Kidney injury prediction system based on big data

Abstract

The invention relates to the technical field of medical data analysis, in particular to a kidney injury prediction system based on big data, which comprises a physiological data processing module, a fluctuation analysis construction module, a dynamic weight evaluation module, a cross-domain data scheduling module and a risk trend prediction module. According to the method, through timestamp standardization and continuous recombination of the acute kidney injury physiological data, the consistency of multi-source monitoring data in the time dimension is ensured, drifting and dislocation are eliminated, the time sequence integrity is improved, fluctuation points of serum creatinine and urine volume are extracted, sensitive time slices are recognized, and high-risk time point early warning is achieved; a rate change comparison mechanism is introduced, a physiological factor weight sequence is generated, factor influence evaluation is enhanced, uploading frequency and a time delay state are combined, high-reliability nodes are screened, data continuity and real-time performance are improved, index trend deviation is tracked, a mode label is generated, and timeliness and accuracy of renal function change prediction are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of medical data analysis, and in particular to a kidney injury prediction system based on big data. Background Art

[0002] The field of medical data analysis technology encompasses the entire process of collecting, organizing, analyzing, and assisting in decision-making regarding patient health information. The core of this technology lies in utilizing multi-source medical data, such as clinically collected data, physiological monitoring indicators, laboratory reports, and medical records, to conduct in-depth exploration of risk factor responses in the disease development process through a specific analytical framework, thereby providing support for disease early warning, early screening, and optimization of diagnostic pathways. This field includes multiple links, including data acquisition and standardization, data cleaning and integration, feature extraction and modeling, as well as result interpretation and clinical feedback. It involves the multidisciplinary integration of statistical medical knowledge and computational modeling, and has significant data-driven and clinical-oriented characteristics.

[0003] Among them, the kidney injury prediction system refers to a system that identifies the risk of renal function deterioration and provides early intervention prompts by analyzing the patient's past medical history, laboratory indicators, medication records and physiological monitoring data. The technical matters targeted by this system include the identification of the pre-onset state of acute kidney injury, the risk quantification of high-risk groups, and the analysis of dynamic changes in renal function indicators. It completes the selection of predictive features through clinical feature screening based on established evaluation rules, and combines laboratory indicator threshold comparison and multi-time node indicator fluctuation analysis to achieve forward-looking judgment of potential renal function damage. The analysis process is based on standardized structured data, supplemented by cross-reference of retrospective medical record data to enhance the accuracy and pertinence of risk identification.

[0004] Existing technologies often use static assessment methods to extract indicators during data processing, lacking the ability to synchronize and reorganize physiological data in a continuous time series dimension. This leads to time misalignment between the collection of multiple indicators from the same patient, compromising the integrity and consistency of data analysis. In terms of fluctuation identification, reliance on absolute thresholds makes it difficult to reveal the dynamic frequency and mutation patterns of indicators within a local time period, which can easily lead to the omission of important time periods. In the weight allocation process, clinical experience or preset weights are often used as the main approach, ignoring the objective reflection of the indicator's own rate of change and degree of effect, and failing to dynamically reflect the forces of different physiological factors under different states. In terms of data scheduling, upload nodes are not selected based on multidimensional judgment criteria, and insufficient consideration is given to latency and frequency, resulting in large differences in data source stability and insufficient real-time performance. At the trend assessment level, existing technologies often use a single time point or short-term window as a reference, lacking the ability to mine the consistent deviation behavior of indicators over longer time periods. This limits the accurate prediction of the future direction of changes in renal function status, easily leading to delayed warnings or misjudgment of risk trends, making it difficult to meet the dynamic tracking and intervention needs of high-risk patients in clinical practice. Summary of the Invention

[0005] The purpose of the present invention is to solve the shortcomings of the existing technology and propose a kidney damage prediction system based on big data.

[0006] To achieve the above objectives, the present invention adopts the following technical solutions: A kidney injury prediction system based on big data includes:

[0007] The physiological data processing module is based on acute kidney injury monitoring data, including blood pressure, urine volume, and serum creatinine. It standardizes and sorts the collected data according to timestamps, synchronizes the data, and reorganizes the time-series monitoring data stream to generate a continuous synchronous monitoring sequence;

[0008] The fluctuation analysis construction module extracts the time period change points of serum creatinine and urine volume based on the continuous synchronous monitoring sequence, determines the frequency of change of each indicator in the continuous time period, selects the abnormal fluctuation value interval as the sensitive time segment, and obtains the renal function fluctuation time window;

[0009] The dynamic weight assessment module selects the continuous value change rate of blood pressure, urine volume, and creatinine within the time period according to the renal function fluctuation time window, compares each rate according to the proportional relationship and assigns a value to sort, assigns a dynamic weight level to each item according to the sort, and generates a dynamic weight table of physiological factors;

[0010] The cross-domain data scheduling module calls the data types with the highest weight ratios in the dynamic weight table of physiological factors, cross-judges the upload frequency and node delay status of medical nodes in differentiated areas, selects nodes with high upload frequency and low delay and prioritizes them into the warehouse, and obtains a priority set of stable nodes.

[0011] As a further solution of the present invention, the continuous synchronous monitoring sequence includes a unified time axis, a standardized indicator sequence, and data synchronization accuracy parameters; the renal function fluctuation time window includes the change frequency distribution, the abnormal value dense interval, and the indicator fluctuation amplitude range; the physiological factor dynamic weight table includes the indicator weight ranking, the rate comparison ratio, and the dynamic assignment level; the stable node priority set includes the node response delay, the data upload frequency, and the node selection priority.

[0012] As a further solution of the present invention, the physiological data processing module includes:

[0013] The time series standardization submodule is based on acute kidney injury monitoring data, including original monitoring records and timestamps of blood pressure, urine volume, and serum creatinine. It sorts each monitoring record according to the timestamp, removes duplicate timestamps and time-missing records, determines the consistency of adjacent time intervals, and generates a standard time series set.

[0014] The synchronization screening submodule sets a synchronization screening interval threshold based on the time node of each monitoring indicator in the standard time series set, determines the valid records of blood pressure, urine volume, and serum creatinine at each time node, eliminates nodes that are missing any indicator, and generates a synchronization observation node set;

[0015] The indicator flow construction submodule calls the indicators under the time nodes in the synchronous observation node set, combines the indicators into indicator vectors in chronological order, and concatenates each vector in sequence to generate a continuous synchronous monitoring sequence.

[0016] As a further solution of the present invention, the fluctuation analysis building block includes:

[0017] The period change identification submodule extracts adjacent measurement values of serum creatinine and urine volume based on the continuous synchronous monitoring sequence, calculates the absolute difference and compares it with the change threshold, selects the measurement nodes that meet the mutation condition, and generates a period change point group of serum creatinine and urine volume;

[0018] The change frequency judgment submodule calls the time period change point group of serum creatinine and urine volume, counts the number of changes of each indicator within the time period, extracts the time periods of synchronous change, and generates a sequence of synchronous change frequency values of the indicators;

[0019] The change sensitivity intensity identification submodule calls the indicator synchronous change frequency value sequence, identifies the change amplitude difference between adjacent time nodes, and normalizes the fluctuation amount using the formula:

[0020]

[0021] Calculate the change sensitivity intensity value, identify the time interval exceeding the abnormal threshold, and obtain the renal function fluctuation time window;

[0022] Among them, F i represents the frequency of change in the i-th period, is the average frequency of change, V i is the combined volatility difference of the i-th period, D i is the change difference between the i-th period and the previous period, S i is the change sensitivity intensity value of the i-th period, and n represents the total number of periods.

[0023] As a further solution of the present invention, the dynamic weight evaluation module includes:

[0024] The continuous rate calculation submodule extracts the continuous values of blood pressure, urine volume, and creatinine within the time period according to the renal function fluctuation time window, identifies the change amount of each item per unit time, and obtains a continuous change rate group;

[0025] The rate ratio assignment submodule identifies the ratio difference between urine volume and creatinine rate according to the continuously changing rate group and the blood pressure rate, sorts them according to the ratio, assigns corresponding serial numbers, and generates a rate sorting sequence;

[0026] The dynamic weight generation submodule analyzes the sequence relationship based on the rate sorting sequence and identifies the dynamic performance weight of each physiological indicator using the formula:

[0027]

[0028] Calculate the dynamic weight values corresponding to the indicators, match the weight values with the original indicators, and generate a dynamic weight table of physiological factors;

[0029] Among them, W j Represents the dynamic weight value corresponding to the j-th indicator, ΔR j is the change in the rate of the jth index, Q is the blood pressure rate value, E j represents the reference rate corresponding to the j-th indicator rate, R j represents the rate of the j-th indicator, and N represents the total number of indicators.

[0030] As a further solution of the present invention, the cross-domain data scheduling module includes:

[0031] The dynamic weight extraction submodule calls the physiological factor dynamic weight table, extracts the weight proportion of the data type, sorts them, identifies the upload frequency and node delay status of the top data types, and generates a weighted top data type set;

[0032] The frequency-delay cross judgment submodule calls the upload frequency and node delay status of the weighted top data type set, and performs cross analysis based on the upload frequency, node delay and original data volume, combined with the number of concurrent nodes in the region and the average delay, using the formula:

[0033]

[0034] Calculate the node screening evaluation value, select the top-ranked nodes, and obtain the frequency delay optimal node set;

[0035] Among them, K represents the node screening evaluation value, A x Represents the upload frequency value of the x-th node, B x represents the delay value of the xth node, C x represents the cumulative upload amount of original data of the xth node, D x Represents the number of concurrent nodes in the area where the x-th node is located, represents the average delay value of the node, and M represents the total number of nodes;

[0036] The stable node screening submodule selects priority nodes according to the evaluation value ranking in the frequency delay preferred node set, determines whether the upload frequency and node status are continuous and stable, and screens out nodes that do not meet the conditions to obtain a stable node priority set.

[0037] As a further solution of the present invention, the system also includes a risk trend prediction module:

[0038] The risk trend prediction module, based on the serum creatinine, urine protein, and blood pressure data uploaded in the stable node priority set, calculates the directional deviation of each indicator within a continuous interval, determines whether the deviation pattern remains consistent over multiple time periods, and if so, performs trend prediction to obtain a kidney injury trend prediction pattern label;

[0039] The renal injury trend prediction model label includes indicator deviation directionality, trend maintenance period, and risk prediction type.

[0040] As a further solution of the present invention, the risk trend prediction module includes:

[0041] The data offset identification submodule divides the time interval into chronological order based on the serum creatinine value, urine protein value and blood pressure value uploaded in the stable node priority set, extracts the continuous records of the indicators within the interval, identifies the change direction between adjacent records, and marks them as positive, negative and no change, and generates a directional offset sequence value;

[0042] The offset pattern consistency submodule calls the directional offset sequence value, compares the directional offset state of each indicator in the differentiated time interval, selects the interval sequence with a continuous and consistent offset direction, calculates the proportion of the consistent directional state, and obtains a continuous offset consistency ratio value;

[0043] The trend dynamic prediction submodule identifies the direction offset sequence of each indicator based on the continuous offset consistency ratio value, sets the trend judgment benchmark according to the offset continuation direction and the number of time intervals, screens the indicator paths that meet the trend conditions, and obtains the kidney injury trend prediction pattern label.

[0044] Compared with the prior art, the advantages and positive effects of the present invention are:

[0045] In the present invention, by standardizing and sorting the physiological data related to acute kidney injury by timestamp and reorganizing it into a continuous time series structure, the consistency of multi-source monitoring data in the time dimension is ensured, and the time drift and dislocation problems in the original data records are effectively eliminated. Based on the reorganized time series data, the changing rhythms and fluctuation points of serum creatinine and urine volume in each monitoring segment are deeply explored, and the sensitive time segments in the area of ​​violent fluctuation of the indicators are accurately extracted to achieve early identification of potential high-risk time points in renal function changes. A dynamic comparison mechanism of the indicator change rate is further introduced to convert the rate change intensity of different physiological factors into weight levels, thereby establishing a ranking basis for factors affecting the dynamic state of renal function. In the data scheduling strategy, through the precise judgment of the weight proportion of each type of indicator, combined with the upload frequency and response delay status of the medical node, high-reliability nodes are screened out, so that the summary of monitoring data has higher continuity and real-time performance. On the basis of a stable data source, the trend deviation status of key indicators in the continuous interval is tracked and the consistency judgment is formed to form a forward-looking trend pattern label, so that the early warning capability of changes in the state of renal injury has higher timeliness. The overall processing logic builds a systematic enhancement path through data sorting and reorganization, time period fluctuation identification, rate weight allocation, data source optimization, trend tracking and other links to improve time series integrity, risk identification sensitivity, factor effect analysis accuracy, data flow stability and trend prediction foresight, providing more targeted and stable decision-making support for renal function monitoring. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 is a system flow chart of the present invention;

[0047] Figure 2 This is a flow chart of the physiological data processing module in the present invention;

[0048] Figure 3 This is a flow chart of the fluctuation analysis module in the present invention;

[0049] Figure 4 This is a flow chart of the dynamic weight evaluation module in the present invention;

[0050] Figure 5 This is a flow chart of the cross-domain data scheduling module in the present invention;

[0051] Figure 6This is a flow chart of the risk trend prediction module in the present invention. DETAILED DESCRIPTION

[0052] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0053] In the description of the present invention, it should be understood that the terms "length," "width," "up," "down," "front," "back," "left," "right," "vertical," "horizontal," "top," "bottom," "inside," "outside," and the like, indicating positions or relationships, are based on the positions or relationships shown in the accompanying drawings and are intended only to facilitate the description of the present invention and simplify the description. They do not indicate or imply that the devices or elements referred to must have a specific orientation, be constructed, or operate in a specific orientation. Therefore, they should not be construed as limiting the present invention. Furthermore, in the description of the present invention, "plurality" means two or more, unless otherwise expressly and specifically defined.

[0054] See also Figure 1 , a kidney injury prediction system based on big data includes:

[0055] The physiological data processing module is based on acute kidney injury monitoring data, including blood pressure, urine volume, and serum creatinine. It standardizes and sorts the collected data according to timestamps, synchronizes the data, and reorganizes the time-series monitoring data stream to generate a continuous synchronous monitoring sequence;

[0056] The fluctuation analysis module is based on a continuous synchronous monitoring sequence, extracting the time-varying points of serum creatinine and urine volume, determining the frequency of changes in each indicator over a continuous period, and screening abnormal fluctuation intervals as sensitive time segments to obtain the renal function fluctuation time window.

[0057] The dynamic weight assessment module selects the continuous value change rate of blood pressure, urine volume, and creatinine within the time period according to the renal function fluctuation time window, compares each rate according to the proportional relationship, assigns a value and sorts it, and assigns a dynamic weight level to each item based on the sorting to generate a dynamic weight table of physiological factors;

[0058] The cross-domain data scheduling module calls the data types with the highest weights in the dynamic weight table of physiological factors, cross-judging the upload frequency and node latency status of medical nodes in differentiated regions, and prioritizes nodes with high upload frequency and low latency to obtain a prioritized set of stable nodes.

[0059] The risk trend prediction module is based on the serum creatinine, urine protein and blood pressure data uploaded in the stable node priority set. It counts the directional deviation of each indicator in a continuous interval and determines whether the deviation pattern remains consistent over multiple time periods. If it is consistent, trend prediction is performed to obtain the kidney injury trend prediction pattern label.

[0060] The continuous synchronous monitoring sequence includes a unified time axis, a standardized indicator sequence, and data synchronization accuracy parameters. The renal function fluctuation time window includes the change frequency distribution, the abnormal value concentration interval, and the indicator fluctuation amplitude range. The dynamic weight table of physiological factors includes the indicator weight ranking, the rate comparison ratio, and the dynamic assignment level. The stable node priority set includes the node response delay, the data upload frequency, and the node selection priority. The renal injury trend prediction model label includes the indicator offset directionality, the trend maintenance period, and the risk prediction type.

[0061] See also Figure 2 , the physiological data processing module includes:

[0062] The time series standardization submodule is based on acute kidney injury monitoring data, including original monitoring records and timestamps of blood pressure, urine volume, and serum creatinine. It sorts each monitoring record according to the timestamp, removes duplicate timestamps and time-missing records, determines the consistency of adjacent time intervals, and generates a standard time series set.

[0063] First, the original data of physiological parameters is obtained through regular monitoring using dedicated equipment in hospitals or laboratories. For example, in the intensive care unit, doctors and nurses use electronic blood pressure monitors, urine meters, and serum creatinine meters to collect data for continuous monitoring of critically ill patients. The data is then stored electronically in the hospital information. When doctors view the data, they can obtain the patient's immediate physiological status and make corresponding diagnosis and treatment decisions. The timestamp of the data is automatically generated by the monitoring equipment to ensure the exact moment of data collection, thereby standardizing the data. The data standardization process includes sorting all monitoring data by timestamp, eliminating data with abnormal or duplicate timestamps, and generating a standard time series set. The data set plays a key role in ensuring the integrity and accuracy of the time series and is the basis for subsequent data processing and analysis.

[0064] The synchronization screening submodule sets the synchronization screening interval threshold based on the time node of each monitoring indicator in the standard time series set, determines the valid records of blood pressure, urine volume, and serum creatinine at each time node, eliminates the nodes that are missing any indicator, and generates a set of synchronization observation nodes;

[0065] Based on the standard time series set, all monitored indicators such as blood pressure, urine volume and serum creatinine are recorded. The first step in the process is to set a maximum allowable time interval. This interval is determined based on specific clinical needs and monitoring frequency. For example, in the monitoring of acute kidney injury, complete data must be obtained at least once an hour to assess rapid changes in the patient's condition. In synchronous screening, the setting of the time interval threshold is based on statistical analysis. Generally, past data is used to determine an optimal threshold to ensure the synchronization and integrity of the data. If the data at a certain time point is incomplete, the data at that time point is considered invalid. After eliminating the data, what remains are time nodes with complete data records at each valid time point, generating a set of synchronous observation nodes.

[0066] The indicator flow construction submodule calls the indicators under the time nodes in the synchronous observation node set, combines the indicators into indicator vectors in chronological order, and concatenates each vector in turn to generate a continuous synchronous monitoring sequence;

[0067] Construct an indicator stream, combining different physiological indicator values (blood pressure, urine volume, serum creatinine) at the same time point into an indicator vector. Each vector represents the complete physiological state at a time point. By connecting the vectors in chronological order, a continuous time series is formed. This sequence is very useful in medical research and clinical diagnosis. It can be used to track the dynamic changes in the patient's condition. For doctors, this continuous data stream provides an intuitive way to observe the progression of the disease and generate a continuous synchronous monitoring sequence.

[0068] See also Figure 3 , the fluctuation analysis building blocks include:

[0069] The time period change identification submodule extracts adjacent measurement values of serum creatinine and urine volume based on the continuous synchronous monitoring sequence, calculates the absolute difference and compares it with the change threshold, selects the measurement nodes that meet the mutation conditions, and generates a time period change point group for serum creatinine and urine volume;

[0070] In the context of monitoring renal function, it is necessary to track a patient's serum creatinine and urine volume indicators to assess their renal health. Through automated data processing, doctors can quickly identify key change points and provide timely medical intervention. The raw values of serum creatinine and urine volume are extracted from the patient's continuous monitoring data. The difference between the data at each time point and the data at the previous time point is calculated. If the absolute value of the difference exceeds the set threshold, the point is considered a key change point. For example, if the serum creatinine measurement value suddenly increases by 20% within two consecutive hours, this point will be marked as a key change point. Key points are marked as a set of change points. This automated labeling process reduces the subjectivity of manual judgment and ensures the consistency and accuracy of data processing. Through this method, doctors can quickly obtain key medical information to make more accurate decisions and generate time period change point groups for serum creatinine and urine volume.

[0071] The change frequency judgment submodule calls the time period change point group of serum creatinine and urine volume, counts the number of changes of each indicator within the time period, extracts the time periods of synchronous change, and generates a sequence of synchronous change frequency values of indicators;

[0072] In the actual operation of monitoring kidney disease, it is necessary to understand whether there is any progress or improvement in the patient's condition. By calling the time period change point group of serum creatinine and urine volume, the frequency of occurrence of the change points during the monitoring period is counted, thus constructing a complete change frequency sequence. For example, if a patient's urine volume change point between 8 and 10 in the morning is higher than that of other time periods, this indicates that the patient's fluid metabolism is abnormal. After analyzing this information through an algorithm, the time period with significant change frequency on the timeline is extracted to identify the key periods of worsening or improvement in the condition. Through this analysis, doctors can obtain detailed information about the changes in the condition and make more targeted treatment adjustments, ultimately generating a sequence of synchronized indicator change frequency values.

[0073] The change sensitivity intensity identification submodule calls the indicator synchronization change frequency value sequence, identifies the difference in the change amplitude of adjacent time nodes, and normalizes the fluctuation amount using the formula:

[0074]

[0075] Calculate the change sensitivity intensity value, identify the time interval exceeding the abnormal threshold, and obtain the renal function fluctuation time window;

[0076] Among them, F i represents the frequency of change in the i-th period, is the average frequency of change, V i is the combined volatility difference of the i-th period, D i is the change difference between the i-th period and the previous period, S iis the change sensitivity intensity value of the i-th period, and n represents the total number of periods;

[0077] In the actual monitoring scenario, based on the previously generated indicator synchronous change frequency value sequence, the key interval representing the risk of renal function fluctuation can be further identified. First, the frequency sequence is called to extract the change frequency value F for each hour. i , calculate its average frequency over the entire period The absolute deviation value of the combined fluctuation intensity V is introduced i , which is the square root of the variance of the combined change in serum creatinine and urine volume per hour. Assuming that the total monitored time period is n = 5 hours, the values of each parameter are obtained as follows:

[0078] F i : The frequency of simultaneous changes in serum creatinine and urine volume within each hour, obtained from the first two modules. For example, from the 1st to the 5th hour, there were 2 times, 4 times, 1 time, 3 times, and 5 times respectively;

[0079] F i The arithmetic mean of the sequence, i.e. (2+4+1+3+5) / 5=3;

[0080] V i : The square root of the mean of the square difference between the hourly changes in serum creatinine and urine volume is taken. Assuming that in the third hour, the change in creatinine is 0.8 μmol / L and the change in urine volume is 120 ml, the combined variance is 0.06 (μmol / L 2 ), then the third hour The rest are: V1=0.316, V2=0.447, V4=0.283, V5=0.500;

[0081] D i : The difference in the synchronous change between the current period and the previous period, represented by |F i -F i-1 |Get, the first hour has no previous value set to 0, and the rest are 2, 3, 2, 2 in sequence;

[0082] In terms of dimensionality unification, the frequency unit is "times / hour", the combined fluctuation value is represented by "variation standard deviation / hour", and the difference is a scalar quantity. The three can be combined and multiplied and divided under dimensionless processing;

[0083] Substitute the above values into the formula to calculate S for each period i , the process is as follows:

[0084] Hour 1:

[0085] Hour 2:

[0086] Hour 3:

[0087] Hour 4:

[0088] Hour 5:

[0089] Final result: S i The sequence is: 0.562, 0.223, 0.248, 0.000, 0.471

[0090] The abnormal sensitivity threshold is set to 0.45 (set by analyzing the 90th percentile value in a large sample), and the 1st and 5th hours are the abnormal change intervals, so the renal function fluctuation time window is obtained as the 1st and 5th hours;

[0091] Among them, F i Indicates the frequency of synchronous changes in the i-th hour (times / hour), Indicates the average synchronous change frequency within the entire monitoring period, V i represents the square root of the variance of the combined change in creatinine and urine volume in hour i (dimensionless), D i Indicates the difference (scalar) between the frequency value of the i-th hour and the previous hour, S w It is the change sensitivity intensity value of each hour period, which is used as a numerical reference for calibrating the abnormal time window;

[0092] By simultaneously considering the frequency of changes, the combined fluctuation amplitude of indicators and the degree of mutation, a multi-level change sensitivity intensity calculation logic was constructed to achieve more discriminating sensitive time window identification. The results showed that the identified 1st hour and 5th hour were periods of significant changes in renal function, which can be used as key windows for subsequent intervention or reanalysis.

[0093] See also Figure 4 , the dynamic weight evaluation module includes:

[0094] The continuous rate calculation submodule extracts the continuous values of blood pressure, urine volume, and creatinine within the time period according to the renal function fluctuation time window, identifies the change amount of each item per unit time, and obtains the continuous change rate group;

[0095] Continuous values of blood pressure, urine volume, and creatinine are obtained within the renal function fluctuation time window. For example, in the intensive care unit, the stability of the patient's renal function is assessed by continuously recording the patient's physiological data. The change in each parameter is calculated per unit time to reflect the dynamic changes in the patient's physiological state. Assuming that the blood pressure drops from 120 mmHg to 100 mmHg within an hour, with a time interval of 60 minutes, the blood pressure change rate is -0.33 mmHg / minute. This calculation can not only observe the patient's blood pressure change trend in the short term, but also compare it with physiological parameters such as urine volume and creatinine for early diagnosis of acute kidney injury or related diseases, and obtain a continuous change rate group.

[0096] The rate ratio assignment submodule identifies the ratio difference between urine volume and creatinine rate according to the continuously changing rate group and the blood pressure rate, sorts them according to the ratio, assigns corresponding serial numbers, and generates a rate sorting sequence;

[0097] By comparing and analyzing the three indicators of blood pressure, urine volume, and creatinine, for example, during dialysis, by comparing the changes in these three indicators, the patient's physiological response before and after dialysis can be evaluated. The blood pressure rate is set as the reference item, and the rates and ratios of the other two indicators are calculated. By setting logical judgment, for example, the blood pressure rate is -0.33 mmHg / minute, the urine volume rate is 50 mL / hour, which is converted to 0.83 mL / minute, and the creatinine rate is 0.05 mg / dL, which is converted to 0.00083 mg / dL / minute. The ratios are sorted and assigned corresponding serial numbers. Based on the blood pressure rate, the ratios of urine volume and creatinine are 2.52 and 0.0025, respectively, so they are sorted from high to low to obtain a rate sorting sequence.

[0098] The dynamic weight generation submodule analyzes the sequence relationship based on the rate sorting sequence and identifies the dynamic performance weight of each physiological indicator using the formula:

[0099]

[0100] Calculate the dynamic weight values corresponding to the indicators, match the weight values with the original indicators, and generate a dynamic weight table of physiological factors;

[0101] Among them, W j Represents the dynamic weight value corresponding to the j-th indicator, ΔR j is the change in the rate of the jth index, Q is the blood pressure rate value, E j represents the reference rate corresponding to the j-th indicator rate, R j represents the rate of the jth indicator, and N represents the total number of indicators;

[0102] Dynamic weight calculation was performed on three physiological indicators: blood pressure, urine volume, and creatinine. First, the continuous change value of each indicator within the time window was obtained. The sampling time was set to 60 minutes, and the value changes at the initial and end time points were compared.

[0103] The blood pressure dropped from 120 mmHg to 100 mmHg, with a change of -20 mmHg, and the corresponding rate was ΔR1 = -20 / 60 = -0.333 mmHg / min;

[0104] The urine volume increases from 3500 mL to 3800 mL, a change of 300 mL, which is converted to a rate per unit time of ΔR2 = 300 / 60 = 5 mL / min;

[0105] Creatinine increases from 1.0 mg / dL to 1.05 mg / dL, a change of 0.05 mg / dL, corresponding to a rate of ΔR3 = 0.05 / 60 = 0.00083 mg / dL / min;

[0106] Before calculation, different units need to be converted into dimensionless ratios. The reference standard dimensions are set as follows: blood pressure 100 mmHg, urine volume 1000 mL, and creatinine 1 mg / dL.

[0107] Then the rate is dimensionless: R1 = -0.0033, R2 = 0.005, R3 = 0.00083;

[0108] Select the blood pressure rate as the reference rate, denoted as Q = -0.0033, and substitute the above parameters into the formula;

[0109] Calculate each term:

[0110] Blood pressure item: |ΔR1|=0.0033, |ΔR1-Q|=|-0.0033+0.0033|=0,

[0111]

[0112] The reference value is set to the rate measured the previous day: -0.0025, then |R1-E1|=|-0.0033+0.0025|=0.0008;

[0113] Urine volume: |ΔR2|=0.005, |ΔR2-Q|=|0.005+0.0033|=0.0083,

[0114]

[0115] The reference value is 0.0040, then |R2-E2|=|0.005-0.0040|=0.001;

[0116] Creatinine item: |ΔR3|=0.00083, |ΔR3-Q|=|0.00083+0.0033|=0.00413,

[0117]

[0118] The reference value is 0.0010, then |R3-E3|=|0.00083-0.0010|=0.00017;

[0119] Sum of all items:

[0120] Blood pressure weight:

[0121] Urine volume weight:

[0122] Creatinine Weight:

[0123] Through the aggregation calculation of absolute difference, square root operation and rate difference, the dynamic change trends of different indicators can be refined and distinguished, which is particularly suitable for evaluating the relative importance of physiological factors under rapid fluctuations. The final result is to establish a dynamic weight table of physiological factors. The results show that within this window period, the dynamic performance of urine volume is the strongest and has a greater impact on the evaluation indicators.

[0124] See also Figure 5 , the cross-domain data scheduling module includes:

[0125] The dynamic weight extraction submodule calls the physiological factor dynamic weight table, extracts the weight proportion of the data type, sorts it, identifies the upload frequency and node latency status of the top data types, and generates a weighted top data type set;

[0126] Extract the weight proportion of each data type, call the upload frequency and node latency status of the weighted top data types. To more carefully demonstrate the implementation of the dynamic weight extraction submodule, we can consider an actual medical data analysis scenario. For example, when conducting kidney injury risk assessment, medical data includes physiological indicators such as patients' blood pressure and heart rate. The upload frequency and latency of data at different time points can significantly affect the real-time performance and accuracy of the model. The uploaded data of each physiological indicator is weighted. For example, heart rate data is given a higher weight due to its rapid change rate and high importance for early injury warning. Then the data is sorted. For example, the weight of heart rate and blood pressure data is higher than that of body temperature. Based on the top-ranked data types, their upload frequency and latency are further analyzed to determine which nodes can provide stable data with higher frequency and lower latency, and generate a set of weighted top data types. Each step is obtained through detailed data monitoring and calculation. For example, its performance is evaluated by calculating the number of uploads and average latency of each data type.

[0127] The frequency-delay cross judgment submodule calls the upload frequency and node delay status of the weighted top data type set, and performs cross analysis based on the upload frequency, node delay, and original data volume, combined with the number of concurrent nodes in the region and the average delay, using the formula:

[0128]

[0129] Calculate the node screening evaluation value, select the top-ranked nodes, and obtain the frequency delay optimal node set;

[0130] Among them, K represents the node screening evaluation value, A x Represents the upload frequency value of the x-th node, B x represents the delay value of the xth node, C x represents the cumulative upload amount of original data of the xth node, D x Represents the number of concurrent nodes in the area where the x-th node is located, represents the average delay value of the node, and M represents the total number of nodes;

[0131] First, the parameters are standardized and dimensionally unified. The upload frequency is in "times / second", the node delay is in "milliseconds", the upload volume is in "GB", the number of concurrent nodes is a dimensionless integer, and the average delay is also in "milliseconds". In order to unify the units, the delay is converted to a unified unit of "seconds" before entering the formula calculation, that is, 1 millisecond is equal to 0.001 seconds, and the original upload volume is kept in "GB" as a stock reference item. The actual application scenario is set as follows. Assume that the upload frequency A of node A in the current medical data acquisition platform is x is 12 times / second, node B is 8 times / second, and node A has a delay of B x is 300 milliseconds (0.3 seconds), the original upload amount C x The node is located in an area with a concurrent node number D. x is 10, the average node delay 320 milliseconds (0.32 seconds);

[0132] The product of the upload frequency of all participating nodes and the original upload volume is:

[0133]

[0134] Substituting into the formula:

[0135] The results show that the comprehensive evaluation value of the node is 0.0224, and its dimension is a dimensionless relative indicator, which is used as the basis for node screening and sorting. The larger the value, the higher the node upload frequency and original activity and the more reasonable the delay. According to the evaluation value, the nodes can be compared and sorted to screen out the frequency-delay preferred node set. By mixing the product and square root of the upload frequency, delay, and original upload volume, and combining the weighted denominator with the number of regional nodes and the average delay to offset the weighted denominator, a nonlinear ratio expression of multiple factors is formed, thereby enhancing the sensitivity to the fluctuation of upload quality and achieving more reasonable data selection between nodes.

[0136] The stable node screening submodule selects priority nodes based on the evaluation value ranking in the frequency-delay preferred node set, determines whether the upload frequency and node status are continuous and stable, and screens out nodes that do not meet the conditions to obtain a priority set of stable nodes;

[0137] When screening for node stability, a specific application example is that when monitoring the upload of vital signs data of patients undergoing kidney surgery, priority is given to devices with high upload frequency and low latency to ensure real-time acquisition of key physiological parameters. Through real-time data stream analysis, the patient's heart rate, blood pressure and other indicators can be monitored in real time. If the data upload is found to be unstable or there is a high latency, the node is automatically excluded and a more stable node is selected to continue monitoring to ensure the accuracy and real-time nature of the data, thereby improving the response speed and accuracy of the patient's health status in actual medical situations and obtaining a priority set of stable nodes.

[0138] See also Figure 6 , the risk trend prediction module includes:

[0139] The data offset identification submodule divides the time interval into chronological order based on the serum creatinine values, urine protein values, and blood pressure values uploaded in the stable node priority set, extracts the continuous records of the indicators within the interval, identifies the change direction between adjacent records, and marks them as positive, negative, and no change, thereby generating a directional offset sequence value;

[0140] Extract relevant physiological indicator data from the prioritized set of stable nodes, specifically including serum creatinine, urine protein, and blood pressure values. Data extraction is based on continuous recordings from medical monitoring devices. For example, a typical application scenario is the long-term monitoring of chronic kidney disease patients, where portable monitoring devices regularly upload physiological indicator data to a cloud database. After data extraction, the data is sorted chronologically and then divided into continuous time intervals, such as weekly or monthly monitoring intervals. For each time interval, the change trend of each indicator within the interval is further calculated. The specific calculation method is to compare the data values at the start and end of the interval and determine the increase or decrease trend based on the set threshold. For example, if two consecutive blood pressure readings show an upward trend that exceeds 5 mmHg, it is judged as a positive change. This judgment is based not only on the increase or decrease in numerical value but also on the medical clinical significance. The core of this process is to mark the direction of change as positive, negative, or no change, and construct a directional offset sequence based on the marking results.

[0141] Table 1: Sample time interval data of serum creatinine, urine protein and blood pressure

[0142]

[0143] As shown in Table 1, serum creatinine, urine protein, and blood pressure showed a gradual upward trend over four consecutive months. By determining the direction of data changes at adjacent time points, all indicators can be marked as continuously shifting positively. Through detailed analysis and calculation of this phase, key shift trend information can be obtained. This information is used to predict the patient's health trends and ultimately generate a directional shift sequence value.

[0144] The offset pattern consistency submodule calls the directional offset sequence value, compares the directional offset state of each indicator within the differentiated time interval, selects the interval sequence with a continuous and consistent offset direction, calculates the proportion of consistent directional states, and obtains the continuous offset consistency ratio value;

[0145] Call the sequence value and perform further data processing on it. The specific operations include comparing the directional offset status of each continuous time interval to determine whether it remains consistent in multiple time intervals. For example, if a patient's blood pressure continues to rise for three consecutive months, it is recorded as a continuous positive offset. During this comparison process, the number and proportion of times each indicator maintains the same change direction in the continuous interval are calculated. For example, the interval in which blood pressure continues to rise accounts for 80% of the total monitoring interval. The calculation involves not only simple statistics, but also the setting of reasonable thresholds to judge the significance of the changes. The setting of this threshold is based on clinical needs. How to set the threshold will be adjusted according to different medical parameters and patient conditions. This judgment of continuous offset provides an important basis for subsequent predictions. Through refined processing, the continuous offset consistency ratio value is obtained to provide data support for the final trend prediction.

[0146] The trend dynamic prediction submodule identifies the directional offset sequence of each indicator based on the continuous offset consistency ratio value, sets the trend judgment benchmark based on the offset continuation direction and the number of time intervals, screens the indicator paths that meet the trend conditions, and obtains the kidney injury trend prediction pattern label;

[0147] The continuous deviation consistency ratio value is called, and then the trend persistence and time span of each indicator are analyzed according to the value, and the corresponding judgment benchmark is set. The benchmark value is a threshold with statistical significance obtained based on the analysis of the original data. For example, if the deviation consistency ratio of an indicator exceeds 70% in four consecutive time intervals, its future trend is predicted to be continuous growth or decline. This prediction is not only based on statistical data, but also needs to take into account the biomedical significance of the indicator. For example, if blood pressure is continuously higher than a certain threshold, it indicates that the treatment plan needs to be adjusted. Through comprehensive analysis, a kidney injury trend prediction is constructed for each patient. The prediction model comprehensively considers multiple factors to ensure the accuracy and reliability of the prediction, and finally generates a kidney injury trend prediction pattern label, which provides doctors with important decision-making basis and helps to better understand the patient's disease development trend.

[0148] The above are merely preferred embodiments of the present invention and do not limit the present invention in any other form. Any technician familiar with the profession may use the technical content disclosed above to change or modify it into an equivalent embodiment with equivalent changes and apply it to other fields. However, any simple modification, equivalent change and modification made to the above embodiment based on the technical essence of the present invention without departing from the content of the technical solution of the present invention shall still fall within the scope of protection of the technical solution of the present invention.

Claims

1. A kidney injury prediction system based on big data, characterized in that: The system comprises: The physiological data processing module is based on acute kidney injury monitoring data, including blood pressure, urine volume, and serum creatinine. It standardizes and sorts the collected data according to timestamps, synchronizes the data, and reorganizes the time-series monitoring data stream to generate a continuous synchronous monitoring sequence; The fluctuation analysis construction module extracts the time period change points of serum creatinine and urine volume based on the continuous synchronous monitoring sequence, determines the frequency of change of each indicator in the continuous time period, selects the abnormal fluctuation value interval as the sensitive time segment, and obtains the renal function fluctuation time window; The dynamic weight assessment module selects the continuous value change rate of blood pressure, urine volume, and creatinine within the time period according to the renal function fluctuation time window, compares each rate according to the proportional relationship and assigns a value to sort, assigns a dynamic weight level to each item according to the sort, and generates a dynamic weight table of physiological factors; The cross-domain data scheduling module calls the data types with the highest weight ratios in the dynamic weight table of physiological factors, cross-judges the upload frequency and node delay status of medical nodes in differentiated areas, selects nodes with high upload frequency and low delay and prioritizes them into the warehouse, and obtains a priority set of stable nodes.

2. The kidney injury prediction system based on big data according to claim 1, characterized in that: The continuous synchronous monitoring sequence includes a unified time axis, a standardized indicator sequence, and data synchronization accuracy parameters; the renal function fluctuation time window includes the change frequency distribution, the abnormal value concentration interval, and the indicator fluctuation amplitude range; the physiological factor dynamic weight table includes the indicator weight ranking, the rate comparison ratio, and the dynamic assignment level; the stable node priority set includes the node response delay, the data upload frequency, and the node selection priority.

3. The kidney injury prediction system based on big data according to claim 1, characterized in that: The physiological data processing module includes: The time series standardization submodule is based on acute kidney injury monitoring data, including original monitoring records and timestamps of blood pressure, urine volume, and serum creatinine. It sorts each monitoring record according to the timestamp, removes duplicate timestamps and time-missing records, determines the consistency of adjacent time intervals, and generates a standard time series set. The synchronization screening submodule sets a synchronization screening interval threshold based on the time node of each monitoring indicator in the standard time series set, determines the valid records of blood pressure, urine volume, and serum creatinine at each time node, eliminates nodes that are missing any indicator, and generates a synchronization observation node set; The indicator flow construction submodule calls the indicators under the time nodes in the synchronous observation node set, combines the indicators into indicator vectors in chronological order, and concatenates each vector in sequence to generate a continuous synchronous monitoring sequence.

4. The kidney injury prediction system based on big data according to claim 3, characterized in that: The fluctuation analysis building block includes: The period change identification submodule extracts adjacent measurement values of serum creatinine and urine volume based on the continuous synchronous monitoring sequence, calculates the absolute difference and compares it with the change threshold, selects the measurement nodes that meet the mutation condition, and generates a period change point group of serum creatinine and urine volume; The change frequency judgment submodule calls the time period change point group of serum creatinine and urine volume, counts the number of changes of each indicator within the time period, extracts the time periods of synchronous change, and generates a sequence of synchronous change frequency values of the indicators; The change sensitivity intensity identification submodule calls the indicator synchronous change frequency value sequence, identifies the change amplitude difference between adjacent time nodes, and normalizes the fluctuation amount using the formula: Calculate the change sensitivity intensity value, identify the time interval exceeding the abnormal threshold, and obtain the renal function fluctuation time window; Among them, F i represents the frequency of change in the i-th period, is the average frequency of change, V i is the combined volatility difference of the i-th period, D i is the change difference between the i-th period and the previous period, S i is the change sensitivity intensity value of the i-th period, and n represents the total number of periods.

5. The kidney injury prediction system based on big data according to claim 4, characterized in that: The dynamic weight evaluation module includes: The continuous rate calculation submodule extracts the continuous values of blood pressure, urine volume, and creatinine within the time period according to the renal function fluctuation time window, identifies the change amount of each item per unit time, and obtains a continuous change rate group; The rate ratio assignment submodule identifies the ratio difference between urine volume and creatinine rate according to the continuously changing rate group and the blood pressure rate, sorts them according to the ratio, assigns corresponding serial numbers, and generates a rate sorting sequence; The dynamic weight generation submodule analyzes the sequence relationship based on the rate sorting sequence and identifies the dynamic performance weight of each physiological indicator using the formula: Calculate the dynamic weight values corresponding to the indicators, match the weight values with the original indicators, and generate a dynamic weight table of physiological factors; Among them, W j Represents the dynamic weight value corresponding to the j-th indicator, ΔR j is the change in the rate of the jth index, Q is the blood pressure rate value, E j represents the reference rate corresponding to the j-th indicator rate, R j represents the rate of the j-th indicator, and N represents the total number of indicators.

6. The big data-based kidney injury prediction system according to claim 5, characterized in that: The cross-domain data scheduling module includes: The dynamic weight extraction submodule calls the physiological factor dynamic weight table, extracts the weight proportion of the data type, sorts them, identifies the upload frequency and node delay status of the top data types, and generates a weighted top data type set; The frequency-delay cross judgment submodule calls the upload frequency and node delay status of the weighted top data type set, and performs cross analysis based on the upload frequency, node delay and original data volume, combined with the number of concurrent nodes in the region and the average delay, using the formula: Calculate the node screening evaluation value, select the top-ranked nodes, and obtain the frequency delay optimal node set; Among them, K represents the node screening evaluation value, A x Represents the upload frequency value of the x-th node, B x represents the delay value of the xth node, C x represents the cumulative upload amount of original data of the xth node, D x Represents the number of concurrent nodes in the area where the x-th node is located, represents the average delay value of the node, and M represents the total number of nodes; The stable node screening submodule selects priority nodes according to the evaluation value ranking in the frequency delay preferred node set, determines whether the upload frequency and node status are continuous and stable, and screens out nodes that do not meet the conditions to obtain a stable node priority set.

7. The kidney injury prediction system based on big data according to claim 1, characterized in that: The system also includes a risk trend prediction module: The risk trend prediction module, based on the serum creatinine, urine protein, and blood pressure data uploaded in the stable node priority set, calculates the directional deviation of each indicator within a continuous interval, determines whether the deviation pattern remains consistent over multiple time periods, and if so, performs trend prediction to obtain a kidney injury trend prediction pattern label; The renal injury trend prediction model label includes indicator deviation directionality, trend maintenance period, and risk prediction type.

8. The big data-based kidney injury prediction system according to claim 7, characterized in that: The risk trend prediction module includes: The data offset identification submodule divides the time interval into chronological order based on the serum creatinine value, urine protein value and blood pressure value uploaded in the stable node priority set, extracts the continuous records of the indicators within the interval, identifies the change direction between adjacent records, and marks them as positive, negative and no change, and generates a directional offset sequence value; The offset pattern consistency submodule calls the directional offset sequence value, compares the directional offset state of each indicator in the differentiated time interval, selects the interval sequence with a continuous and consistent offset direction, calculates the proportion of the consistent directional state, and obtains a continuous offset consistency ratio value; The trend dynamic prediction submodule identifies the direction offset sequence of each indicator based on the continuous offset consistency ratio value, sets the trend judgment benchmark according to the offset continuation direction and the number of time intervals, screens the indicator paths that meet the trend conditions, and obtains the kidney injury trend prediction pattern label.

Images