​a fourth-order matrix, where each element represents the conditional probability of transitioning from one risk state to another at the current moment, and the sum of the probabilities in each row is 1. The long-term weighted feature vector is obtained by exponentially smoothing the unified observation vector over several historical time steps. Its weight coefficients reflect the time decay relationship, thus emphasizing the impact of recent trends on future risk evolution. Furthermore, a functional mapping relationship is established between the long-term weighted feature vector and preset transition weight parameters, making the transition probabilities between states a time-dependent function. Finally, by recursively calculating the risk state distribution, the long-term risk state distribution results for future time periods are obtained.

[0069] Understandably, the technical effect of this step lies in using the endpoint risk state identified in the short-term window as the starting point for long-term evolution, and characterizing the delayed manifestation process of occult bleeding risk through a trend-driven dynamic transition probability mechanism. Since occult bleeding after renal biopsy often exhibits characteristics such as delayed hemoglobin decline, delayed imaging changes, and a hemodynamic compensation phase, relying solely on the current risk level is insufficient to accurately predict future risk evolution. This invention, by correlating the transition probability with recent characteristic trends, makes risk evolution depend not only on the current state but also on the trend direction and rate of change, thereby achieving dynamic modeling of the "short-term stability—long-term deterioration" transition process and improving the prospective identification capability of delayed bleeding risk.

[0070] It should be understood that, compared to traditional homogeneous Markov models that use a fixed transition probability matrix, this invention constructs a feature-driven non-homogeneous Markov mechanism, making the state transition probability a function of time. Traditional methods typically calculate fixed transition probabilities based on historical population statistical frequencies, reflecting only the overall average transition pattern and failing to capture the real-time trend changes of the current individual. This invention establishes a functional relationship between the transition probability and a real-time long-term weighted feature vector, thereby increasing the probability of transitioning to a medium- or high-risk state even when heart rate is continuously rising, blood pressure is continuously falling, but hemoglobin has not yet significantly decreased. This avoids the problem of risk identification lag and significantly improves individualized prediction and dynamic response capabilities.

[0071] Step S40: Based on the long-term risk status distribution results, a risk transfer acceleration judgment mechanism is used to execute the hierarchical disposal decision-making construction task and output the review and scheduling plan;

[0072] It should be noted that the "risk transfer acceleration determination mechanism" refers to: after obtaining the long-term risk status distribution results, not only analyzing the instantaneous probability value of a certain risk level, but also quantitatively calculating the changing trend of the risk status over time, including the first-order rate of change and the second-order trend of risk probability, thereby constructing a comprehensive judgment index reflecting the speed and acceleration of risk evolution. Risk transfer acceleration includes the time difference calculation results of medium-risk and high-risk probabilities, used to characterize the dynamic characteristics of risk transfer from low to high levels; the hierarchical treatment decision-making construction task includes classifying treatment levels according to the risk acceleration interval, generating monitoring frequency adjustment plans, scheduling hemoglobin re-examination times, and combining imaging review strategies; the review scheduling plan includes, but is not limited to, specific scheduling measures such as shortening the interval of vital sign monitoring, advancing laboratory testing time, increasing the number of ultrasound re-examinations, and activating the notification mechanism for responsible medical personnel.

[0073] Understandably, integrating the "absolute level of risk probability" with the "dynamic speed of risk evolution" can identify situations where risk is in a rapidly deteriorating phase. Occult bleeding often progresses rapidly within a short period, making it difficult to determine whether a sudden change in risk is imminent based solely on a probability value at a given moment. This invention, by introducing a risk transfer acceleration index, can detect risk escalation trends in advance and initiate enhanced monitoring or review measures before the probability reaches its highest threshold, thereby improving the timeliness and safety of risk intervention. Compared to traditional techniques that only use fixed probability thresholds for grading, this invention adds risk change rate and acceleration assessment steps, making the decision-making logic more consistent with the actual clinical risk evolution. Traditional methods typically trigger treatment strategies only after the high-risk probability exceeds a preset value, failing to distinguish between "stable high-risk states" and "rapidly rising risk states," easily leading to intervention delays or unreasonable resource allocation. This invention, through a dynamic trend assessment mechanism, adjusts the treatment level during the rapid risk escalation phase, thus avoiding the lag problem caused by relying solely on static numerical judgments. For example, a patient's high-risk probability was 0.25 at 10 hours post-surgery, rising to 0.33 at 11 hours, and further increasing to 0.48 at 12 hours. Although the probabilities at 10 and 11 hours had not yet reached the traditional high-risk threshold, their continuous upward trend was obvious, and the rate of increase gradually widened, indicating that the risk was in an accelerating phase. According to the risk transfer acceleration determination mechanism of this invention, the patient could be identified as being in the "risk acceleration phase" at 11 hours, and the review scheduling plan would be automatically upgraded from the routine monitoring mode to the enhanced monitoring mode. For example, the hemoglobin retest time would be advanced by 2 hours, the number of bedside ultrasound retests would be increased, and the interval between vital sign collections would be shortened. Subsequently, the patient showed a significant decrease in hemoglobin at 13 hours, and the early-triggered scheduling decision was highly consistent, verifying the practical effect of this step in early risk identification and decision optimization.

[0074] Step S50: Based on the review scheduling plan, use the consistency calibration feedback mechanism to perform the risk probability correction task and output the final occult bleeding risk level.

[0075] It should be noted that the "consistency calibration feedback mechanism" refers to the following: after completing the review scheduling plan and obtaining new objective test results, a multi-dimensional consistency comparison is performed between the reviewed hemoglobin changes, reviewed imaging hematoma changes, and the reviewed hematoma maximum diameter mapping results and the previous long-term risk status distribution results. Based on the proportional gain principle, a review calibration factor is constructed to correct the original risk probability in both direction and magnitude. The consistency comparison includes numerical consistency judgment (whether the reviewed indicators match the predicted probability level), trend consistency judgment (whether the direction of change in the review results is consistent with the predicted trend), and magnitude consistency judgment (whether the reviewed change has reached the expected deterioration level). The review calibration factor is used to amplify or attenuate the gain of the original risk distribution and is normalized after correction to maintain the rationality of the risk distribution probability structure. Through this mechanism, a closed-loop correction is achieved between the risk prediction results and the actual review data.

[0076] Understandably, the technical effect of this step lies in constructing a dynamic adaptive risk correction mechanism, transforming the risk assessment system from a one-way predictive model into one with "self-learning feedback adjustment" capabilities. When the review results align with the predicted trend, the confidence level of the risk grade is enhanced; when the review results deviate from the predicted trend, the corresponding risk probability is automatically reduced or adjusted, thereby preventing the continuous accumulation of risk misjudgments. Especially in clinical scenarios like occult bleeding, which exhibits high individual variability and dynamic fluctuations, the predictive model may be affected by occasional fluctuations in vital signs. By introducing review consistency calibration, the impact of random fluctuations on the final risk grade can be effectively suppressed, improving the overall assessment stability.

[0077] It should be understood that, compared to methods that rely solely on predetermined model outputs without subsequent feedback corrections, this invention achieves online dynamic adjustment of risk probabilities by establishing a closed-loop structure of prediction-review-calibration-re-output. Traditional methods typically maintain the risk level unchanged after generation; if the initial prediction is too high or too low, it may lead to over-allocation of resources or delayed intervention in subsequent management. This invention, by introducing a review and calibration factor, updates the risk probability in real time with actual test results, thereby improving prediction accuracy and optimizing clinical resource allocation. This mechanism not only improves the accuracy of risk assessment but also enhances the tolerance for abnormal test results and the responsiveness to trend reversals.

[0078] For example, in a patient's postoperative 12-hour forecast, the high-risk probability was 0.55, classified as a "high-risk warning." A subsequent review and scheduling plan was implemented. The reviewed hemoglobin level decreased by only 4 g / L, and the hematoma diameter on imaging did not increase, which was not entirely consistent with the predicted deterioration trend. Using a consistency calibration mechanism, the review calibration factor was calculated to be 0.72. The high-risk probability was attenuated and adjusted to 0.40, while the proportion of the medium-risk probability was increased, ultimately correcting the risk level to "medium-risk enhanced monitoring." This adjustment avoided unnecessary intensive intervention measures. In another case, the predicted high-risk probability was 0.48 at postoperative 10 hours, but the reviewed hemoglobin level decreased by 16 g / L, and the hematoma diameter on imaging significantly increased, highly consistent with the predicted trend. The calculated calibration factor was 1.35. The high-risk probability was amplified to 0.65, directly outputting a "high-risk emergency treatment" level, and blood tests and imaging reviews were arranged in advance. Subsequent clinical results showed that the patient developed significant bleeding signs at 13 hours, validating the effectiveness of the calibration mechanism in risk enhancement assessment. Therefore, this step, through a consistent calibration feedback mechanism, deeply integrates the predictive model with real-time verification data to form a complete closed-loop control structure. This significantly improves the dynamic adaptability, stability, and clinical decision reliability of occult bleeding risk assessment, demonstrating a technical effect that is significantly superior to traditional static risk output models.

[0079] Example 2: Furthermore, the dynamic assessment system for the risk of occult bleeding after renal biopsy based on multi-source data fusion provided by the present invention employs the dynamic assessment method for the risk of occult bleeding after renal biopsy based on multi-source data fusion described in the above examples, and can solve the technical problem of dynamic assessment of the risk of occult bleeding after renal biopsy based on multi-source data fusion. The beneficial effects of the dynamic assessment system for the risk of occult bleeding after renal biopsy based on multi-source data fusion provided by the present invention are the same as those of the dynamic assessment method for the risk of occult bleeding after renal biopsy based on multi-source data fusion provided in the above examples, and other technical features of the dynamic assessment system for the risk of occult bleeding after renal biopsy based on multi-source data fusion are the same as those disclosed in the methods of the above examples, and will not be repeated here.

[0080] Example 3: This invention provides a dynamic assessment device for the risk of occult bleeding after renal biopsy based on multi-source data fusion. Please refer to... Figure 5The device for dynamic assessment of the risk of occult bleeding after renal biopsy based on multi-source data fusion includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute the method for dynamic assessment of the risk of occult bleeding after renal biopsy based on multi-source data fusion described in Embodiment 1 above. The device for dynamic assessment of the risk of occult bleeding after renal biopsy based on multi-source data fusion in this embodiment of the invention may include, but is not limited to, mobile terminals such as mobile phones, laptops, digital radio receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Description), PMPs (Portable Media Players), in-vehicle terminals (e.g., in-vehicle navigation terminals), and fixed terminals such as digital TVs and desktop computers. The device for dynamic assessment of the risk of occult bleeding after renal biopsy based on multi-source data fusion is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of the invention. The dynamic assessment device for the risk of occult bleeding after renal biopsy based on multi-source data fusion may include a processing unit 1001 (e.g., a central processing unit, a graphics processing unit, etc.), which can perform various appropriate actions and processes according to a program stored in a read-only memory 1002 or a program loaded from a storage device 1003 into a random access memory 1004. The random access memory 1004 also stores various programs and data required for the operation of the dynamic assessment device for the risk of occult bleeding after renal biopsy based on multi-source data fusion. The processing unit 1001, the read-only memory 1002, and the random access memory 1004 are interconnected via a bus 1005. An I / O interface 1006 is also connected to the bus. Typically, the following systems can be connected to I / O interface 1006: input devices 1007 including, for example, touchscreens, touchpads, keyboards, mice, image sensors, microphones, accelerometers, gyroscopes, etc.; output devices 1008 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices 1003 including, for example, magnetic tapes, hard disks, etc.; and communication devices 1009. Communication device 1009 allows the multi-source data fusion-based dynamic assessment device for the risk of occult bleeding after renal biopsy to exchange data wirelessly or via wired communication with other devices. Although the figure shows a multi-source data fusion-based dynamic assessment device for the risk of occult bleeding after renal biopsy with various systems, it should be understood that implementing or having all the systems shown is not required. More or fewer systems can be implemented alternatively.

[0081] Example 4: This invention also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the above-described method for dynamic assessment of the risk of occult bleeding after renal biopsy based on multi-source data fusion. The computer program product provided by this invention can solve the technical problem of dynamic assessment of the risk of occult bleeding after renal biopsy based on multi-source data fusion. Compared with the prior art, the beneficial effects of the computer program product provided by this invention are the same as those of the method for dynamic assessment of the risk of occult bleeding after renal biopsy based on multi-source data fusion provided in the above embodiments, and will not be repeated here.

[0082] In particular, according to the embodiments disclosed in this invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of this invention include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device, or installed from storage device 1003, or installed from read-only memory 1002. When the computer program is executed by processing device 1001, it performs the functions defined in the methods of the embodiments disclosed in this invention.

[0083] It should be understood that the various parts disclosed in this invention can be implemented using hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics can be combined in any suitable manner in one or more embodiments or examples.

[0084] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for dynamic assessment of the risk of occult bleeding after renal biopsy based on multi-source data fusion, characterized in that, The methods include: Step S10: Obtain postoperative multi-source observation data, and perform multi-source feature construction task based on postoperative multi-source observation data using a unified time axis alignment and baseline difference fusion mechanism, and output a standardized unified observation vector sequence; Step S20: Based on the standardized unified observation vector sequence, a short-time exponential decay weighted fusion mechanism is used to perform a short-time instantaneous risk modeling task and output the short-time endpoint risk status; Step S30: Based on the short-term endpoint risk state, a time-varying state transition probability construction mechanism is used to perform the long-term lag risk transition modeling task, and output the long-term risk state distribution results; Step S40: Based on the long-term risk status distribution results, a risk transfer acceleration judgment mechanism is used to execute the hierarchical disposal decision-making construction task and output the review and scheduling plan; Step S50: Based on the review scheduling plan, use the consistency calibration feedback mechanism to perform the risk probability correction task and output the final occult bleeding risk level.

2. The method for dynamic assessment of the risk of occult bleeding after renal biopsy based on multi-source data fusion as described in claim 1, characterized in that, Step S10 involves acquiring postoperative multi-source observation data, performing multi-source feature construction based on the postoperative multi-source observation data using a unified time axis alignment and baseline difference fusion mechanism, and outputting a standardized unified observation vector sequence. Specifically, this includes: Step S101: Acquire postoperative multi-source observation data, including vital sign data collected within 0 to 6 hours postoperatively with a sampling step of 5 minutes, vital sign data collected within 6 to 72 hours postoperatively with a sampling step of 30 minutes, laboratory hemoglobin test data, bedside ultrasound imaging feature data, and puncture baseline parameters; among them, vital sign data include systolic blood pressure (SBP), diastolic blood pressure (DBP), heart rate (HR), and mean arterial pressure (MAP); laboratory hemoglobin test data includes hemoglobin concentration (Hb); bedside ultrasound imaging feature data includes the maximum diameter of the hematoma; and puncture baseline parameters include the number of punctures, needle diameter, and puncture depth. Step S102: First, perform time axis alignment processing on the postoperative multi-source observation data, output the aligned postoperative multi-source observation data, and then extract the hemoglobin change based on the laboratory hemoglobin detection data. Based on the change in hemoglobin Hemoglobin production trend characteristic variables; Step S103: Obtain laboratory testing delay information, encode the laboratory testing delay information as missing label variables, and perform feature fusion and splicing of missing label variables, hemoglobin trend feature variables and post-alignment multi-source observation data to output a standardized unified observation vector sequence.

3. The method for dynamic assessment of the risk of occult bleeding after renal biopsy based on multi-source data fusion as described in claim 1, characterized in that, Step S20, which involves performing short-term real-time risk modeling based on a standardized unified observation vector sequence using a short-term exponential decay weighted fusion mechanism and outputting the short-term endpoint risk state, specifically includes: Step S201: Preset a short window and exponentially decaying weights Based on exponential decay weights For short window The standardized unified observation vector sequence within the sequence undergoes exponential decay weighted fusion to generate a weighted short-time feature vector. ; Step S202: Introduce the logistic probability mapping function, and apply it to the weighted short-time feature vector. Perform risk probability calculations and output short-term real-time risk probabilities; Step S203: Based on the preset multi-threshold gating judgment rules, the short-term immediate risk probability is discretized into four levels of short-term risk status; all four levels of short-term risk status within the short-term window are merged, and the short-term endpoint risk status is finally output.

4. The method for dynamic assessment of the risk of occult bleeding after renal biopsy based on multi-source data fusion as described in claim 3, characterized in that, Step S203, which involves discretizing the short-term immediate risk probability into four levels of short-term risk states based on a preset multi-threshold gating judgment rule, specifically includes: Step S2031: Preset a multi-threshold set, the multi-threshold set including a first risk threshold. Second risk threshold and the third risk threshold Among them, the first risk threshold Used to distinguish between stable and low-risk states; second risk threshold Used to distinguish between low-risk and medium-risk states; third risk threshold Used to distinguish between medium-risk and high-risk states; and meets the following conditions: ; Step S2032: Based on a multi-threshold set, and combined with a nonlinear mapping model with time decay weights, perform gated discretization processing on the short-term instantaneous risk probability, and output the discrete risk probability p; when At that time, the determination of the short-term immediate risk probability corresponds to the first short-term risk state. ; when At that time, the determination of the short-term immediate risk probability corresponds to the second short-term risk state. ; when At that time, the determination of the short-term immediate risk probability corresponds to the third short-term risk state. ; when At that time, the determination of the short-term immediate risk probability corresponds to the fourth short-term risk state. ; Step S2033: Output a Level 4 short-term risk status based on the above judgment results.

5. The method for dynamic assessment of the risk of occult bleeding after renal biopsy based on multi-source data fusion as described in claim 1, characterized in that, Step S30, which involves performing a long-term lag risk transition modeling task based on a time-varying state transition probability construction mechanism using the short-term endpoint risk state, and outputting the long-term risk state distribution results, specifically includes: Step S301: Encode the short-term endpoint risk state into the long-term initial state and generate the initial state distribution; Step S302: Obtain the standardized unified observation vector sequence of the past 6 hours, and construct a long-term weighted feature vector based on the standardized unified observation vector sequence of the past 6 hours. Construct a time-varying state transition probability matrix based on the long-term weighted feature vector using the exponential mapping Markov chain transition rate modeling method. Step S303: Based on the initial state distribution and the time-varying state transition probability matrix, the future state distribution is calculated recursively using matrix multiplication, and the long-term risk state distribution result is output.

6. The method for dynamic assessment of the risk of occult bleeding after renal biopsy based on multi-source data fusion as described in claim 1, characterized in that, In step S40, the steps of constructing a graded disposal decision based on the risk transfer acceleration judgment mechanism according to the long-term risk state distribution results, and outputting the review and scheduling plan, specifically include: Step S401 extracts the short-term instantaneous risk probability difference between adjacent time points in the long-term risk state distribution results as a risk transfer acceleration index; Step S402: Based on the threshold range of the risk transfer acceleration index, perform a classification judgment and generate the corresponding disposal level; Step S403: Generate a treatment decision vector and a review scheduling plan according to the corresponding treatment level. The review scheduling plan includes at least a vital signs monitoring frequency adjustment plan, a hemoglobin retesting time plan, and an image review plan.

7. The method for dynamic assessment of the risk of occult bleeding after renal biopsy based on multi-source data fusion as described in claim 1, characterized in that, Step S50, which involves executing a risk probability correction task based on the consistency calibration feedback mechanism according to the review scheduling plan and outputting the final occult bleeding risk level, specifically includes: Step S501: After executing the review scheduling plan, obtain the review hemoglobin change value, review image hematoma change information, and review hematoma maximum diameter mapping amount; Step S502: Based on the changes in hemoglobin, changes in hematoma in the re-examination images, and the maximum diameter mapping of the re-examination hematoma, a re-examination calibration factor is constructed using the proportional gain calibration method. Based on the re-examination calibration factor, the long-term risk status distribution results are adjusted by gain, and the final occult bleeding risk level is output.

8. A dynamic assessment system for the risk of occult bleeding after renal biopsy based on multi-source data fusion, applied to the dynamic assessment method for the risk of occult bleeding after renal biopsy based on multi-source data fusion as described in any one of claims 1 to 7, characterized in that, The dynamic assessment system for the risk of occult bleeding after renal biopsy based on multi-source data fusion includes: The multi-source data structuring module is used to acquire postoperative multi-source observation data. Based on the postoperative multi-source observation data, a unified time axis alignment and baseline difference fusion mechanism is used to perform multi-source feature construction tasks and output a standardized unified observation vector sequence. The short-term risk modeling module is used to perform short-term real-time risk modeling tasks based on a standardized unified observation vector sequence and a short-term exponential decay weighted fusion mechanism, and outputs the short-term endpoint risk status. The long-term risk transfer modeling module is used to perform long-term lag risk transfer modeling tasks based on short-term endpoint risk states using a time-varying state transition probability construction mechanism, and outputs long-term risk state distribution results. The risk acceleration decision module is used to construct a graded disposal decision based on the risk transfer acceleration judgment mechanism according to the long-term risk status distribution results, and outputs a review and scheduling plan. The risk consistency calibration module is used to perform risk probability correction tasks according to the review scheduling plan using a consistency calibration feedback mechanism, and output the final occult bleeding risk level.

9. A dynamic assessment device for the risk of occult bleeding after renal biopsy based on multi-source data fusion, characterized in that, The device for dynamic assessment of the risk of occult bleeding after renal biopsy based on multi-source data fusion includes: a memory, a processor, and a program for dynamic assessment of the risk of occult bleeding after renal biopsy based on multi-source data fusion stored in the memory and executable on the processor. When the program for dynamic assessment of the risk of occult bleeding after renal biopsy based on multi-source data fusion is executed by the processor, it implements the method for dynamic assessment of the risk of occult bleeding after renal biopsy based on multi-source data fusion as described in any one of claims 1 to 7.

10. A computer program product, characterized in that, The computer program product includes a dynamic assessment program for the risk of occult bleeding after renal biopsy based on multi-source data fusion. When the processor executes the dynamic assessment program for the risk of occult bleeding after renal biopsy based on multi-source data fusion, it implements the method for dynamic assessment of the risk of occult bleeding after renal biopsy based on multi-source data fusion as described in any one of claims 1 to 7.

Images