Presepsis early warning method based on multi-parameter time sequence pattern recognition

Abstract

The invention discloses a sepsis early warning method based on multi-parameter time sequence pattern recognition, and relates to the technical field of medical information processing, and the sepsis early warning method comprises the following steps: obtaining urine volume time sequence data and turbidity time sequence data of a target object; performing double-sequence pairing and change measurement on the urine volume time sequence data and the turbidity time sequence data to generate quality weights, and performing credible convergence on the quality weights to generate effective section marks; and in the effective section mark, extracting a stable repetition form from the urine volume time sequence data according to the mass weight and a continuous time window, and converging to generate a rhythm baseline. According to the scheme, the stable repeated form is extracted from the urine volume time sequence data in the effective section mark according to the time window to generate the rhythm baseline, and the deviation index is generated through same-window comparison, so that individualized rhythm reference and deviation quantification are realized, report missing caused by individual difference is reduced, and rhythm damage is captured in advance.

Description

Technical Field

[0001] This invention relates to the field of medical information processing technology, specifically to an early warning method for sepsis based on multi-parameter temporal pattern recognition. Background Technology

[0002] Sepsis is a critical illness caused by infection that can lead to systemic inflammatory response and organ dysfunction. It progresses rapidly and has a short window period. Clinically, it is usually identified and managed by vital signs, laboratory inflammatory markers, and organ function-related indicators. Especially in the monitoring scenario of long-term indwelling catheterization in urology, the risk of urinary tract infection is higher and the course of the disease is insidious. Urogenital sepsis often shows observable early changes in urine characteristics and excretion rhythm before obvious fever, hemodynamic abnormalities, or significant abnormalities in laboratory indicators.

[0003] Existing technologies mainly rely on monitoring methods based on changes in urine volume or urine turbidity. For example, they assess kidney function or infection risk by monitoring the absolute change in urine volume per unit time, or they use turbidity thresholds to indicate the possibility of urinary tract infection. The advantage of these methods is that data acquisition is relatively simple and easy to deploy at the bedside for a long time. However, these methods often use uniform thresholds or uniform change ranges as the basis for judgment, which is difficult to adapt to the individual differences in urine volume rhythm formed by the combined effects of individual baseline levels and diurnal rhythms. Therefore, it is easy to misjudge normal rhythm fluctuations as abnormal or mask true abnormalities as normal, resulting in false alarms and false negatives. Moreover, abnormal urine volume is not only caused by infection progression. Catheter obstruction or kinking can directly change the urine excretion process and cause morphological changes such as sudden drops in urine volume, cessation of flow, and short-term rebound after repositioning. This further amplifies the risk of false alarms and false negatives caused by the uniform thresholds, thereby reducing the stability and reliability of the early warning judgment. Summary of the Invention

[0004] The purpose of this invention is to provide an early warning method for sepsis based on multi-parameter temporal pattern recognition, so as to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the technical solution of the present invention is as follows:

[0006] In a first aspect, this invention discloses an early warning method for sepsis based on multi-parameter temporal pattern recognition, applied to the early warning of urinary sepsis under long-term catheterization monitoring in urology, comprising the following steps:

[0007] Acquire time-series data of urine volume and turbidity of the target object;

[0008] The urine volume time series data and the turbidity time series data are paired and the change is measured to generate quality weights. The quality weights are then reliably aggregated to generate valid segment labels.

[0009] Within the effective segment marker, stable repetitive patterns are extracted from the urine volume time series data according to the quality weight and aggregated according to the continuous time window to generate a rhythm baseline. The urine volume time series data and the rhythm baseline are compared with each other in the same window and then combined with the quality weight to express the deviation and generate a deviation index.

[0010] Within the effective segment marker, the urine volume time series data is morphologically scanned and cross-validated with the rhythm baseline and the deviation index to generate an obstruction index. Based on the obstruction index, the turbidity time series data is subjected to transition extraction processing for contradictory evidence constraints to generate turbidity transition evidence.

[0011] The deviation index, the blockage index, and the turbidity transition evidence are fused together to generate an evidence matrix. The evidence matrix is ​​then subjected to quality weighting and deviation accumulation constraints based on the quality weight and the deviation index to generate an early warning result.

[0012] Secondly, this invention discloses an early warning system for sepsis based on multi-parameter temporal pattern recognition, comprising:

[0013] The data acquisition module is used to acquire time-series data of urine volume and turbidity of the target object;

[0014] The effective segment segmentation module is used to perform dual-sequence pairing and change measurement on the urine volume time series data and the turbidity time series data, generate quality weights, and perform reliable aggregation on the quality weights to generate effective segment labels.

[0015] The deviation analysis module is used to extract stable repetitive patterns from the urine volume time series data according to the quality weight within the effective segment marker, and aggregate them to generate a rhythm baseline. The urine volume time series data and the rhythm baseline are compared with each other in the same window and then combined with the quality weight to express the deviation and generate a deviation index.

[0016] The transition analysis module is used to perform morphological scanning on the urine volume time series data within the effective segment marker, and then cross-validate it with the rhythm baseline and the deviation index to generate an obstruction index. Based on the obstruction index, the module performs transition extraction processing on the turbidity time series data with contradictory constraints to generate turbidity transition evidence.

[0017] The early warning generation module is used to perform window difference fusion on the deviation index, the blocking index and the turbidity transition evidence to generate an evidence matrix, and to perform quality weighting and deviation accumulation constraints on the evidence matrix according to the quality weight and the deviation index to generate an early warning result.

[0018] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0019] 1. This scheme generates a rhythm baseline by extracting stable and repetitive patterns from urine volume time-series data within the effective segment markers according to time windows, and generates a deviation index by comparing with peers. This achieves individualized rhythm reference and deviation quantification, reduces underreporting caused by individual differences, and detects rhythm disruption in advance. By performing morphological scanning on urine volume time-series data within the effective segment markers and cross-validating it with the rhythm baseline and deviation index to generate an obstruction index, it can distinguish between obstructive urine volume abnormalities and infectious changes, avoid mistaking obstruction for infection, and improve the consistency of judgment.

[0020] 2. This scheme generates stable candidate segments by judging adjacent differences that are zero and corresponding sums that are greater than zero, and then merging them to form stable intervals. This aggregates continuous flow cessation patterns from fragmented points into stable intervals, reducing false alarms of short pauses being mistaken for obstruction. By contouring the stable candidate segments with the rhythm baseline to generate baseline violation expressions and aggregating deviation support expressions within the deviation index window, normal low urine volume is excluded using dual evidence of individual baseline and abnormal intensity, and the discrimination of true abnormal flow cessation is strengthened. After the stable candidate segments, the adjacent differences are continuously incrementally scanned according to the alignment index to generate rising candidate segments, and the obstruction index is generated through alignment mapping. This achieves the capture of the release pattern of rising flow after cessation and outputs an interpretable degree of obstruction to distinguish between obstructive drops and infectious changes. Attached Figure Description

[0021] The disclosure of this invention is illustrated with reference to the accompanying drawings. It should be understood that the drawings are for illustrative purposes only and are not intended to limit the scope of protection of this invention. In the drawings, the same reference numerals are used to refer to the same parts. Wherein:

[0022] Figure 1 A flowchart illustrating the steps of the sepsis early warning method based on multi-parameter temporal pattern recognition provided by the present invention;

[0023] Figure 2 A schematic diagram of the process for generating the blocking index provided by the present invention;

[0024] Figure 3 This is a schematic diagram of the process for generating reconstructed urine volume data provided by the present invention;

[0025] Figure 4 A schematic diagram of the process for generating an evidence matrix provided by the present invention;

[0026] Figure 5 This is a schematic diagram of the module functions of the sepsis early warning system based on multi-parameter temporal pattern recognition provided by the present invention. Detailed Implementation

[0027] It is readily understood that, based on the technical solution of this invention, those skilled in the art can propose various interchangeable structural methods and implementations without altering the essential spirit of the invention. Therefore, the following detailed embodiments and accompanying drawings are merely illustrative examples of the technical solution of this invention and should not be considered as the entirety of the invention or as limitations or restrictions on the technical solution of this invention.

[0028] Application Overview:

[0029] Traditional early warning methods for sepsis rely on fixed thresholds to monitor and determine changes in urine volume. This makes it difficult to adapt to individual differences in baseline levels and the circadian rhythm fluctuations in urine volume. This can lead to normal physiological changes being misjudged as abnormal events or true pathological changes being masked as normal. Furthermore, changes in urine output caused by catheter events such as obstruction or kinking, including sudden drops in urine volume, flow cessation, and short-term rebound after repositioning, further interfere with the threshold determination logic. As a result, the risk of false alarms and missed alarms in the early warning system increases, affecting the stability and reliability of the judgment.

[0030] For example, in a long-term catheterization monitoring scenario in urology, a patient with chronic renal insufficiency may have a baseline urine output below the standard threshold. Furthermore, due to circadian rhythms, nighttime urine output exhibits a physiological decrease. Current technology may identify this normal rhythmic fluctuation as abnormal, triggering false alarms. Simultaneously, when the catheter experiences a slight kink, urine output may plummet to zero, leading the system to misinterpret this as infection progression. However, early changes in actual urinary sepsis may not be captured due to individual differences, resulting in missed reports. Specifically, in this scenario, a uniform threshold makes it difficult to distinguish between physiological rhythmic fluctuations and pathological changes. Moreover, changes in urine output patterns caused by catheter events overlap with infection-related changes in data representation, leading to inaccurate early warning judgments and increasing the complexity of clinical management.

[0031] If the above problems are not addressed, the reliability of the early warning system will be insufficient to meet the clinical needs for early identification of sepsis. False alarms may lead to unnecessary medical interventions, while missed alarms may cause patients to miss the critical treatment window, thereby allowing the disease to progress to an irreversible stage, endangering patient safety and reducing the efficiency of medical resource utilization.

[0032] After introducing the basic concept of the present invention, the embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0033] Example 1:

[0034] Please see Figure 1 An early warning method for sepsis based on multi-parameter temporal pattern recognition is applied to the early warning of urinary sepsis under long-term catheterization monitoring in urology, including the following steps:

[0035] Acquire time-series data of urine volume and turbidity of the target object;

[0036] The urine volume time series data and turbidity time series data were paired and the change was measured to generate quality weights. The quality weights were then reliably aggregated to generate valid segment labels.

[0037] Within the effective segment markers, stable repetitive patterns of urine volume time-series data are extracted and aggregated according to continuous time windows based on quality weights to generate a rhythm baseline. The urine volume time-series data and the rhythm baseline are compared with each other in the same window and then combined with quality weights to express the deviation index.

[0038] Within the effective segment marker, the urine volume time series data is morphologically scanned and cross-validated with the rhythm baseline and deviation index to generate the obstruction index. Based on the obstruction index, the turbidity time series data is subjected to the transition extraction process of the counter-evidence constraint to generate turbidity transition evidence.

[0039] The evidence of deviation index, blockage index and turbidity transition is fused by peer difference to generate evidence matrix. The evidence matrix is ​​then subjected to quality weighting and deviation accumulation constraint according to quality weight and deviation index to generate early warning results.

[0040] Among them, urine volume time series data refers to raw numerical sequence data that can reflect the changes in urine output of the target object on a continuous time axis;

[0041] Turbidity time series data refers to a data sequence that reflects the relationship between urine turbidity and time, obtained by continuous optical detection of urine excreted by a target object under natural flow conditions.

[0042] Quality weights refer to numerical weights used to characterize the reliability of data at sampling locations or within sampling intervals.

[0043] Valid segment markers refer to time-limited data that simultaneously satisfy the requirements of high consistency and reliable continuity of the two sequences in time-series data.

[0044] Rhythm baseline refers to individualized urine output rhythm reference data used to characterize the target subject under long-term catheterization.

[0045] Deviation index refers to time-series data used to reflect the magnitude and trend of deviation of the current urine output change of a target subject relative to its own stable rhythm state;

[0046] The obstruction index is a single numerical time-series data used to characterize whether abnormal urine output can be explained by catheter obstruction or torsion.

[0047] Turbidity transition evidence refers to data results formed by jointly characterizing the structural level switching behavior and the significant change in fluctuation intensity of turbidity over time, which are used to characterize whether turbidity changes from a stable state to an abnormal evolution state.

[0048] An evidence matrix is ​​a data structure used to characterize the peer relationship between evidence of abnormal urine volume and evidence of abnormal turbidity.

[0049] Early warning results refer to a set of risk assessment information used for supplementary reference.

[0050] This scheme pairs urine volume time-series data and turbidity time-series data and generates quality weights by measuring changes, and reliably aggregates them to generate effective segment markers. It suppresses weighing jitter and optical transient interference by measuring cross-series consistency, automatically delineates reliable continuous intervals, reduces false alarms caused by noise, extracts stable repetitive patterns according to quality weights within the effective segment markers to generate a rhythm baseline, and generates a deviation index by peer comparison, forming an individualized urine volume rhythm reference and expressing abnormalities as relative deviations and continuous accumulation, weakening the impact of individual differences and exposing hidden rhythm disruptions in advance.

[0051] Within the effective segment marker, morphological scanning is performed and cross-validated with the rhythm baseline and deviation index to generate an obstruction index. Turbidity transition evidence is extracted using the contra-evidence constraint to distinguish between sudden changes in urine volume caused by obstruction / torsion and suppress its interference with false transitions in turbidity. This improves the accuracy and lead time for infection transition identification. The deviation index, obstruction index, and turbidity transition evidence are fused together to generate an evidence matrix. The results are then output as early warnings with quality weighting and deviation accumulation constraints. This achieves consistency judgment and conflict suppression of multiple evidences and outputs verifiable graded early warnings.

[0052] The above describes the complete scheme for an early warning method for sepsis based on multi-parameter time-series pattern recognition. The following describes the acquisition of time-series urine volume and turbidity data of the target subjects, specifically including:

[0053] Urine volume time-series data of the target object is obtained through a urinary catheter flow meter; the urine volume time-series data includes, but is not limited to, the sampling time information, the corresponding urine volume value, the sampling sequence number, the sampling interval identifier, and the working status identifier of the acquisition unit for each sampling;

[0054] Turbidity time-series data of the target object is acquired through an optical turbidity acquisition device; the turbidity time-series data includes, but is not limited to, the sampling time information, the corresponding turbidity value, the sampling sequence number, the sampling interval identifier, and the working status identifier of the acquisition unit for each sampling.

[0055] The above describes how to obtain time-series data on urine volume and turbidity of the target object. The following describes how to perform two-series pairing and change measurement on the urine volume and turbidity time-series data to generate quality weights, specifically including:

[0056] The urine volume time series data and turbidity time series data are arranged coaxially and a point-by-point correspondence is established to generate an alignment index;

[0057] Based on the alignment index, consistency and inconsistency measures are performed on pairs of urine volume time series data and turbidity time series data to generate quality weights.

[0058] The alignment index refers to a data object used to describe the point-to-point correspondence between urine volume time series data and turbidity time series data in the same time dimension.

[0059] The above content will be described in detail below:

[0060] The urine volume time series data and turbidity time series data are arranged coaxially and a point-by-point correspondence is established to generate an alignment index:

[0061] Extract the sampling time series and corresponding numerical series from the urine volume time series data and the turbidity time series data respectively. Obtain the urine volume time interval series by subtracting adjacent sampling times in the urine volume sampling time series, and obtain the turbidity time interval series by subtracting adjacent sampling times in the turbidity sampling time series.

[0062] The median of the urine volume time interval sequence and the turbidity time interval sequence are taken to obtain the typical intervals of urine volume and turbidity, respectively. The smaller of the typical intervals of urine volume and turbidity is taken as the alignment step size. Based on this, the larger of the first time of the urine volume sampling time sequence and the first time of the turbidity sampling time sequence is taken as the alignment start time, and the smaller of the last time of the urine volume sampling time sequence and the last time of the turbidity sampling time sequence is taken as the alignment end time. The alignment start time to the alignment end time is divided equally according to the alignment step size to generate an alignment time sequence.

[0063] For each alignment time in the alignment time series, calculate the absolute value of the time difference between it and each sampling time in the urine sampling time series, and take the urine sampling number corresponding to the smallest absolute value of the time difference. Calculate the absolute value of the time difference between it and each sampling time in the turbidity sampling time series, and take the turbidity sampling number corresponding to the smallest absolute value of the time difference. This forms a point-to-point correspondence between urine sampling number and turbidity sampling number for each alignment time. Assign consecutive alignment numbers to each point-to-point correspondence according to the chronological order of the alignment time series. Summarize and output each record in the form of alignment number, alignment time, urine sampling number, and turbidity sampling number to obtain the alignment index.

[0064] Based on the alignment index, consistency and inconsistency measures are performed on the pairwise urine volume time series data and turbidity time series data to generate quality weights:

[0065] The urine sampling points and turbidity sampling points corresponding to the same sampling number are paired according to the alignment index to form a pairing sequence;

[0066] The difference between the urine volume sampling points of two adjacent sampling numbers is calculated. The urine volume difference is obtained by subtracting the urine volume value of the previous sampling number from the urine volume value of the later sampling number. The absolute value of the urine volume difference is then taken to obtain the change in urine volume.

[0067] The difference is calculated for the turbidity sampling points of two adjacent sampling numbers. The turbidity difference is obtained by subtracting the turbidity value of the previous sampling number from the turbidity value of the later sampling number, and the absolute value of the turbidity difference is obtained to obtain the turbidity change.

[0068] The difference between the change in urine volume and the change in turbidity at the same sampling number is calculated. The change in urine volume is subtracted from the change in turbidity and the absolute value is taken to obtain the mutual verification difference. At the same time, the consistency of the positive and negative directions of the difference in urine volume and the difference in turbidity is measured. Specifically, when the difference in urine volume and the difference in turbidity are both positive or both negative, it is recorded as consistent direction. When the difference in urine volume and the difference in turbidity are one positive and the other negative, it is recorded as inconsistent direction. Consistent direction is recorded as 0 and inconsistent direction is recorded as 1 to form the inconsistent direction quantity.

[0069] The inconsistency metric is obtained by adding the mutual verification difference and the direction inconsistency metric, and then mapped to a quality weight. The specific calculation formula is as follows:

[0070] ;

[0071] In the formula, This represents a measure of inconsistency; all the data above have been normalized during the calculation.

[0072] The quality weights are reliably aggregated to generate valid segment labels:

[0073] Calculate the sum of quality weights and the total number of sampling points, divide the sum of quality weights by the total number of sampling points to obtain the mean of quality weights, and then divide the mean of quality weights by two to obtain the baseline weights.

[0074] For each sampling point, a baseline deduction operation is performed on the mass weight. Specifically, the deduction weight sequence is obtained by subtracting the baseline weight from the mass weight at that sampling point.

[0075] A prefix convergence sequence is generated from the deducted weight sequence. Specifically, the deducted weight sequence is sequentially accumulated starting from the first sampling point, and the accumulated result at each sampling point is used as the corresponding prefix convergence value.

[0076] At each sampling point, the difference between the prefix convergence value of that sampling point and the minimum value among all the prefix convergence values ​​before that sampling point is taken to obtain the endpoint convergence difference value of that sampling point. At the same time, the sampling point that produces the minimum value is recorded as the candidate sequence number of the interval start point. Among the endpoint convergence differences of all sampling points, the sampling point corresponding to the maximum value is taken as the interval end point number, and the corresponding record of the interval start point candidate sequence number is taken as the interval start point number.

[0077] The starting and ending numbers of the interval, along with the range of continuous sampling points they cover, are summarized and output as valid segment markers.

[0078] This scheme establishes a point-by-point correspondence by coaxially arranging urine volume time-series data and turbidity time-series data, generating an alignment index. This enables point-by-point comparison and window alignment of the two types of time-series data under the same sampling sequence number, eliminating mismatch errors caused by asynchronous sampling. It provides a unified positioning benchmark for subsequent cross-series references. Based on the alignment index, the scheme measures the consistency and inconsistency of paired data, generates quality weights, and quantifies mutually inconsistent segments such as stable urine volume with isolated turbidity spikes or stable turbidity with isolated urine volume jumps as low confidence segments, and quantifies synchronous structural change segments as high confidence segments. This automatically reduces the weight of noisy segments and highlights effective segments, reducing false alarms and missed alarms.

[0079] The above describes the pairing and variation measurement of urine volume and turbidity time-series data to generate quality weights. The following describes the extraction of stable repetitive patterns from urine volume time-series data according to continuous time windows based on these quality weights, followed by aggregation to generate a rhythm baseline. Specifically, this includes:

[0080] Within the valid segment markers, urine volume time series data are sequentially extracted according to the alignment index, and the sequential extraction results are weighted according to the quality weight by equal time windows to generate urine volume change profiles.

[0081] Perform cross-window similarity aggregation on the urine volume change profile, and select the urine volume change profile with the largest cross-window similarity aggregation result as the rhythm baseline.

[0082] Among them, the sequential extraction result refers to the basic data set formed for time window weighting processing after reading the urine time series data in a continuous, non-jumping and non-rearranged order according to the sampling sequence number within the range of valid segment markers based on the alignment index.

[0083] Urine volume change profile refers to an ordered set of data used to describe the pattern of urine volume changes within a time window;

[0084] Cross-window similarity aggregation results refer to the comprehensive results used to measure the degree of similarity between urine volume changes in different time windows.

[0085] The above content will be described in detail below:

[0086] Within the valid segment markers, urine volume time-series data are sequentially extracted according to the alignment index, and the sequential extraction results are weighted according to equal time windows based on quality weights to generate a urine volume change profile:

[0087] Within the sampling point range indicated by the effective segment marker, urine sampling points corresponding to each sampling point are extracted from the urine volume time series data according to the alignment index in ascending order of the sampling points. The effective segment urine volume sequence is formed by arranging the sampling points in the extraction order and using the effective segment urine volume sequence as the extraction result.

[0088] The effective urine volume sequence is divided into equal time windows of fixed length. For any time window, the absolute value of the urine volume difference contained in the time window is taken and summed item by item to obtain the total change of the time window. At the same time, the weight values ​​corresponding to each sampling point in the time window are extracted from the quality weights according to the same alignment index and summed item by item to obtain the total weight of the time window. Then, the total change of the time window is divided by the total weight of the time window to obtain the weighted change of the time window.

[0089] Calculate the corresponding weighted change for each time window, and then arrange and summarize the weighted changes for each time window in order of time window to obtain the profile of urine volume change.

[0090] The cross-window similarity aggregation is performed on the urine volume change profile, and the specific calculation formula is as follows:

[0091] ;

[0092] In the formula, Indicates the first Cross-window similarity aggregation results for each time window This represents the total number of time windows participating in the aggregation. This represents the length of the profile of urine volume changes within each time window. Indicates the first Within the first time window The difference in urine volume at each location, Indicates the first Within the first time window The urine volume difference at each location; all the above data have been normalized during the calculation.

[0093] The urine volume change profile with the largest cross-window similarity aggregation result was selected as the rhythm baseline.

[0094] This scheme extracts urine volume time-series data sequentially within the effective segment markers based on alignment indices, and generates a urine volume change profile by weighting the sequential extraction results according to quality weights and equal time windows. This reduces the influence of low-confidence sampling points on the profile, making the profile mainly composed of high-confidence segments. This reduces the contamination of rhythm features by weighing jitter and transient disturbances and improves profile stability. Cross-window similarity aggregation is performed on the urine volume change profile, and the urine volume change profile with the largest similarity aggregation result is selected as the rhythm baseline. The recurring stable pattern is solidified into an individual baseline, reducing baseline drift caused by individual differences and occasional fluctuations. Subsequent deviation judgments are compared with relative deviation rather than absolute urine volume, thereby reducing false alarms and false negatives and improving rhythm recognition consistency in long-term catheterization scenarios.

[0095] The above describes how stable repetitive patterns are extracted from urine volume time-series data according to continuous time windows and aggregated based on quality weights to generate a rhythm baseline. The following describes how to express the deviation index by combining urine volume time-series data with the rhythm baseline after peer comparison with the rhythm baseline and incorporating quality weights. Specifically, this includes:

[0096] Based on the rhythm baseline, urine volume time series data are aligned and paired within a window to generate a bias intensity expression. The bias intensity expression is then subjected to window weighting constraint based on quality weights to obtain the bias sequence.

[0097] Perform cross-window coherent convergence and scale unification on the deviation sequence to generate the deviation index.

[0098] Among them, the deviation intensity expression refers to the data expression used to characterize the overall degree of deviation of the actual change in urine volume within a time window from the baseline change.

[0099] A deviation sequence is a set of numerical sequences arranged in order of time windows used to characterize the degree of deviation of actual urine volume changes from an individual's rhythm baseline.

[0100] The above content will be described in detail below:

[0101] Based on the rhythm baseline, urine volume time series data are aligned and paired within a window to generate a bias intensity expression. Then, a window-weighted constraint is applied to the bias intensity expression based on quality weights to obtain the bias sequence.

[0102] Within each time window, the urine volume time series data of the first sampling point is paired with the rhythm baseline of the first sampling point in the same window to form the first pair, and the urine volume time series data of the second sampling point is paired with the rhythm baseline of the second sampling point in the same window to form the second pair, until the last sampling point in the time window forms the last pair, thus obtaining the pairing set within the time window. On this pairing set, a deviation intensity expression is generated: for each pair of paired points, the absolute value of the difference between the urine volume time series data and the rhythm baseline is calculated as the deviation value of the paired point, and all deviation values ​​within the time window are accumulated in the pairing order to obtain the total deviation.

[0103] The weighted deviation value of the time window is obtained by dividing the total deviation by the sum of the weights. The weighted deviation value is the expression of the deviation intensity of the time window.

[0104] Repeat the above calculation process of in-window alignment and pairing, total deviation accumulation, total weight accumulation and total deviation divided by total weight for all time windows, and arrange and summarize the deviation intensity expressions of each time window in the order of time windows to obtain the deviation sequence;

[0105] Perform cross-window coherent convergence and scale unification on the biased sequences to generate the bias index:

[0106] The deviation values ​​corresponding to the current time window are retrieved sequentially according to the order of the time windows. The difference between the deviation value and the deviation value of the previous time window is calculated and the absolute value is taken as the inter-window fluctuation. At the same time, the deviation values ​​of the current time window and the deviation values ​​of the previous time window are summed to obtain the inter-window cumulative amount.

[0107] The coherence coefficient is calculated by comparing the cumulative amount between windows with the fluctuation amount between windows. The coherence deviation value is obtained by multiplying the deviation value of the current time window with the coherence coefficient. All coherence deviation values ​​are then summarized in the order of the time windows to form a coherence sequence.

[0108] Standardize the scale for coherent sequences: Search for the maximum and minimum coherence deviation values ​​in the coherent sequence, and calculate the baseline removal value by subtracting the minimum coherence deviation value from each coherence deviation value. Then, calculate the normalized value by dividing the baseline removal value by the difference between the maximum and minimum coherence deviation values. Finally, summarize the normalized values ​​of each time window in the order of the time windows and output them as the deviation index.

[0109] This scheme performs in-window alignment and pairing of urine volume time-series data based on the rhythm baseline to generate a deviation intensity expression. This transforms the absolute urine volume differences between different target subjects and different catheterization stages into relative baseline deviations, ensuring that the deviation only reflects the degree to which the current urine volume change violates the individual rhythm. This improves the sensitivity of anomaly localization and reduces false alarms caused by individual differences. Furthermore, the deviation intensity expression is subjected to window weighting constraints based on quality weights to obtain a deviation sequence. Low-confidence windows affected by jitter, light shading, etc., are automatically downweighted, ensuring that the deviation sequence is dominated by high-confidence segments. This reduces the amplification effect of noise peaks on judgment and improves stability. The deviation sequence is then subjected to cross-window coherent convergence and scale unification to generate a deviation index. This distinguishes short-term fluctuations from continuous deviations, highlights the cumulative characteristics of continuous anomalies, and achieves uniform dimensional comparability. This allows hidden progression to manifest as a continuously rising deviation index in the early stages, enhancing early warning capabilities.

[0110] The above describes how to express deviation by comparing urine volume time-series data with the rhythm baseline using peer review and incorporating quality weights, generating a deviation index. The following describes how to perform morphological scanning on urine volume time-series data and cross-reference it with the rhythm baseline and deviation index to generate an obstruction index. Please refer to [link / reference]. Figure 2 , Figure 2 This is a schematic diagram of the process for generating a blocking index provided in an embodiment of this application. Generating the blocking index specifically includes:

[0111] The sequential extraction results corresponding to the urine volume time series data are subjected to adjacent difference calculation, and the corresponding two quality weights are summed.

[0112] Determine whether the difference between adjacent subtraction results is zero and whether the corresponding summation result is greater than zero. If so, merge the corresponding adjacent subtraction results to generate a stationary candidate segment.

[0113] By contouring the stationary candidate segments with the rhythmic baseline, baseline violation expressions are generated, and deviation indices are aggregated within a window to generate deviation support expressions.

[0114] After the stable candidate segments, the adjacent difference results are continuously incrementally scanned according to the alignment index to generate the rising candidate segments. The baseline violation expression, deviation support expression and rising candidate segments are aligned and mapped to generate the blocking index.

[0115] Among them, the adjacent difference result refers to the data structure used to characterize the change state of urine volume time series data at adjacent sampling positions;

[0116] The summation result refers to the numerical result used to characterize the strength of the mutual credibility of adjacent difference results at the current sampling position;

[0117] A stationary candidate segment is a continuous sampling interval used to characterize urine volume as approximately constant over a continuous period of time.

[0118] Baseline violation expression refers to a data structure used to characterize the degree to which candidate segments of stable urine volume deviate from an individual’s normal urination rhythm over time.

[0119] Deviation support expression refers to in-window aggregated data used to characterize whether the degree of abnormal deviation in urine output corresponding to a certain stationary candidate time period has persistence and concentration.

[0120] The recovery candidate segment refers to a continuous data interval used to characterize the transition of urine volume from an abnormally low state to a state of recovery and growth.

[0121] The above content will be described in detail below:

[0122] The sequential extraction results corresponding to the urine volume time series data are subjected to adjacent difference calculation, and the corresponding two quality weights are summed.

[0123] Determine whether the adjacent difference results are zero and whether the corresponding summation results are greater than zero. If so, merge the corresponding adjacent difference results into continuous intervals according to the sampling order, and map each continuous interval into a stationary candidate segment. The starting position of the stationary candidate segment is determined by the minimum sampling number in the continuous interval, and the ending position is determined by the maximum sampling number in the continuous interval.

[0124] By contouring stationary candidate segments against the rhythmic baseline, baseline violation expressions are generated, and deviation indices are aggregated within a window to generate deviation support expressions.

[0125] The urine volume time series data of the stationary candidate segment at each sampling point is subtracted from the value of the rhythm baseline at the same sampling point, and the absolute value is taken. The point value deviation sequence is obtained according to the sampling order.

[0126] The total point value deviation is obtained by summing all the absolute values ​​in the point value deviation sequence point by point, and the average point value deviation is obtained by dividing the total point value deviation by the number of sampling points contained in the stationary candidate segment.

[0127] The candidate change sequence is obtained by subtracting the urine volume time series data of adjacent sampling points of the stationary candidate segment, and the baseline change sequence is obtained by subtracting the values ​​of adjacent sampling points of the rhythm baseline. The candidate change sequence and the baseline change sequence are then subtracted at the same adjacent positions and the absolute value is taken. The change profile deviation sequence is obtained in the order of adjacent positions.

[0128] The total amount of the change profile deviation is obtained by summing all the absolute values ​​in the change profile deviation sequence point by point, and the average amount of the change profile deviation is obtained by dividing the total amount of the change profile deviation by the number of adjacent positions. The baseline violation expression is obtained by adding the average point value deviation to the average change profile deviation.

[0129] The starting and ending sampling points of the stationary candidate segment are mapped to the time window range corresponding to the deviation index. The deviation index value sequence arranged in time window order within the time window range is extracted, and the cumulative deviation is calculated: all values ​​in the deviation index value sequence are added window by window to obtain the cumulative deviation, and the cumulative deviation is divided by the number of time windows within the time window range to obtain the mean deviation. Finally, the mean deviation is output as the deviation support expression.

[0130] After the stationary candidate segments, the adjacent difference results are continuously incrementally scanned according to the alignment index to generate rising candidate segments. Then, the baseline violation expression, deviation from supporting expression, and rising candidate segments are aligned and mapped to generate the blocking index.

[0131] The interval of consecutive sampling numbers after the termination sampling number of the stable candidate segment is determined by the alignment index and is used as the rising scan interval;

[0132] Within the recovery scan interval, a positive increment sequence is generated on adjacent difference results. The calculation process of the positive increment sequence is as follows: for each difference result of adjacent difference results, take the larger value with zero to obtain the positive increment value corresponding to each sampling point, and summarize them in the order of sampling points so that the negative difference value corresponds to zero in the positive increment sequence. Based on the positive increment sequence, a continuous increment scan is performed to generate a set of candidate recovery intervals. The calculation process of the continuous increment scan is as follows: the sampling points corresponding to the non-zero increment values ​​in the positive increment sequence are divided into connected segments according to the adjacent relationship to obtain incremental connected segments composed of continuous sampling points, and the starting sampling point and ending sampling point of each incremental connected segment are recorded as recovery candidate intervals.

[0133] For each candidate interval of recovery, calculate the cumulative recovery amount. The calculation process of the cumulative recovery amount is as follows: sum the positive increment values ​​within the coverage area of ​​the candidate interval of recovery point by point to obtain the cumulative sum, and use the cumulative sum as the cumulative recovery amount of the candidate interval of recovery. Select the candidate interval of recovery with the largest cumulative recovery amount as the candidate segment of recovery and output its starting sampling point and ending sampling point.

[0134] Based on the alignment index, the sampling point range of the rising candidate segment is mapped to the time window or sampling point range of the baseline violation expression and the deviation support expression, so that the three have the same sampling point coverage interval. Then, the baseline violation aggregation value is obtained by summing the baseline violation expression values ​​within the coverage interval of the rising candidate segment, and the deviation support aggregation value is obtained by summing the deviation support expression values. The cumulative rising value, the baseline violation aggregation value, and the deviation support aggregation value are summed to obtain the blocking mapping value. Finally, the blocking mapping value is normalized by the ratio of the starting sampling point of the rising candidate segment minus the ending sampling point of the stationary candidate segment plus one, to generate the blocking index.

[0135] This scheme performs adjacent subtraction on the sequential extraction results corresponding to urine volume time series data and sums the two quality weights. The difference characterizes the local changes in urine volume and the weights aggregate to characterize the reliability of the changes. It suppresses the interference of jitter points on morphological judgment from the source. When the adjacent subtraction result is zero and the corresponding summation result is greater than zero, it is merged to generate stable candidate segments. The continuous and approximately unchanged flow cessation morphology is aggregated from discrete points into a locatable interval and low-confidence zero-difference segments are eliminated to reduce false detections.

[0136] By comparing stable candidate segments with the rhythm baseline profile to generate baseline violation expressions and aggregating deviation support expressions within the deviation index window, the abnormality degree and the persistence of abnormality are quantified, thereby distinguishing between normal low urine output and obstructive cessation of flow. After stable candidate segments, continuous incremental scanning is performed according to the alignment index to generate rebound candidate segments, and the baseline violation expression, deviation support expression and rebound candidate segments are aligned and mapped to generate an obstruction index. The rebound pattern after cessation of flow is bound to the aforementioned mutually corroborating evidence to form a comprehensive explanation of the obstruction release process, reducing the probability of misjudging infection progression as obstruction or obstruction as infection.

[0137] The above describes the process of performing morphological scanning on urine volume time-series data and cross-validating it with the rhythm baseline and deviation index to generate an obstruction index. The following section describes the subsequent steps after generating the obstruction index, including generating reconstructed urine volume data. Please refer to [link to relevant documentation]. Figure 3 , Figure 3 This is a schematic diagram of the process for generating reconstructed urine volume data provided in an embodiment of this application. Generating reconstructed urine volume data specifically includes:

[0138] Based on the obstruction index, the urine volume time series data are windowed, summarized, and sorted to generate abnormal intervals. Within the abnormal intervals, the rhythm baselines are differentially analyzed to generate baseline increment sequences.

[0139] Within the abnormal interval, the baseline incremental sequence is jointly scaled point by point according to the deviation index and the quality weight to generate a compensated incremental sequence.

[0140] Starting with urine volume time series data, the compensated incremental sequence is accumulated and spliced ​​point by point to generate reconstructed urine volume data.

[0141] Among them, the abnormal interval refers to the set of continuous time windows that are automatically identified in urine volume time series data based on the obstruction index, where the urine volume change behavior deviates significantly from the normal rhythm and the deviation can be reasonably explained by the obstruction pattern.

[0142] Baseline increment sequence refers to a data structure used to characterize the theoretical variation amplitude between adjacent sampling points of the rhythm baseline within the abnormal interval;

[0143] Compensation increment sequence refers to a data structure used to describe the amount of urine volume change that should be compensated between each adjacent sampling point when correcting for changes in urine volume under baseline constraints within an interval deemed abnormal.

[0144] The above content will be described in detail below:

[0145] Urine volume time-series data are windowed, summarized, and sorted according to the obstruction index to generate abnormal intervals. Within these abnormal intervals, adjacent differences are performed on the rhythm baseline to generate baseline increment sequences.

[0146] The time window set corresponding to the rhythm baseline is used to segment the urine volume time series data and the obstruction index into the same window, so that each time window covers the same range of continuous sampling points;

[0147] Within each time window, all obstruction indices within the coverage area of ​​the time window are summed point by point to obtain the obstruction window sum value. The obstruction window sum value is then divided by the number of sampling points within the time window to obtain the obstruction window value. All urine volume time series data values ​​within the coverage area of ​​the time window are summed point by point to obtain the urine volume window sum value. The obstruction window values ​​of all time windows are sorted from largest to smallest to obtain the obstruction sorting sequence, while retaining the time window point corresponding to each obstruction window value.

[0148] Generate an anomaly window set: First, calculate the mean of the blocking window values ​​for all time windows. The mean is calculated by summing the blocking window values ​​of all time windows and then dividing by the total number of time windows. Next, calculate the standard deviation of the blocking window values ​​for all time windows. The standard deviation is calculated by subtracting the mean from the blocking window value of each time window to obtain the deviation value. The deviation values ​​are squared and summed to obtain the sum of squared deviations. The sum of squared deviations is divided by the total number of time windows to obtain the variance. The square root of the variance is then taken to obtain the standard deviation. Subsequently, the sum of the mean and the standard deviation is used as the screening benchmark. Time window points with blocking window values ​​not less than the screening benchmark are collected and formed into an anomaly window set. The time window points in the anomaly window set are arranged in ascending order and then merged according to the continuity of the points. The calculation process for merging intervals is as follows: windows with a difference of one between two adjacent time window points are considered as the same continuous segment, and the start and end time window points of the continuous segment are combined into an interval record. The interval records corresponding to all continuous segments are summarized and output as an anomaly interval.

[0149] Adjacent difference is performed on the rhythm baseline only within the sampling point range covered by the abnormal interval to generate the baseline increment sequence. The calculation process is as follows: take the baseline values ​​of two adjacent sampling points in the rhythm baseline in the order of sampling points, subtract the baseline value of the previous sampling point from the baseline value of the latter sampling point to obtain an increment value, and arrange and summarize the increment values ​​in the order of sampling points to form the baseline increment sequence.

[0150] Within the outlier interval, the baseline increment sequence is jointly scaled point-by-point based on the deviation index and quality weight to generate a compensated increment sequence. The specific calculation formula is as follows:

[0151] ;

[0152] In the formula, This indicates that the compensated increment sequence is at the sampling point The incremental compensation value at the location, This represents the set of sampling points corresponding to the abnormal interval. Indicates the baseline incremental sequence at the sampling point The value at that location, Indicates the quality weight at the sampling point The value at that location, This indicates the deviation index at the sampling point. The values ​​at the specified locations have all been normalized during the calculations.

[0153] Starting with urine volume time-series data, the compensated increment sequence is accumulated and spliced ​​point by point to generate reconstructed urine volume data:

[0154] The urine volume value of the first sampling point in the urine volume time series data is taken as the first reconstructed urine volume value. Then, the first increment value in the compensation increment sequence is added to the first reconstructed urine volume value to obtain the second reconstructed urine volume value. Then, the second increment value in the compensation increment sequence is added to the second reconstructed urine volume value to obtain the third reconstructed urine volume value. This process is repeated point by point in the order of the compensation increment sequence, so that each new reconstructed urine volume value is obtained by adding the previous reconstructed urine volume value to the current increment value, forming a reconstructed urine volume segment. The reconstructed urine volume segment is then joined with the urine volume values ​​of the remaining sampling points in the urine volume time series data that do not participate in the compensation increment accumulation in the order of sampling to complete the splicing, thereby obtaining continuous reconstructed urine volume data.

[0155] This approach summarizes and sorts urine volume time-series data by windowing based on the obstruction index and generates abnormal intervals. By prioritizing time windows driven by obstruction explanatory power, urine volume abnormalities are located in continuous intervals that are more likely to be caused by tubing factors. This reduces the probability of misidentifying scattered fluctuations as abnormalities and stabilizes the boundaries of abnormal intervals. Within the abnormal intervals, adjacent differences are performed on the rhythm baseline to generate baseline increment sequences. This transforms individualized baseline trends into incremental constraints that can be used for local reconstruction, ensuring that the reconstruction follows the direction and amplitude of the patient's own rhythm changes and avoiding trend drift caused by using fixed increments.

[0156] Within the abnormal interval, the baseline incremental sequence is jointly scaled point by point according to the deviation index and quality weight to generate a compensated incremental sequence. At the same time, the abnormality intensity and data credibility are used to suppress overcompensation and weaken the influence of noise points, so that the compensation amplitude adapts to the persistence of abnormality and credibility. Starting with urine volume time series data, the compensated incremental sequence is accumulated and spliced ​​point by point to generate reconstructed urine volume data. Without changing the continuity of the original sequence, the observation gap of the flow stop section is repaired and the comparable urine volume trend is restored, thus providing a more stable and consistent input for subsequent evidence fusion.

[0157] As described above, after generating the obstruction index, the process also includes generating reconstructed urine volume data. The following section describes the peer-to-peer fusion of the deviation index, obstruction index, and turbidity transition evidence to generate an evidence matrix. Please refer to [link / reference]. Figure 4 , Figure 4 This is a schematic diagram of the process for generating an evidence matrix provided in an embodiment of this application. Generating the evidence matrix specifically includes:

[0158] The deviation index, blocking index, and turbidity transition evidence are converged and normalized to generate deviation window value sequences, blocking window value sequences, and transition window value sequences, respectively.

[0159] The reconstructed urine volume data and urine volume time series data were aligned point by point and differential analysis was performed to generate a differential window value sequence.

[0160] After positive extraction of the blocking window value sequence and the transition window value sequence by the same window difference, they are compared with the deviation window value sequence to generate the conflict window value sequence. The conflict window value sequence and the difference window value sequence are then combined to generate the evidence matrix.

[0161] Among them, the deviation window value sequence refers to the numerical sequence arranged in time window order after performing window aggregation and scale unification on the generated deviation index;

[0162] The obstruction window value sequence refers to a time series data sequence formed by quantifying the degree to which abnormal urine output can be explained by catheter obstruction or torsion;

[0163] The transition window sequence refers to the quantitative result used to characterize whether turbidity changes from a stationary state to an anomalous state over a continuous time window scale;

[0164] The difference window value sequence refers to the difference expression sequence formed by quantifying and aggregating the numerical differences between reconstructed urine volume and urine volume time series data within the same sampling sequence number range, and arranging them in time window order;

[0165] Conflict window sequence refers to a data sequence used to characterize whether there is a directional contradiction and its intensity between the urine volume-side obstruction interpretation and the turbidity-side infection transition interpretation within the same time window.

[0166] The above content will be described in detail below:

[0167] The deviation index, blocking index, and turbidity transition evidence were converged and normalized using a common window method to generate deviation window value sequences, blocking window value sequences, and transition window value sequences.

[0168] To generate a deviation window value sequence, the deviation index is normalized by performing window convergence: all deviation index values ​​within the coverage area of ​​the time window are extracted sequentially according to the time window order. The deviation index values ​​within the time window are added point by point to obtain the total deviation within the window. The number of sampling points covered by the time window is then counted to obtain the number of points within the window. The deviation window value of the time window is obtained by dividing the total deviation within the window by the number of points within the window. All deviation window values ​​are arranged in the time window order to form an initial deviation window value sequence. A normalization operation is performed on the initial deviation window value sequence to obtain the deviation window value sequence. The normalization operation is calculated as follows: the maximum and minimum values ​​of all deviation window values ​​in the initial deviation window value sequence are taken. For each deviation window value, the minimum value is subtracted from the deviation window value to obtain the deviation difference. The maximum value is then subtracted from the minimum value and added to one to obtain the normalization denominator. Finally, the deviation difference is divided by the normalization denominator to obtain the normalized deviation window value. All normalized deviation window values ​​are arranged in the time window order as the deviation window value sequence.

[0169] The blocking index and turbidity transition evidence were subjected to window convergence and normalization according to the above process to generate blocking window value sequences and transition window value sequences, respectively.

[0170] Point-by-point alignment and difference analysis were performed on the reconstructed urine volume data and the urine volume time series data to generate a difference window value sequence:

[0171] Using the sampling points corresponding to both reconstructed urine volume data and urine volume time-series data as the basis for point-by-point alignment, the reconstructed urine volume values ​​and urine volume values ​​at the same sampling point are paired to form a point-by-point control set. Differential analysis is performed on the point-by-point control set to generate an intermediate data differential value sequence: For each sampling point, the absolute value of the difference between the reconstructed urine volume value and the urine volume value at that sampling point is taken as the differential value of that sampling point. All differential values ​​are arranged in the order of the sampling points to form a differential value sequence. Within each time window, the differential value sequence is aggregated to generate a differential window value sequence: For each time window, all differential values ​​covered by the time window are summed point by point to obtain the total differential value within the window. The total differential value within the window is then divided by the number of sampling points covered by the time window to obtain the differential window value of the time window. All differential window values ​​are arranged and output in the order of the time windows to form a differential window value sequence.

[0172] After positive extraction of the blocking window value sequence and the transition window value sequence using the same window difference, they are compared with the deviation window value sequence to generate a conflict window value sequence. The conflict window value sequence and the difference window value sequence are then combined to generate an evidence matrix.

[0173] The blocking window value sequence, transition window value sequence, deviation window value sequence, and difference window value sequence are aligned window by window using the same time window number. Then, for each time window, the same window difference positive extraction is performed: first, the same window difference is obtained by subtracting the transition window value of the time window from the blocking window value of the time window. Then, the same window difference is maximized by zero to obtain the positive difference, thereby setting the negative part of the same window difference to zero and retaining the positive part. The positive difference is compared with the deviation window value of the same time window to generate the conflict window value. The reference comparison is completed by a deterministic arithmetic combination: first, the positive difference is multiplied by the deviation window value to obtain the reference product. Then, the deviation window value is added to one to obtain the reference denominator. Finally, the reference product is divided by the reference denominator to obtain the conflict window value of the time window. The conflict window values ​​of all time windows are summarized in order of time window number to form the conflict window value sequence.

[0174] The conflict window sequence and the difference window sequence are combined to generate an evidence matrix.

[0175] This scheme generates deviation window sequences, obstruction window sequences, and turbidity transition window sequences by performing window convergence and normalization on evidence of deviation index, obstruction index, and turbidity transition, respectively. This unifies evidence from different time granularities and amplitude scales under the same time window and the same dimension, reducing the interference of individual differences and transient noise on judgment and improving cross-evidence comparability. The reconstructed urine volume data and urine volume time-series data are aligned point-by-point and difference analysis is performed to generate difference window sequences, quantifying observational biases caused by obstruction or distortion as evidence of difference within the same window. This allows urine volume abnormalities to be explained and suppressed. To address misjudgments caused by a sudden drop in urine output, a conflict window sequence is generated by positively extracting the difference between the blockage window sequence and the transition window sequence and comparing it with the deviation window sequence. This explicitly characterizes the degree of contradiction between strong blockage and weak transition and strong transition and weak blockage, and uses the deviation level to calibrate the confidence of the conflict, thereby reducing false positives of blockage and improving the stability of infection transition identification. The conflict window sequence and the difference window sequence are combined to generate an evidence matrix, forming a verifiable structured evidence carrier, which provides a basis for conflict suppression in subsequent risk synthesis and improves the consistency of early warning interpretation.

[0176] The above describes the window difference fusion of evidence from deviation index, blockage index, and turbidity transition to generate an evidence matrix. The following describes the generation of early warning results by applying quality weighting and deviation accumulation constraints to the evidence matrix based on quality weights and deviation indexes, specifically including:

[0177] A continuity analysis is performed on the evidence matrix, combining quality weights and deviation indices, to generate a consistency score. The evidence matrix is ​​then weighted and combined with the consistency score, and conflict penalties are applied to generate a risk score.

[0178] The evidence matrix, consistency score, and risk score are combined to generate early warning results.

[0179] Among them, the consistency score refers to the comprehensive score data used to characterize whether the evidence related to sepsis risk remains consistent in direction, continuous in intensity, and jointly enhanced under credible constraints in multiple consecutive time windows.

[0180] Risk scoring refers to the comprehensive risk quantification result obtained by weighting and combining the evidence items representing the degree of support for infection and the evidence items representing the degree of conflict in the evidence matrix.

[0181] The above content will be described in detail below:

[0182] A continuity analysis is performed on the evidence matrix, combining quality weights and deviation indices, to generate a consistency score. Then, a weighted composite of the evidence matrix and the consistency score, along with conflict penalties, is applied to generate a risk score.

[0183] For each time window, calculate the window confidence value: sum the quality weights covered by the time window point by point to obtain the weight sum, and then divide the weight sum by the number of sampling points covered by the time window to obtain the mean weight. For the same time window, calculate the window deviation value: sum the deviation indexes covered by the time window point by point to obtain the deviation sum, and then divide the deviation sum by the number of sampling points covered by the time window to obtain the deviation mean. For the same time window, calculate the continuity weight: multiply the mean weight by the deviation mean. For the same time window, calculate the support sum: extract the deviation window value and the transition window value corresponding to the time window from the evidence matrix and add them to obtain the support sum. For all time windows, calculate the continuity weight and support sum respectively to generate the consistent support value: multiply the continuity weight of each time window by the support sum of the time window to obtain the weighted support value, and then add the weighted support values ​​of all time windows to obtain the consistent support value. Sum the continuity weights of all time windows respectively to obtain the weight sum, and divide the consistent support value by the weight sum to obtain the consistent score.

[0184] Consistency score based on evidence matrix Weighted synthesis and conflict penalty are applied to generate a risk score. The specific calculation formula is as follows:

[0185] ;

[0186] In the formula, This represents the total number of time windows corresponding to the evidence matrix. Indicates the first Window weight of each time window This represents the set of supporting evidence items. Represents the set of conflicting evidence items. Indicates supporting evidence. The item weights, Indicates the first The first time window The numerical value of each piece of evidence. Indicating conflicting evidence items The item weights, Indicates the first The first time window The values ​​of each piece of evidence are all normalized during the calculation.

[0187] Range mapping and joint gating are performed on the consistency score and risk score to generate an early warning level:

[0188] The consensus score is subjected to interval truncation mapping, which involves comparing the consensus score with zero and taking the larger value, and then comparing it with one and taking the smaller value, thereby obtaining the truncated consensus value that falls within the closed interval of zero to one. The same interval truncation mapping is applied to the risk score to obtain the truncated risk value that falls within the closed interval of zero to one.

[0189] To generate a consistency level, a scaling quantization mapping is performed on the truncated consistency value. Specifically, the truncated consistency value is multiplied by three and rounded down to obtain an initial consistency level. The initial consistency level is then compared with two and the smaller value is taken to obtain a consistency level with a value of zero, one, or two.

[0190] The same scaling quantization mapping is performed on the truncated risk value to generate a risk level. Specifically, the truncated risk value is multiplied by three and rounded down to obtain an initial risk level. The initial risk level is then compared with two and the smaller value is taken to obtain a risk level with a value of zero, one, or two.

[0191] The minimum value between the consistency level and the risk level is used as the gating result, and the gating result is output as the warning level.

[0192] The evidence matrix is ​​processed window by window to perform support convergence and conflict suppression, generating a window contribution sequence. Peak location and continuous expansion of the window contribution sequence are then performed to generate trigger time windows. Within these trigger time windows, the evidence matrix is ​​subjected to primary evidence filtering and conflict-based extraction to generate evidence combinations.

[0193] For each time window, the deviation window value, transition window value, and difference window value are added together to obtain the window support value for that time window, and the conflict window value corresponding to that time window is used as the window conflict value.

[0194] The window contribution of a time window is obtained by subtracting the window conflict value from the window support value and then normalizing it. The window contribution values ​​of each time window are then arranged and summarized in the order of the time windows to form a window contribution sequence.

[0195] Peak localization is performed on the window contribution sequence to generate peak localization results: all window contributions in the window contribution sequence are sorted by size and the time window corresponding to the window contribution with the largest value is selected as the peak time window, and the window contribution with the largest value is recorded as the peak contribution.

[0196] A continuous expansion of the window contribution sequence is performed to generate a trigger time window: a valley point marker sequence is generated based on the comparison of the window contributions of adjacent time windows. The generation process of the valley point marker sequence is as follows: for each intermediate time window in the three-window structure, its window contribution is taken and compared with the window contributions of its adjacent preceding and following time windows. If the window contribution of the intermediate time window is not greater than the window contributions of its two adjacent time windows, the intermediate time window is marked as a valley point. The first and last time windows of the sequence are compared with their unique adjacent time windows. If the window contribution of the last time window is not greater than the window contribution of its adjacent time windows, the last time window is marked as a valley point. Based on the valley point marker sequence, the valley point time window closest to the peak time window is found in reverse order on the left side of the peak time window and determined as the left boundary time window. The valley point time window closest to the peak time window is found in order on the right side of the peak time window and determined as the right boundary time window. The left boundary time window, the peak time window and the right boundary time window are connected in order to form a continuous time window set and output as the trigger time window.

[0197] Within the trigger time window, the evidence matrix is ​​subjected to main evidence screening and conflict parallel extraction to generate evidence combinations: Within each time window covered by the trigger time window, the deviation window value, transition window value, difference window value, and blocking window value are read respectively, and the evidence score of the same evidence row is obtained by summing all window values ​​within the trigger time window, thereby forming an evidence score table containing deviation evidence score, transition evidence score, difference evidence score, and blocking evidence score. The evidence score table is sorted by evidence score from smallest to largest, and the set of evidence rows corresponding to the second half of the middle position of the sort is determined as the main evidence set, thereby completing the main evidence screening. At the same time, within the trigger time window, the conflict window value is read and a conflict window value table is formed according to the time window order. The conflict window value table is sorted by conflict row window value from smallest to largest, and the set of time windows corresponding to the second half of the middle position of the sort is determined as the conflict parallel time window set, thereby completing the conflict parallel extraction.

[0198] For each evidence row in the main evidence set, a location description is generated. The process of generating the location description is to take the time window corresponding to the maximum value of the window value of the evidence row within the trigger time window as the main location time window, and write the main location time window and the evidence score of the evidence row into the evidence entry. All evidence entries corresponding to the main evidence set are merged and output with the conflict entries corresponding to the conflict parallel time window set to obtain the evidence combination.

[0199] The evidence and warning levels are combined in a structured manner to generate warning results.

[0200] This solution generates a consistency score by performing continuous analysis of the evidence matrix combined with quality weights and deviation indices. This weakens the abnormal contribution of low-confidence time windows and strengthens time windows with continuous deviations, thereby reducing false resonances caused by noise such as urine bag shaking and transient shading. This improves the stable detection and advance warning of hidden progression. By weighted synthesis and conflict penalty of the evidence matrix combined with the consistency score, a risk score is generated. This allows the risk to be aggregated and amplified when there is consistent support from multiple windows, and the risk to be suppressed and reduced when obstruction conflict is dominant. This reduces false alarms of misdiagnosing obstructive urine volume abnormalities as infection and improves the discrimination confidence when infection evidence is consistent. The evidence matrix, consistency score, and risk score are combined to generate early warning results. The trigger time window and key evidence items can be output simultaneously, making the early warning basis verifiable and interpretable, reducing the uncertainty of treatment and improving the usability of the early warning.

[0201] The above describes the generation of early warning results by applying quality weighting and deviation accumulation constraints to the evidence matrix based on quality weights and deviation indices. The following describes the generation of early warning results by combining the evidence matrix, consistency score, and risk score, specifically including:

[0202] Interval mapping and joint gating are performed on consistency scores and risk scores to generate early warning levels;

[0203] Support convergence and conflict suppression are performed on the evidence matrix window by window to generate a window contribution sequence. Peak location and continuous expansion of the window contribution sequence are performed to generate a trigger time window. Then, within the trigger time window, the evidence matrix is ​​screened for main evidence and conflict is extracted in parallel to generate evidence combination.

[0204] The evidence and warning levels are combined in a structured manner to generate warning results.

[0205] Among them, the warning level refers to the discrete level data used to characterize the overall credibility and urgency of the abnormal state reflected by the evidence matrix within the current triggering time window.

[0206] The window contribution sequence refers to an ordered numerical sequence that is calculated for each time window based on the supporting and conflicting evidence corresponding to each time window in the evidence matrix, and combined with the constraint results of consistency score and risk score on the validity of time window evidence. This sequence is used to characterize the degree of contribution of the time window to the final early warning result.

[0207] A trigger time window is a set of continuous time windows used to characterize the process of risk formation and accumulation.

[0208] An evidence set refers to a collection of multi-source evidence units extracted from the evidence matrix and structurally aggregated within the triggering time window.

[0209] This part has already been described in detail above, so I will not repeat it here.

[0210] This solution generates early warning levels by performing interval mapping and joint gating on consistency scores and risk scores. It simultaneously constrains the strength of persistence and overall risk to the same level output, ensuring the level changes with both scores rather than jumping to a single threshold. This reduces false alarms caused by boundary fluctuations and stabilizes the early warning triggering sequence. Support convergence and conflict suppression are performed window-by-window on the evidence matrix, generating a window contribution sequence. Peak location and continuous expansion of the window contribution sequence generate trigger time windows. Within the trigger time window, the evidence matrix undergoes primary evidence screening and conflict parallel extraction to generate evidence combinations. The enhanced segments of multi-window evidence are precisely located as trigger intervals while simultaneously retaining conflict items. This enables verifiable definition of the early warning period and suppresses occasional triggers caused by single-window spikes. The evidence combinations and early warning levels are structured and combined to generate early warning results. The level and evidence chain are output as a unified whole, ensuring the early warning results simultaneously possess interpretable primary evidence paths and verifiable conflict parallel information, facilitating rapid review and consistency judgment.

[0211] Example 2:

[0212] Please see Figure 5 An early warning system for sepsis based on multi-parameter temporal pattern recognition includes:

[0213] The data acquisition module is used to acquire time-series data of urine volume and turbidity of the target object;

[0214] The effective segment segmentation module is used to perform dual-series pairing and change measurement on urine volume time series data and turbidity time series data, generate quality weights, and perform reliable aggregation of quality weights to generate effective segment labels.

[0215] The deviation analysis module is used to extract stable repetitive patterns from urine volume time series data according to continuous time windows and aggregate them based on quality weights within the effective segment marking, generate a rhythm baseline, and express the deviation index by comparing the urine volume time series data with the rhythm baseline in the same window and combining it with quality weights.

[0216] The transition analysis module is used to perform morphological scanning on urine volume time series data within the effective segment marker, cross-validate it with the rhythm baseline and deviation index, generate the obstruction index, and perform transition extraction processing on turbidity time series data with counter-evidence constraints based on the obstruction index to generate turbidity transition evidence.

[0217] The early warning generation module is used to perform window difference fusion on the evidence of deviation index, blockage index and turbidity transition to generate an evidence matrix. Based on the quality weight and deviation index, the evidence matrix is ​​subjected to quality weighting and deviation accumulation constraints to generate early warning results.

[0218] This embodiment has the same technical effects as Embodiment 1.

[0219] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. The data mentioned in this application have undergone normalization and other preprocessing to unify dimensions during formula calculations.

[0220] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for early warning of sepsis based on multi-parameter temporal pattern recognition, applied to the early warning of urinary sepsis under long-term catheterization monitoring in urology, characterized in that... Includes the following steps: Acquire time-series data of urine volume and turbidity of the target object; The urine volume time series data and the turbidity time series data are paired and the change is measured to generate quality weights. The quality weights are then reliably aggregated to generate valid segment labels. Within the effective segment marker, stable repetitive patterns are extracted from the urine volume time series data according to the quality weight and aggregated according to the continuous time window to generate a rhythm baseline. The urine volume time series data and the rhythm baseline are compared with each other in the same window and then combined with the quality weight to express the deviation and generate a deviation index. Within the effective segment marker, the urine volume time series data is morphologically scanned and cross-validated with the rhythm baseline and the deviation index to generate an obstruction index. Based on the obstruction index, the turbidity time series data is subjected to transition extraction processing for contradictory evidence constraints to generate turbidity transition evidence. The deviation index, the blockage index, and the turbidity transition evidence are fused together to generate an evidence matrix. The evidence matrix is ​​then subjected to quality weighting and deviation accumulation constraints based on the quality weight and the deviation index to generate an early warning result.

2. The sepsis early warning method based on multi-parameter temporal pattern recognition according to claim 1, characterized in that: The process of pairing and measuring changes in the urine volume time-series data and the turbidity time-series data to generate quality weights specifically includes: The urine volume time series data and the turbidity time series data are arranged coaxially and a point-by-point correspondence is established to generate an alignment index; Based on the alignment index, the consistency and inconsistency of the urine volume time series data and the turbidity time series data are measured in pairs, and quality weights are generated.

3. The sepsis early warning method based on multi-parameter temporal pattern recognition according to claim 2, characterized in that: Based on the quality weight, stable repetitive patterns are extracted from the urine volume time-series data according to continuous time windows and aggregated to generate a rhythm baseline, specifically including: Within the effective segment marker, the urine volume time series data is sequentially extracted according to the alignment index, and the sequential extraction results are weighted according to the quality weight by an equal time window to generate a urine volume change profile. Perform cross-window similarity aggregation on the urine volume change profile, and select the urine volume change profile with the largest cross-window similarity aggregation result as the rhythm baseline.

4. The sepsis early warning method based on multi-parameter temporal pattern recognition according to claim 1, characterized in that: The urine volume time-series data and the rhythm baseline are compared with peers and then combined with the quality weights to express the deviation, generating a deviation index, which specifically includes: Based on the rhythm baseline, the urine volume time series data is aligned and paired within a window to generate a deviation intensity expression, and the deviation intensity expression is subjected to window weighting constraint based on the quality weight to obtain a deviation sequence; The deviation sequence is subjected to cross-window coherent convergence and scale unification to generate a deviation index.

5. The sepsis early warning method based on multi-parameter temporal pattern recognition according to claim 3, characterized in that: After performing morphological scanning on the urine volume time-series data, cross-validation is performed with the rhythm baseline and the deviation index to generate the obstruction index, specifically including: The sequential extraction results corresponding to the urine volume time series data are subjected to adjacent difference calculation, and the corresponding two quality weights are summed. Determine whether the difference between adjacent subtraction results is zero and whether the corresponding summation result is greater than zero. If so, merge the corresponding adjacent subtraction results to generate a stationary candidate segment. The stationary candidate segments are compared with the rhythmic baseline to generate baseline violation expressions, and the deviation index is aggregated within a window to generate deviation support expressions. After the stable candidate segment, the adjacent difference results are continuously incrementally scanned according to the alignment index to generate the rising candidate segment. The baseline violation expression, the deviation support expression and the rising candidate segment are aligned and mapped to generate the blocking index.

6. The sepsis early warning method based on multi-parameter temporal pattern recognition according to claim 1, characterized in that: After generating the obstruction index, the process also includes generating reconstructed urine volume data, specifically including: The urine volume time series data is windowed, summarized, and sorted according to the obstruction index to generate abnormal intervals. Adjacent differences are then performed on the rhythm baseline within the abnormal intervals to generate baseline increment sequences. Within the abnormal interval, the baseline incremental sequence is jointly scaled point-by-point according to the deviation index and the quality weight to generate a compensated incremental sequence; Starting with the urine volume time series data, the compensation increment sequence is accumulated and spliced ​​point by point to generate reconstructed urine volume data.

7. The sepsis early warning method based on multi-parameter temporal pattern recognition according to claim 6, characterized in that: The specific steps for generating an evidence matrix by performing window difference fusion on the deviation index, the blocking index, and the turbidity transition evidence include: The deviation index, the blocking index, and the turbidity transition evidence are respectively converged and normalized to generate deviation window value sequence, blocking window value sequence, and transition window value sequence; The reconstructed urine volume data and the urine volume time series data are aligned point by point and subjected to difference analysis to generate a difference window value sequence; After positive extraction of the blocking window value sequence and the transition window value sequence by the same window difference, they are compared with the deviation window value sequence to generate a conflict window value sequence. The conflict window value sequence and the difference window value sequence are then combined to generate an evidence matrix.

8. The sepsis early warning method based on multi-parameter temporal pattern recognition according to claim 7, characterized in that: Based on the quality weights and deviation indices, the evidence matrix is ​​subjected to quality weighting and deviation accumulation constraints to generate early warning results, specifically including: A continuity analysis is performed on the evidence matrix in conjunction with the quality weights and the deviation index to generate a consistency score. The evidence matrix is ​​then combined with the consistency score using a weighted composite and conflict penalty to generate a risk score. The evidence matrix, the consistency score, and the risk score are combined to generate an early warning result.

9. The sepsis early warning method based on multi-parameter temporal pattern recognition according to claim 8, characterized in that: The process of combining the evidence matrix, the consistency score, and the risk score to generate an early warning result specifically includes: The consistency score and the risk score are subjected to interval mapping and joint gating to generate an early warning level; Support convergence and conflict suppression are performed on the evidence matrix window by window to generate a window contribution sequence. Peak location and continuous expansion of the window contribution sequence are performed to generate a trigger time window. Then, within the trigger time window, the evidence matrix is ​​subjected to main evidence screening and conflict parallel extraction to generate evidence combination. The evidence combination and the warning level are combined in a structured manner to generate a warning result.

10. A sepsis early warning system based on multi-parameter temporal pattern recognition, characterized in that, include: The data acquisition module is used to acquire time-series data of urine volume and turbidity of the target object; The effective segment segmentation module is used to perform dual-sequence pairing and change measurement on the urine volume time series data and the turbidity time series data, generate quality weights, and perform reliable aggregation on the quality weights to generate effective segment labels. The deviation analysis module is used to extract stable repetitive patterns from the urine volume time series data according to the quality weight within the effective segment marker, and aggregate them to generate a rhythm baseline. The urine volume time series data and the rhythm baseline are compared with each other in the same window and then combined with the quality weight to express the deviation and generate a deviation index. The transition analysis module is used to perform morphological scanning on the urine volume time series data within the effective segment marker, and then cross-validate it with the rhythm baseline and the deviation index to generate an obstruction index. Based on the obstruction index, the module performs transition extraction processing on the turbidity time series data with contradictory constraints to generate turbidity transition evidence. The early warning generation module is used to perform window difference fusion on the deviation index, the blocking index and the turbidity transition evidence to generate an evidence matrix, and to perform quality weighting and deviation accumulation constraints on the evidence matrix according to the quality weight and the deviation index to generate an early warning result.

Images