An impact quantification method and system for acute kidney injury assessment

Abstract

The present invention relates to the field of medical data processing technology, and in particular to an impact quantification method and system for acute kidney injury assessment. The method obtains a data vector composed of at least two detection data indicators of any patient, obtains a data vector sequence within a preset time period, performs nonlinear data conversion on each data vector sequence respectively, and obtains a corresponding target data sequence; according to the acute kidney injury score corresponding to each data vector in the two target data sequences, obtains a dynamic weight matrix corresponding to the two target data sequences; uses the dynamic weight matrix to obtain the DTW optimized distance between every two target data sequences of all patients, and based on all DTW optimized distances, clusters all target data sequences to obtain at least one clustering cluster; and performs real-time assessment of the acute kidney injury of newly added patients according to each clustering cluster, thereby improving the efficiency of identifying groups with similar disease progression patterns.

Description

Technical Field

[0001] The present invention relates to the technical field of medical data processing, and in particular to an impact quantification method and system for acute kidney injury assessment. Background Art

[0002] Acute kidney injury (AKI) is a common critical disease in clinical practice, and its accurate assessment is crucial for the patient's prognosis. In recent years, with the widespread application of machine learning technology in the medical field, K-means clustering algorithm has gradually been introduced into the evaluation and classification research of AKI. As an unsupervised learning method, K-means clustering can divide patients into different subgroups based on their multidimensional clinical data, which helps to identify different AKI phenotypes or risk groups.

[0003] However, the development of AKI presents significant individualization and dynamic characteristics, which is represented by the fact that various physiological indicators of patients are often inconsistent in time. Secondly, the development of AKI often presents nonlinear characteristics. For example, some patients will experience a sharp rise in creatinine levels, while other patients may show a slow but continuous rise, which makes the importance of various physiological indicators at different stages of AKI also change dynamically. Therefore, traditional K-means clustering has limitations when applied to AKI assessment. Specifically, the Euclidean distance used by the traditional K-means algorithm cannot effectively handle the irregular sampling and nonlinear characteristics of time series data, making it difficult to capture the dynamic changes in the importance of physiological indicators, thereby causing deviations in the results of AKI assessment based on clustering results.

[0004] Therefore, how to improve the accuracy of AKI assessment using K-means clustering has become an urgent problem to be solved. Summary of the invention

[0005] In view of this, an embodiment of the present invention provides an impact quantification method and system for acute kidney injury assessment to solve the problem of how to improve the accuracy of AKI assessment using K-means clustering.

[0006] In a first aspect, an embodiment of the present invention provides an impact quantification method for acute kidney injury assessment, the method comprising the following steps:

[0007] Obtain a data vector composed of at least two detection data indicators of any patient to obtain a data vector sequence within a preset time period, and perform nonlinear data conversion on each of the data vector sequences to obtain a corresponding target data sequence;

[0008] According to the importance of each detection data indicator in the data vector, an acute kidney injury score corresponding to each data vector in each of the target data sequences is obtained, and in the process of dynamic time warping of any two target data sequences, a dynamic weight matrix corresponding to the two target data sequences is obtained according to the acute kidney injury score corresponding to each data vector in the two target data sequences;

[0009] Utilizing the dynamic weight matrix, optimizing the DTW distance between the two target data sequences to obtain a DTW optimized distance, obtaining the DTW optimized distance between every two target data sequences in all patients, and clustering all target data sequences based on all DTW optimized distances to obtain at least one cluster;

[0010] According to the acute kidney injury progression pattern corresponding to each of the clusters, the acute kidney injury of new patients is assessed in real time.

[0011] Preferably, performing nonlinear data conversion on each of the data vector sequences to obtain a corresponding target data sequence includes:

[0012] For any data vector in any data vector sequence, a hyperbolic tangent function is used to perform data conversion on each detection data indicator in the data vector to obtain a converted data vector, and all the converted data vectors are combined into a target data sequence corresponding to the data vector sequence.

[0013] Preferably, obtaining the acute kidney injury score corresponding to each data vector in each target data sequence according to the importance of each detection data indicator in the data vector comprises:

[0014] For any data vector in any target data sequence, obtaining at least one cross strength characterizing the importance of interaction between two detection data indicators according to the correlation between the detection data indicators in the data vector;

[0015] Through the L1 regularization term, the weight of each detection data indicator in the data vector and the weight of each cross strength are adaptively obtained. According to the weight of each detection data indicator in the data vector and the weight of each cross strength, all detection data indicators and all cross strengths in the data vector are weighted and summed, and the corresponding result is used as the acute kidney injury score corresponding to the data vector.

[0016] Preferably, obtaining at least one cross strength characterizing the importance of interaction between two detection data indicators according to the correlation between the detection data indicators in the data vector comprises:

[0017] For any two detection data indicators in the data vector, according to the data types of the two detection data indicators, extract the corresponding detection data indicators from all target data sequences to form an indicator sequence of the corresponding data type, and calculate the indicator mean and indicator standard deviation of each indicator sequence respectively;

[0018] Obtain the differences between the two detection data indicators and the corresponding indicator means, obtain a first product between the two differences, obtain a second product between the two indicator standard deviations, and calculate the ratio between the first product and the second product as the cross-intensity that characterizes the importance of the interaction between the two detection data indicators.

[0019] Preferably, the adaptively obtaining the weight of each detection data indicator in the data vector and the weight of each cross strength through the L1 regularization term includes:

[0020] Constructing a corresponding L1 regularization term using the weight of each detection data indicator in the data vector and the weight of each cross strength, obtaining the variance of the acute kidney injury score according to the acute kidney injury score corresponding to each data vector in all target data sequences, calculating the indicator variance of all detection data indicators in all target data sequences, calculating the ratio of the variance of the acute kidney injury score to the indicator variance, and constructing a weight optimization function using the negative of the ratio and the L1 regularization term;

[0021] According to the weight optimization function, dynamically iteratively adjust the weight of each detection data indicator in the data vector and the weight of each cross strength, and use the weight corresponding to the minimized weight optimization function as the weight of each detection data indicator in the data vector and the weight of each cross strength;

[0022] The calculation expression corresponding to the dynamic iterative adjustment is:

[0023] ;

[0024] in, represents the kth weight after the v+1th dynamic iteration adjustment, represents the kth weight after the vth dynamic iteration adjustment, that is, the kth weight before the v+1th dynamic iteration adjustment. represents the partial derivative of the weight optimization function J to the kth weight after the vth dynamic iteration adjustment, Represents the preset learning rate.

[0025] Preferably, the step of obtaining the dynamic weight matrix corresponding to the two target data sequences according to the acute kidney injury score corresponding to each data vector in the two target data sequences comprises:

[0026] For any two data vectors in the two target data sequences, obtaining the addition result of the acute kidney injury scores corresponding to the two data vectors, and mapping the product between the preset control weight and the addition result by using an exponential function with a natural constant as the base to obtain a mapping value between the two data vectors;

[0027] Obtain the proportion of the mapping value in the sum of the mapping values ​​corresponding to each two data vectors in the two target data sequences, obtain the dynamic weight between the two data vectors, and form a dynamic weight matrix with the dynamic weights corresponding to each two data vectors in the two target data sequences.

[0028] Preferably, the using the dynamic weight matrix to optimize the DTW distance between the two target data sequences to obtain the DTW optimized distance includes:

[0029] The calculation expression of DTW optimization distance is:

[0030]

[0031] in, Represents the target data sequence and the target data sequence The DTW optimized distance between Represents the target data sequence The i-th data vector and target data sequence in The Euclidean distance between the j-th data vectors, Represents the target data sequence The i-th data vector and target data sequence in The dynamic weights corresponding to the j-th data vectors in the dynamic weight matrix, min() represents the minimum function, Represents the target data sequence and the target data sequence The shortest path length from point (i, j) to point (i-1, j) in the distance matrix, Represents the target data sequence and the target data sequence The shortest path length from point (i, j) to point (i, j-1) in the distance matrix, Represents the target data sequence and the target data sequence The shortest path length from point (i, j) to point (i-1, j-1) in the distance matrix, represents the preset gap penalty coefficient, Indicates the preset matching reward coefficient.

[0032] In a second aspect, an embodiment of the present invention provides an impact quantification system for acute kidney injury assessment, comprising a memory and a processor, wherein the processor executes a computer program stored in the memory to implement the above-mentioned impact quantification method for acute kidney injury assessment.

[0033] Compared with the prior art, the embodiments of the present invention have the following beneficial effects:

[0034] The present invention obtains a data vector composed of at least two detection data indicators of any patient, obtains a data vector sequence within a preset time period, and performs nonlinear data conversion on each of the data vector sequences to obtain a corresponding target data sequence;

[0035] According to the importance of each detection data indicator in the data vector, the acute kidney injury score corresponding to each data vector in each of the target data sequences is obtained; in the process of dynamic time warping of any two target data sequences, according to the acute kidney injury score corresponding to each data vector in the two target data sequences, the dynamic weight matrix corresponding to the two target data sequences is obtained; using the dynamic weight matrix, the DTW distance between the two target data sequences is optimized to obtain the DTW optimized distance, the DTW optimized distance between every two target data sequences of all patients is obtained, and based on all DTW optimized distances, all target data sequences are clustered to obtain at least one clustering cluster; according to the acute kidney injury progression pattern corresponding to each of the clustering clusters, the acute kidney injury of the newly added patients is evaluated in real time. First, nonlinear data transformation and dynamic weight matrix are used to achieve effective comparison of different sampling modes and acute kidney injury progression characteristics. Secondly, an acute kidney injury score is introduced, which comprehensively considers multiple key data indicators and their cross-strength. Through the automatic weight optimization process, it accurately captures the characteristic changes of acute kidney injury at different stages, which is used to optimize the DTW distance in the K-means clustering process. Finally, the K-means clustering method is used to group all patients based on the optimized DTW distance to identify groups with similar disease progression patterns. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0037] Figure 1 This is a method flow chart of an impact quantification method for acute kidney injury assessment provided in Example 1 of the present invention. DETAILED DESCRIPTION

[0038] Embodiments of the present disclosure are described in detail below, and examples of the embodiments are shown in the accompanying drawings. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present disclosure, but should not be understood as limiting the present disclosure.

[0039] It should be noted that the terms "first", "second", etc. in the specification of the present disclosure and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present disclosure described herein can be implemented in an order other than those illustrated or described herein. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present disclosure. Instead, they are merely examples of devices and methods consistent with some aspects of the present disclosure.

[0040] In order to illustrate the technical solution of the present invention, specific embodiments are provided below for illustration.

[0041] See also Figure 1 , is a method flow chart of a method for quantifying the impact of acute kidney injury assessment provided by Embodiment 1 of the present invention, such as Figure 1 As shown, the method may include:

[0042] Step S101, obtaining a data vector consisting of at least two detection data indicators of any patient, obtaining a data vector sequence within a preset time period, and performing nonlinear data conversion on each data vector sequence to obtain a corresponding target data sequence.

[0043] According to the sampling frequency, each patient's blood creatinine, urine volume, blood potassium, and blood creatinine change rate are collected to obtain a data vector composed of multiple test data indicators at each sampling time, which is recorded as ,in, represents the data vector corresponding to the i-th sampling moment, represents the blood creatinine at the i-th sampling moment, represents the urine volume at the i-th sampling moment, represents the blood potassium at the i-th sampling moment, Represents the rate of change of blood creatinine at the i-th sampling moment, and then obtains the data vector sequence formed by the time series ,in represents the data vector corresponding to the first sampling moment, represents the data vector corresponding to the second sampling moment, Represents the data vector corresponding to the i-th sampling moment.

[0044] It should be noted that when hour, ,when hour, ,in, It refers to the baseline creatinine value. The baseline creatinine value refers to the daily blood creatinine level range, which is 53μmol / L-106μmol / L for men and 44μmol / L-90μmol / L for women. Therefore, the baseline creatinine value of each patient can be set within the normal level range. At the same time, the baseline creatinine value of all patients can be set to the same value, which is not required here.

[0045] Similarly, data vector sequences of multiple patients are obtained. In the traditional way, the DTW distance between each two data vector sequences is calculated, and all data vector sequences are clustered using the K-means clustering algorithm to obtain multiple clusters. One cluster corresponds to one acute kidney injury (AKI) progression pattern. For new patients, the AKI progression pattern of new patients can be evaluated in real time based on the clusters.

[0046] However, in the evaluation of acute kidney injury (AKI), the development of AKI presents significant individualization and dynamic characteristics. In clinical practice, it is found that the test data of patients are often inconsistent in time. According to the KDIGO (Kidney Disease: Improving Global Outcomes) guidelines, the diagnosis and staging of AKI requires continuous monitoring of serum creatinine and urine volume. However, the actual test frequency varies depending on the patient's condition and medical resources. Some patients undergo multiple tests a day, while others may only undergo a test every few days. This irregular data collection pattern significantly affects the direct comparison of the course of different patients. On this basis, AKI has a nonlinear progression problem. The reason is that studies have shown that the progression of AKI is not a linear process. For example: A study of patients in the intensive care unit found that about 40% of AKI patients experienced a sharp increase in creatinine levels. This rapid change is difficult to accurately predict with a linear model. In contrast, other patients generally have a slow but continuous increase in creatinine levels. This nonlinear change pattern exceeds the capture capacity of traditional linear models. This also leads to the dynamic changes in the importance of various physiological indicators in different stages of AKI. According to the KDIGO guidelines, changes in urine volume are a sensitive indicator in the early stages of AKI, because changes in urine volume can reflect acute changes in renal function earlier than serum creatinine. In the later stages, changes in serum creatinine levels become a key indicator for assessing the recovery of renal function, and the importance of this dynamic change characteristic increases the complexity of the assessment.

[0047] Based on the above characteristics, the embodiment of the present invention optimizes the DTW distance in the K-means clustering algorithm by introducing nonlinear data transformation and dynamic weight matrix, so that for patients who are examined once a day and patients who are examined once every two days, the best correspondence between the two data vector sequences can be found from DTW, and a meaningful comparison can be achieved. First, nonlinear data transformation is performed on each data vector sequence to obtain the corresponding target data sequence. The nonlinear data transformation process is:

[0048] For any data vector in any data vector sequence, a hyperbolic tangent function is used to perform data conversion on each detection data indicator in the data vector to obtain a converted data vector, and all the converted data vectors are combined into a target data sequence corresponding to the data vector sequence.

[0049] In one embodiment, taking the i-th data vector in the data vector sequence X as an example, a conversion function is used to perform nonlinear conversion on each detection data indicator of the i-th data vector, and the nonlinear conversion function is:

[0050]

[0051] in, represents the converted value corresponding to the t-th detection data indicator in the ith data vector, tanh() represents the hyperbolic tangent function, Represent different learning parameters, represents the tth detection data indicator in the ith data vector, and sigmoid() represents the activation function.

[0052] It should be noted that tanh() compresses the original test data indicators into the range of (-1, 1), reduces the impact of extreme values, and enhances sensitivity near the key thresholds for AKI diagnosis and staging (such as a 1.5-fold increase in creatinine or 0.3 mg / dL as defined in the KDIGO guidelines), so that small changes within this range can be amplified after data conversion, thereby producing a more significant impact in the subsequent DTW distance calculation; the addition of sigmoid() allows further fine-tuning of the conversion effect by adjusting and Different sensitivity intervals are set for different detection data indicators. Therefore, the purpose of designing the nonlinear transformation function is to adjust the scale and sensitivity of the original detection data indicators to better reflect the Clinical features and progression stages.

[0053] Preferably, in the embodiment of the present invention, the learning parameter The introduction of can adaptively adjust the conversion function according to actual data, adapt to the characteristics of AKI in different populations and clinical environments, and set , , , The ranges are , , , , the conversion of different features requires adaptive adjustment of settings.

[0054] Similarly, the converted values ​​corresponding to each detection data indicator in the i-th data vector are obtained to form the converted i-th data vector, which is recorded as: , where n is the number of detection data indicators contained in the data vector. Similarly, nonlinear data transformation is performed on each data vector in the data vector sequence X, and the transformed data vectors are combined into the target data sequence corresponding to the data vector sequence X .

[0055] At this point, the target data sequence corresponding to each data vector sequence can be obtained, which can better reflect the clinical significance of AKI in the subsequent calculation of DTW distance, rather than just the difference in raw values. It can make the transition from normal to early AKI have a significant impact in the DTW distance metric, while the further distinction between cases that have been diagnosed with severe AKI is relatively weakened, which helps to more accurately classify patients into different risk groups or prognosis groups.

[0056] Step S102, according to the importance of each detection data indicator in the data vector, obtain the acute kidney injury score corresponding to each data vector in each target data sequence, and in the process of dynamic time warping of any two target data sequences, according to the acute kidney injury score corresponding to each data vector in the two target data sequences, obtain the dynamic weight matrix corresponding to the two target data sequences.

[0057] The development and progression of AKI involves complex interactions of multiple indicators such as creatinine, urine volume, and electrolytes, and the relative importance of these indicators changes significantly at different stages of the disease. For example, in early AKI, small changes in urine volume are more predictive than changes in creatinine; in late AKI, electrolyte disorders become a more critical prognostic indicator. Since nonlinear data transformation mainly focuses on the time series of a single test data indicator, it is difficult to describe the dynamic change characteristics of the interaction of multiple test data indicators. Therefore, in order to solve the above problems, the AKI evaluation process is further refined on the basis of obtaining the target data sequence. The specific refinement method is: according to the importance of each test data indicator in the data vector, the acute kidney injury score corresponding to each data vector in each target data sequence is obtained.

[0058] In actual clinical scenarios, the manifestation of AKI varies from patient to patient. According to the KDIGO guidelines, an increase in creatinine levels to 1.5-1.9 times the baseline is defined as AKI stage 1. However, for patients of different ages and baseline renal function, the same degree of creatinine increase represents different degrees of renal function damage. In the process of obtaining the acute kidney injury score corresponding to each data vector in each target data sequence, this individual difference is adapted by considering multiple test data indicators and their interactions. That is, the synergistic effect between test data indicators is quantified by introducing cross-strength calculations. For example: when an increase in creatinine is accompanied by a significant decrease in urine volume, this situation usually indicates more severe AKI and a worse prognosis, and the importance of each indicator will change at different stages of AKI. In early AKI, changes in urine volume are particularly important, and in late AKI, electrolyte disorders become a more critical prognostic indicator. Therefore, the importance of each test data indicator is automatically adjusted at different stages of AKI through the process of automatically optimizing weights.

[0059] Wherein, according to the importance of each detection data indicator in the data vector, obtaining the acute kidney injury score corresponding to each data vector in each target data sequence includes:

[0060] (1) For any data vector in any target data sequence, at least one cross strength characterizing the importance of the interaction between two detection data indicators is obtained according to the correlation between the detection data indicators in the data vector.

[0061] Specifically, for any two detection data indicators in the data vector, according to the data types of the two detection data indicators, corresponding detection data indicators are extracted from all target data sequences to form indicator sequences of corresponding data types, and the indicator mean and indicator standard deviation of each indicator sequence are calculated respectively;

[0062] Obtain the differences between the two detection data indicators and the corresponding indicator means, obtain a first product between the two differences, obtain a second product between the two indicator standard deviations, and calculate the ratio between the first product and the second product as the cross-intensity that characterizes the importance of the interaction between the two detection data indicators.

[0063] In one embodiment, taking the i-th data vector in any target data sequence as an example, since the i-th data vector contains , then the calculation expression of the cross strength that characterizes the importance of the interaction between any two detection data indicators in the i-th data vector is:

[0064]

[0065] in, represents the cross strength of the interaction importance between detection data index a and detection data index b in the i-th data vector, represents the detection data index a in the i-th data vector, represents the detection data index b in the i-th data vector, represents the mean of all detection data indicators a in all target data sequences, represents the mean of all detection data indicators b in all target data sequences, Represents the standard deviation of all detection data indicators a in all target data sequences, represents the standard deviation of all detection data indicators b in all target data sequences, where the values ​​of a and b are contained in the i-th data vector .

[0066] It should be noted that the purpose of calculating the standardized covariance between different test data indicators is to quantify the dynamic relationship between different test data indicators, especially the abnormal coordinated changes. Specifically, when creatinine and urine volume show abnormal changes at the same time (for example, creatinine increases significantly while urine volume decreases sharply), will produce larger positive values, which will lead to larger changes in the subsequent analysis of acute kidney injury scores and highlight this clinically highly relevant collaborative abnormality in the DTW distance metric; for creatinine and serum potassium, if they change in the same direction (such as both increase), Will produce a positive value, if the change is in the opposite direction, then This produces negative values, which enables the distinction of pure prerenal (creatinine elevation only) and more complex with electrolyte disturbances Standardization in the formula (subtracting the mean and dividing by the standard deviation) ensures that measures of different magnitude are comparable on the same scale, with the goal that small clinically significant changes in creatinine are not masked by large fluctuations in urine output in subsequent distance measures.

[0067] (2) Adaptively obtain the weight of each detection data indicator and the weight of each cross-intensity in the data vector through the L1 regularization term, and perform weighted summation on all detection data indicators and all cross-intensities in the data vector according to the weight of each detection data indicator and the weight of each cross-intensity in the data vector, and use the corresponding result as the acute kidney injury score corresponding to the data vector.

[0068] In one embodiment, the calculation expression of the acute kidney injury score corresponding to the i-th data vector in the target data sequence is:

[0069]

[0070] in, represents the acute kidney injury score corresponding to the i-th data vector, n represents the number of test data indicators contained in the i-th data vector, represents the tth detection data index in the ith data vector, represents the weight of the t-th detection data indicator in the ith data vector, N represents the number of cross-strengths in the ith data vector, represents the Lth cross-strength in the ith data vector, represents the weight of the Lth cross-strength in the ith data vector.

[0071] It should be noted that obtaining The goal is to provide a more comprehensive AKI severity score and to determine the best weight combination through an automated optimization process to create a scoring system that can accurately distinguish different AKI stages and types. The initial value of each weight involved is 1, and then through the automatic weight optimization process, all weights will be adjusted to maximize the discriminatory power of the score. The significance of the subsequent DTW distance metric is that different types and stages of AKI can be better distinguished in the DTW distance calculation. For example, the distinction between acute kidney injury that has begun to develop and acute kidney injury that has already developed will become more obvious, because the weight optimization process will enhance the sensitivity to early changes, and AKI with sufficient urine output and oliguric AKI will be clearly separated in the distance space, and In the distance calculation, emphasis will be placed on the feature combination that can best distinguish different AKI severities, rather than just the change in a single test data indicator.

[0072] For example: for the i-th data vector in the target data sequence in the embodiment of the present invention , then the calculation expression of its acute kidney injury score is:

[0073]

[0074] in, represents the creatinine level in the ith data vector, represents the amount of urine in the ith data vector, represents the blood potassium level in the ith data vector, represents the creatinine change rate in the ith data vector, represents the cross intensity of creatinine and urine volume in the ith data vector, represents the cross intensity between creatinine and serum potassium in the ith data vector.

[0075] Cross Strength and The weight of and ) reflects the importance of the interaction between these test data indicators, which will highlight that certain specific interaction patterns (such as an increase in creatinine accompanied by a decrease in urine volume) are particularly important for assessing the severity of AKI, and the weights will be adjusted accordingly. Automatic optimization of weights will make The score can adapt to the complex patterns and nonlinear relationships that may exist in the data. , optimization will make The importance of nonlinearity across the range of creatinine levels, for example, would identify key thresholds in the KDIGO criteria (e.g. ), and give a higher weight within this range. This is not achieved through manual setting, but the result of automatic adjustment by the optimization algorithm based on data characteristics; for urine volume , In the low urine output range (such as ), it gets a higher weight, which will make The score is more sensitive to oliguria and anuria, and this sensitivity is data-driven rather than pre-specified.

[0076] Among them, the weight of each detection data indicator in the data vector and the weight of each cross strength are adaptively obtained through the L1 regularization term, including:

[0077] Constructing a corresponding L1 regularization term using the weight of each detection data indicator in the data vector and the weight of each cross strength, obtaining the variance of the acute kidney injury score according to the acute kidney injury score corresponding to each data vector in all target data sequences, calculating the indicator variance of all detection data indicators in all target data sequences, calculating the ratio of the variance of the acute kidney injury score to the indicator variance, and constructing a weight optimization function using the negative of the ratio and the L1 regularization term;

[0078] According to the weight optimization function, the weight of each detection data indicator in the data vector and the weight of each cross-intensity are dynamically iteratively adjusted, and the weight corresponding to the minimized weight optimization function is used as the weight of each detection data indicator in the data vector and the weight of each cross-intensity.

[0079] In one embodiment, the calculation expression of the weight optimization function is:

[0080]

[0081] in, represents the weight optimization function, represents the variance of the acute kidney injury scores corresponding to all data vectors in all target data sequences, Represents the indicator variance of all detection data indicators in all target data sequences, represents the L1 regularization term, represents the kth weight, represents the L1 regularization parameter, .

[0082] It should be noted that the weight optimization function aims to find the optimal weight combination to maximize score( ) by automatically adjusting the weights to improve the discrimination ability of acute kidney injury score ( ) can distinguish different Case. Among them, maximize The optimized It can capture the most variability in the original detection data indicators, which will lead to different types and stages of DTW distance metrics in the subsequent are clearly distinguished.

[0083] L1 regularization term ( ) is to promote the sparsity of weights, which means that in the process of weight optimization, the weights of some less important test data indicators will be reduced to near zero, so as to highlight the truly critical indicators, and then in the subsequent DTW distance calculation, the clustering algorithm will pay more attention to those test data indicators that best reflect the severity and prognosis of AKI.

[0084] The process of obtaining the optimal weight combination of the weight optimization function is an iterative process. Therefore, the calculation expression corresponding to the dynamic iterative adjustment is used to adjust the weight by gradient descent to find the optimal weight that can minimize the weight optimization function. The calculation expression corresponding to the dynamic iterative adjustment is:

[0085]

[0086] in, represents the kth weight after the v+1th dynamic iteration adjustment, represents the kth weight after the vth dynamic iteration adjustment, that is, the kth weight before the v+1th dynamic iteration adjustment. represents the partial derivative of the weight optimization function J to the kth weight after the vth dynamic iteration adjustment, represents the preset learning rate, .

[0087] It should be noted that through this iterative process, a set of weights can be obtained that makes the score more sensitive to small changes in creatinine and urine volume in the early AKI stage (such as KDIGO grade 1), which will amplify the differences between early AKI cases in the subsequent DTW distance metric and help identify high-risk patients early. Secondly, in more severe AKI stages (such as KDIGO grade 3), the score is more sensitive to the coordinated changes in electrolyte disorders and creatinine-urine volume, which will highlight those complex cases requiring renal replacement therapy in the subsequent DTW distance calculation.

[0088] Furthermore, in the process of performing dynamic time warping (DTW algorithm) on any two target data sequences, a dynamic weight matrix corresponding to the two target data sequences is obtained according to the acute kidney injury score corresponding to each data vector in the two target data sequences, so as to dynamically adjust the importance between different data vectors in the two target data sequences during the dynamic time warping process.

[0089] Wherein, according to the acute kidney injury score corresponding to each data vector in the two target data sequences, obtaining the dynamic weight matrix corresponding to the two target data sequences includes:

[0090] For any two data vectors in the two target data sequences, obtaining the addition result of the acute kidney injury scores corresponding to the two data vectors, and mapping the product between the preset control weight and the addition result by using an exponential function with a natural constant as the base to obtain a mapping value between the two data vectors;

[0091] Obtain the proportion of the mapping value in the sum of the mapping values ​​corresponding to each two data vectors in the two target data sequences, obtain the dynamic weight between the two data vectors, and form a dynamic weight matrix with the dynamic weights corresponding to each two data vectors in the two target data sequences.

[0092] In one embodiment, the target data sequence The data vector in and the target data sequence The data vector in For example, calculate the data vector and data vector The calculation expression of the dynamic weight between is:

[0093]

[0094] in, Represents data vector and data vector The dynamic weight between them, exp() represents an exponential function with a natural constant as the base, represents the preset control weight, , Represents the target data sequence The data vector in Acute kidney injury score, Represents the target data sequence The data vector in Acute kidney injury score, n represents the number of data vectors in the target data sequence.

[0095] Similarly, get the target data sequence and the target data sequence The dynamic weights between every two data vectors form a dynamic weight matrix W. At this point, the dynamic weight matrix between every two target data sequences can be obtained.

[0096] Step S103, using a dynamic weight matrix, optimizing the DTW distance between two target data sequences to obtain a DTW optimized distance, obtaining the DTW optimized distance between every two target data sequences in all patients, and clustering all target data sequences based on all DTW optimized distances to obtain at least one cluster.

[0097] Through the method of step S101 and step S102, a nonlinear conversion function and an acute kidney injury score are introduced into every two data vector sequences, and a dynamic weight matrix between every two target data sequences is obtained. Then, the DTW distance between the two target data sequences is optimized by using the dynamic weight matrix to obtain the DTW optimized distance, and the DTW optimized distance between every two target data sequences in all patients is obtained, wherein the calculation expression of the DTW optimized distance is:

[0098]

[0099] in, Represents the target data sequence and the target data sequence The DTW optimized distance between Represents the target data sequence The i-th data vector and target data sequence in The Euclidean distance between the j-th data vectors, Represents the target data sequence The i-th data vector and target data sequence in The dynamic weights corresponding to the j-th data vectors in the dynamic weight matrix, min() represents the minimum function, Represents the target data sequence and the target data sequence The shortest path length from point (i, j) to point (i-1, j) in the distance matrix, Represents the target data sequence and the target data sequence The shortest path length from point (i, j) to point (i, j-1) in the distance matrix, Represents the target data sequence and the target data sequence The shortest path length from point (i, j) to point (i-1, j-1) in the distance matrix, represents the preset gap penalty coefficient, Indicates the preset matching reward coefficient.

[0100] Preferably, gap penalty The introduction of is used to deal with different sampling frequencies or missing data. The size of can control the tolerance of the algorithm to the timeline stretching, the initial value =0.5, value range [0, 2]; matching reward The introduction of encourages the algorithm to perform direct data vector matching as much as possible, by adjusting Can affect the behavior of the algorithm when dealing with fast and slow changes, set the initial value , the value range is [-2, 0]. By adjusting and To adjust the behavior of the algorithm according to specific clinical needs (such as whether to focus more on overall trends or local changes).

[0101] It should be noted that the role of the nonlinear conversion function is to capture nonlinear patterns in the progression of AKI. For example, it can identify a sudden increase in creatinine levels after a stable period, and distinguish between disease courses that appear similar on the surface but actually represent different risk levels. The introduction of the dynamic weight matrix is ​​to solve the problem of feature importance changing over time. The weights are dynamically adjusted according to the current severity of AKI. Its role is to emphasize the most relevant indicators at different stages of the disease. For example, in early AKI, the model gives a higher weight to changes in urine volume, and in the later stages, it pays more attention to changes in creatinine levels. Therefore, in the process of measuring the DTW distance, the optimized DTW distance (DTW optimized distance) can generate a distance metric that reflects the clinical characteristics of AKI and can adapt to data irregularities. This distance metric can more accurately quantify the similarity between patients and accurately capture the characteristic changes of AKI at different stages, thereby making the subsequent K-means clustering results based on the DTW optimized distance more able to reflect different individualized progression patterns of AKI.

[0102] For example, in actual application, if there are two AKI patients, the creatinine level of patient A rises slowly, and the creatinine level of patient B suddenly rises sharply after a few days of stability. Traditional methods tend to classify these two patients as similar because the final creatinine levels of the two patients are similar. The optimized distance metric can identify the sudden deterioration pattern of patient B and classify him into a higher risk group. This distinction is crucial for timely intervention and personalized treatment plan formulation. It is in line with the principle of early identification and intervention emphasized in the KDIGO guidelines.

[0103] It is worth noting that the DTW optimization distance is obtained by introducing a dynamic weight matrix, a gap penalty coefficient, and a matching reward coefficient in the DTW algorithm, and the DTW algorithm belongs to the prior art and will not be described in detail here.

[0104] Referring to the calculation expression of the above DTW optimized distance, the DTW optimized distance between the target data sequences corresponding to each two patients is obtained respectively. After all the DTW optimized distances are obtained, all the target data sequences are clustered based on all the DTW optimized distances using K-means clustering to obtain multiple clusters, where the number of clusters (K value) is obtained using the elbow rule, where the elbow rule and K-means clustering are both existing technologies and will not be described in detail here.

[0105] Step S104: performing a real-time assessment of the acute kidney injury of the newly added patients according to the acute kidney injury progression pattern corresponding to each cluster.

[0106] A cluster represents an acute kidney injury progression pattern. Therefore, after obtaining multiple clusters, the characteristics of each cluster can be analyzed, including but not limited to the average creatinine increase rate, urine volume change pattern and electrolyte disorder degree. Based on these characteristics, each cluster is assigned a clinically significant label, such as "rapid progression type" or "slow recovery type". After collecting the data vector sequence of the new patients, the data vector sequence is nonlinearly transformed to obtain the corresponding target data sequence, and the Euclidean distance between the target data sequence and the cluster center of each cluster is calculated to determine the cluster closest to the new patients. According to the determined cluster, the new patients are evaluated for acute kidney injury in real time to predict their AKI development trend and potential risks.

[0107] It is worth noting that the focus of the present invention is on how to optimize the DTW distance, thereby improving the accuracy of the clustering results of K-means clustering. How to perform real-time evaluation of new patients according to each cluster can refer to the existing technology and will not be repeated here.

[0108] In summary, the embodiment of the present invention obtains a data vector composed of at least two detection data indicators of any patient, obtains a data vector sequence within a preset time period, and performs nonlinear data conversion on each of the data vector sequences to obtain a corresponding target data sequence;

[0109] According to the importance of each detection data indicator in the data vector, the acute kidney injury score corresponding to each data vector in each of the target data sequences is obtained; in the process of dynamic time warping of any two target data sequences, according to the acute kidney injury score corresponding to each data vector in the two target data sequences, the dynamic weight matrix corresponding to the two target data sequences is obtained; using the dynamic weight matrix, the DTW distance between the two target data sequences is optimized to obtain the DTW optimized distance, the DTW optimized distance between every two target data sequences of all patients is obtained, and based on all DTW optimized distances, all target data sequences are clustered to obtain at least one clustering cluster; according to the acute kidney injury progression pattern corresponding to each of the clustering clusters, the acute kidney injury of the newly added patients is evaluated in real time. First, nonlinear data transformation and dynamic weight matrix are used to achieve effective comparison of different sampling modes and acute kidney injury progression characteristics. Secondly, an acute kidney injury score is introduced, which comprehensively considers multiple key data indicators and their cross-strength. Through the automatic weight optimization process, it accurately captures the characteristic changes of acute kidney injury at different stages, which is used to optimize the DTW distance in the K-means clustering process. Finally, the K-means clustering method is used to group all patients based on the optimized DTW distance to identify groups with similar disease progression patterns.

[0110] Based on the same inventive concept as the above method, an embodiment of the present invention also provides an impact quantification system for acute kidney injury assessment, including a memory and a processor, wherein the processor executes a computer program stored in the memory to implement the steps of any one of the above-mentioned impact quantification methods for acute kidney injury assessment.

[0111] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention, and should be included in the protection scope of the present invention.

Claims

1. A method for quantifying the impact of acute kidney injury assessment, characterized in that: The method comprises: Obtain a data vector composed of at least two detection data indicators of any patient to obtain a data vector sequence within a preset time period, and perform nonlinear data conversion on each of the data vector sequences to obtain a corresponding target data sequence; According to the importance of each detection data indicator in the data vector, an acute kidney injury score corresponding to each data vector in each of the target data sequences is obtained, and in the process of dynamic time warping of any two target data sequences, a dynamic weight matrix corresponding to the two target data sequences is obtained according to the acute kidney injury score corresponding to each data vector in the two target data sequences; Utilizing the dynamic weight matrix, optimizing the DTW distance between the two target data sequences to obtain a DTW optimized distance, obtaining the DTW optimized distance between every two target data sequences in all patients, and clustering all target data sequences based on all DTW optimized distances to obtain at least one cluster; According to the acute kidney injury progression pattern corresponding to each of the clusters, a real-time assessment of the acute kidney injury of newly added patients is performed; The step of obtaining, according to the importance of each detection data indicator in the data vector, an acute kidney injury score corresponding to each data vector in each target data sequence comprises: For any data vector in any target data sequence, obtaining at least one cross strength characterizing the importance of interaction between two detection data indicators according to the correlation between the detection data indicators in the data vector; Adaptively obtain the weight of each detection data indicator and the weight of each cross strength in the data vector through the L1 regularization term, and perform weighted summation on all detection data indicators and all cross strengths in the data vector according to the weight of each detection data indicator and the weight of each cross strength in the data vector, and use the corresponding result as the acute kidney injury score corresponding to the data vector; The adaptively obtaining the weight of each detection data indicator in the data vector and the weight of each cross strength through the L1 regularization term includes: Constructing a corresponding L1 regularization term using the weight of each detection data indicator in the data vector and the weight of each cross strength, obtaining the variance of the acute kidney injury score according to the acute kidney injury score corresponding to each data vector in all target data sequences, calculating the indicator variance of all detection data indicators in all target data sequences, calculating the ratio of the variance of the acute kidney injury score to the indicator variance, and constructing a weight optimization function using the negative of the ratio and the L1 regularization term; According to the weight optimization function, dynamically iteratively adjust the weight of each detection data indicator in the data vector and the weight of each cross strength, and use the weight corresponding to the minimized weight optimization function as the weight of each detection data indicator in the data vector and the weight of each cross strength; The calculation expression corresponding to the dynamic iterative adjustment is: ； in, represents the kth weight after the v+1th dynamic iteration adjustment, represents the kth weight after the vth dynamic iteration adjustment, that is, the kth weight before the v+1th dynamic iteration adjustment. represents the partial derivative of the weight optimization function J to the kth weight after the vth dynamic iteration adjustment, Represents the preset learning rate.

2. The method for quantifying the impact of acute kidney injury assessment according to claim 1, characterized in that: The performing nonlinear data conversion on each of the data vector sequences to obtain a corresponding target data sequence comprises: For any data vector in any data vector sequence, a hyperbolic tangent function is used to perform data conversion on each detection data indicator in the data vector to obtain a converted data vector, and all the converted data vectors are combined into a target data sequence corresponding to the data vector sequence.

3. The method for quantifying the impact of acute kidney injury assessment according to claim 1, characterized in that: The step of obtaining at least one cross strength characterizing the importance of the interaction between two detection data indicators according to the correlation between the detection data indicators in the data vector comprises: For any two detection data indicators in the data vector, according to the data types of the two detection data indicators, extract the corresponding detection data indicators from all target data sequences to form an indicator sequence of the corresponding data type, and calculate the indicator mean and indicator standard deviation of each indicator sequence respectively; Obtain the differences between the two detection data indicators and the corresponding indicator means, obtain a first product between the two differences, obtain a second product between the two indicator standard deviations, and calculate the ratio between the first product and the second product as the cross-intensity that characterizes the importance of the interaction between the two detection data indicators.

4. The impact quantification method for acute kidney injury assessment according to claim 1, characterized in that: The step of obtaining a dynamic weight matrix corresponding to the two target data sequences according to the acute kidney injury score corresponding to each data vector in the two target data sequences comprises: For any two data vectors in the two target data sequences, obtaining the addition result of the acute kidney injury scores corresponding to the two data vectors, and mapping the product between the preset control weight and the addition result by using an exponential function with a natural constant as the base to obtain a mapping value between the two data vectors; Obtain the proportion of the mapping value in the sum of the mapping values ​​corresponding to each two data vectors in the two target data sequences, obtain the dynamic weight between the two data vectors, and form a dynamic weight matrix with the dynamic weights corresponding to each two data vectors in the two target data sequences.

5. The impact quantification method for acute kidney injury assessment according to claim 1, characterized in that: The step of optimizing the DTW distance between the two target data sequences by using the dynamic weight matrix to obtain the DTW optimized distance includes: The calculation expression of DTW optimization distance is: in, Represents the target data sequence and the target data sequence The DTW optimized distance between Represents the target data sequence The i-th data vector and target data sequence in The Euclidean distance between the j-th data vectors, Represents the target data sequence The i-th data vector and target data sequence in The dynamic weights corresponding to the j-th data vectors in the dynamic weight matrix, min() represents the minimum function, Represents the target data sequence and the target data sequence The shortest path length from point (i, j) to point (i-1, j) in the distance matrix, Represents the target data sequence and the target data sequence The shortest path length from point (i, j) to point (i, j-1) in the distance matrix, Represents the target data sequence and the target data sequence The shortest path length from point (i, j) to point (i-1, j-1) in the distance matrix, represents the preset gap penalty coefficient, Indicates the preset matching reward coefficient.

6. An impact quantification system for acute kidney injury assessment, comprising a memory and a processor, characterized in that: The processor executes the computer program stored in the memory to implement the impact quantification method for acute kidney injury assessment as described in any one of claims 1-5.

Images