Indirect estimation method and system for reference intervals of children's test indicators

Abstract

This invention provides an indirect estimation method and system for reference intervals of pediatric laboratory indicators, applied in the field of artificial intelligence technology. The method includes: acquiring multiple laboratory indicator data; inputting the multi-indicator data sample dataset into a pre-trained weighted learning network to obtain the correlation between the indicators; performing weighted fusion on each multi-indicator data sample in the dataset according to a preset similarity function to obtain patient graph structure data; inputting the patient graph structure data into a pre-trained graph neural network for cluster analysis to obtain clustering results; determining a graph reconstruction learning objective based on the clustering results; jointly training the graph reconstruction learning objective and the correlation learning objective, and outputting the reference interval result of the target laboratory indicator when the preset joint learning objective is met. This invention can be effectively used in multi-indicator data scenarios without label information.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a method and system for indirectly estimating reference intervals of children's test indicators. Background Art

[0002] The indirect estimation method of reference intervals for children's test indicators aims to use routine laboratory data in laboratory information systems that do not have health labels attached. These data come from a variety of occasions, including but not limited to blood samples from blood donors, participants in regular physical examinations, individuals undergoing lead exposure screening, and patients undergoing minor surgical procedures.

[0003] By analyzing these data for a certain test indicator, clustering technology is used to identify and distinguish individuals with different health status, and then a reference range of the test indicator is constructed; however, most existing methods usually only estimate the reference range based on a single test indicator, which makes them unable to cope with multi-test indicator data scenarios well.

[0004] This shows that the estimation method of the reference interval of children's test indicators in the relevant technology has the technical problem of being difficult to cope with multiple test indicator data scenarios. Summary of the invention

[0005] The present invention provides an indirect estimation method and system for reference intervals of children's test indicators, so as to solve the defect of the prior art that it is difficult to cope with multi-test indicator data scenarios, and realize that it can be effectively used in multi-test indicator data scenarios without label information.

[0006] The present invention provides an indirect estimation method for the reference interval of a child's test index, comprising the following steps: obtaining multiple test index data, wherein the multiple test index data are multiple clinical test index data for children; preprocessing the multiple test index data to obtain a multiple test index data sample data set; inputting the multiple test index data sample data set into a pre-trained weight learning network to obtain the correlation between the test indexes output by the pre-trained weight learning network; iteratively fine-tuning the correlation between other test indexes in the correlation between the test indexes to the target test index based on the correlation learning target; weighted fusion of each multiple test index data sample in the multiple test index data sample data set based on the correlation between the test indexes according to a preset similarity function to obtain patient graph structure data; inputting the patient graph structure data into a pre-trained graph neural network for cluster analysis to obtain a clustering result; determining a graph reconstruction learning target corresponding to the patient graph structure data based on the clustering result; jointly training the graph reconstruction learning target and the correlation learning target, and outputting the reference interval result of the target test index when the preset joint learning target is met.

[0007] According to an indirect estimation method for children's test indicator reference intervals provided by the present invention, when weighted fusion is performed on each multiple test indicator data sample in the multiple test indicator data sample data set based on the correlation between the test indicators according to a preset similarity function, the method also includes: determining a hyperparameter, wherein the hyperparameter is used to adjust the weight of each multiple test indicator data sample in the multiple test indicator data sample data set.

[0008] According to an indirect estimation method for reference intervals of children's test indicators provided by the present invention, after determining the hyperparameter, the method further includes: determining a proportionality coefficient based on the ratio of the data size of the multiple test indicator data sample data set to the hyperparameter; determining a correlation weight matrix based on the correlation of the other test indicators to the target test indicator; determining a first vector composed of all test indicators of the i-th sample in the multiple test indicator data sample data set, and a second vector composed of all test indicators of the j-th sample in the multiple test indicator data sample data set; determining the similarity between the i-th sample and the j-th sample based on the proportionality coefficient, the correlation weight matrix, the first vector and the second vector.

[0009] According to an indirect estimation method for reference intervals of children's test indicators provided by the present invention, the multiple test indicator data sample data sets are input into a pre-trained weight learning network to obtain the correlation between the test indicators output by the pre-trained weight learning network, including: inputting the multiple test indicator data sample data sets into a pre-trained weight learning network; determining the distribution characteristics of the class centers of the multiple test indicator data sample data sets through the pre-trained weight learning network; determining the feature similarity between the target test indicator and other test indicators through a weight learning module based on an attention mechanism in the pre-trained weight learning network to construct a weight matrix; and exponentially enhancing and normalizing the weight matrix through a weight sharpening module of the pre-trained weight learning network to obtain a correlation matrix between the test indicators.

[0010] According to an indirect estimation method for the reference interval of a child test indicator provided by the present invention, before iteratively fine-tuning the correlation of other test indicators in the correlation between the test indicators with the target test indicator based on the correlation learning target, the method also includes: determining the distribution characteristics of the target test indicator of the i-th sample in the multi-test indicator data sample data set, and the weighted latent embedding representation of the target test indicator of the i-th sample; determining the correlation learning target of the target test indicator based on the number of samples of the multi-test indicator data sample data set, the distribution characteristics and the weighted latent embedding representation.

[0011] According to an indirect estimation method for reference intervals of children's test indicators provided by the present invention, the multiple test indicator data are preprocessed to obtain a multiple test indicator data sample data set, including: deleting the multiple test indicator data with missing test indicators to obtain filtered multiple test indicator data; screening the filtered multiple test indicator data based on a preset age range to obtain children's multiple test indicator data; grouping the children's multiple test indicator data according to age and gender respectively to obtain a multiple test indicator data sample data set.

[0012] The present invention also provides an indirect estimation system for reference intervals of children's test indicators, comprising the following modules: an acquisition module, used to acquire multiple test indicator data, wherein the multiple test indicator data are multiple clinical test indicator data for children; a preprocessing module, used to preprocess the multiple test indicator data to obtain a sample data set of multiple test indicator data; a correlation module, used to input the sample data set of multiple test indicator data into a pre-trained weight learning network to obtain the correlation between the test indicators output by the pre-trained weight learning network; a fine-tuning module, used to adjust the correlation between other test indicators in the correlation between the test indicators to the target test indicator based on the correlation learning target; Perform iterative fine-tuning; a fusion module, used to perform weighted fusion on each multi-test indicator data sample in the multi-test indicator data sample data set based on the correlation between the test indicators according to a preset similarity function, to obtain patient graph structure data; a clustering module, used to input the patient graph structure data into a pre-trained graph neural network for clustering analysis, to obtain clustering results; a determination module, used to determine the graph reconstruction learning target corresponding to the patient graph structure data based on the clustering results; an output module, used to jointly train the graph reconstruction learning target and the correlation learning target, and output the reference interval result of the target test indicator when the preset joint learning target is met.

[0013] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, an indirect estimation method for a reference interval of a child's test index as described in any one of the above-mentioned methods is implemented.

[0014] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements a method for indirect estimation of a reference interval of a child's test index as described in any one of the above.

[0015] The present invention also provides a computer program product, including a computer program, which, when executed by a processor, implements the indirect estimation method of the reference interval of a child's test indicator as described in any one of the above.

[0016] The indirect estimation method and system of the reference interval of children's test indicators provided by the present invention cleans the children's test indicator data to construct a multi-test indicator data sample data set; the multi-test indicator data sample data set is input into a pre-trained weight learning network to obtain the correlation between the test indicators output by the pre-trained weight learning network, and according to the correlation learning target, the correlation of other test indicators relative to the target test indicator is iteratively fine-tuned; according to a preset similarity function, each multi-test indicator data sample in the multi-test indicator data sample data set is weightedly fused based on the correlation between the test indicators to obtain patient graph structure data, thereby being able to integrate the information of multiple test indicators to form a patient graph structure data. Into a more comprehensive and rich patient representation, not only the value of a single indicator is considered, but also the interaction and correlation between indicators, so as to more accurately reflect the patient's overall health status; the patient's graph structure data is input into the pre-trained graph neural network for clustering analysis, and the clustering results are obtained, and then the graph reconstruction learning objectives are obtained; the graph reconstruction learning objectives and the correlation learning objectives are jointly trained until the convergence criteria are met, and the final reference interval results are output. Therefore, it can be effectively used in multi-test indicator data scenarios in the absence of label information, thereby solving the technical problem that the estimation method of the reference interval of children's test indicators in the related technology is difficult to cope with multi-test indicator data scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced one by one below. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0018] Figure 1 It is a flow chart of the indirect estimation method of the reference interval of the children's test index provided by the present invention.

[0019] Figure 2 It is a flow chart of the indirect estimation method of the reference interval for children based on multiple clinical test indicators and the auxiliary diagnosis system provided by the present invention.

[0020] Figure 3 It is a schematic structural diagram of the main steps T3 to T8 provided by the present invention.

[0021] Figure 4 It is a structural schematic diagram of the indirect estimation system of the reference interval of children's test indicators provided by the present invention.

[0022] Figure 5 It is a schematic diagram of the physical structure of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0023] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0024] The indirect estimation method of the reference interval of children's test indicators aims to use routine laboratory data in the laboratory information system without health labels. These data come from a variety of occasions, including but not limited to blood samples from blood donors, participants in regular physical examinations, individuals undergoing lead exposure screening, and patients undergoing minor surgical operations. By analyzing these samples for a certain test indicator, clustering technology is used to identify and distinguish individuals with different health conditions, and then the reference interval of the test indicator is constructed, in order to provide a more reliable standard for children's health assessment.

[0025] For example, the currently commonly used indirect estimation method uses the modified Hofimann method, which collects serum creatinine data from 14 laboratories in two regions to establish a normal reference range for serum creatinine. However, the technical difficulties in establishing reference ranges for children's test indicators include the annotation of data sets and the collection restrictions of test indicators.

[0026] Most existing methods rely on direct methods, which require a large amount of data to be labeled. However, in real scenarios, obtaining this labeled data requires huge manpower and financial costs.

[0027] Most existing methods usually estimate the reference interval based on a single test indicator, which makes them unable to cope well with multi-test indicator data scenarios.

[0028] In order to at least partially solve the above problems, the present invention proposes an indirect estimation method for reference intervals of children's test indicators, which can be effectively used in multiple test indicator data scenarios without label information and can be integrated into an auxiliary diagnosis system.

[0029] refer to Figure 1 , Figure 1 It is a flow chart of the indirect estimation method of the reference interval of the children's test index provided by the present invention, such as Figure 1 As shown, the method includes the following steps.

[0030] Step 101, obtaining multiple test index data, wherein the multiple test index data are multiple clinical test index data for children.

[0031] Step 102, preprocessing the multiple test index data to obtain a multiple test index data sample data set.

[0032] In the embodiment of the present invention, after collecting a variety of children's test indicator data, the children's test indicator data are cleaned to construct a multi-test indicator data sample data set.

[0033] Specifically, data cleaning included deleting data on missing indicators, excluding data whose age was out of the study range and grouping the data by age and gender, and individuals with two or more records in one year were represented by the earliest record.

[0034] Step 103, inputting the multiple test index data sample data sets into the pre-trained weight learning network to obtain the correlation between the test indexes output by the pre-trained weight learning network.

[0035] In an embodiment of the present invention, a plurality of test index data sample data sets are input into a weight learning network to obtain the correlation between the test indexes, wherein the training process of the weight learning network is as follows.

[0036] The multi-test indicator data sample data set is input into the preset weight learning network. The input data is first converted into distribution features about the class center. Then, the weight learning module based on the attention mechanism calculates the similarity between the target test indicator features and other test indicator features to form a weight matrix. The weight sharpening module exponentially enhances the weight matrix and normalizes it to obtain the correlation matrix between the test indicators.

[0037] Among them, the weight learning network has a feature importance evaluation function (such as feature importance in decision trees and random forests) or a built-in attention mechanism (commonly used in Transformer models in natural language processing or image recognition) to evaluate the contribution of different test indicators to model prediction, thereby indirectly inferring the correlation between different test indicators.

[0038] Step 104 , based on the correlation learning target, iteratively fine-tune the correlations of other test indicators in the correlations between test indicators to the target test indicator.

[0039] In an embodiment of the present invention, a relevance learning target is designed, and the relevance of each inspection indicator of the sample with respect to the target indicator is iteratively fine-tuned. The training process of the model is guided by the relevance learning target, and the model parameters are iteratively fine-tuned.

[0040] For example, gradient descent or other optimization algorithms are used to update model parameters through correlation learning objectives to minimize the loss function; after each iteration, the correlation between other test indicators and the target indicator is calculated and evaluated, and the optimization objective or model parameters are adjusted according to the evaluation results.

[0041] Through the embodiments of the present invention, the relevance measurement can be effectively integrated into the training process of the model so as to iteratively fine-tune the model parameters.

[0042] Step 105 , weighted fusion is performed on each of the multiple test indicator data samples in the multiple test indicator data sample data set based on the correlation between the test indicators according to a preset similarity function to obtain patient graph structure data.

[0043] In an embodiment of the present invention, a patient similarity function is designed to weightedly fuse multiple indicators of samples according to their relevance, and hyperparameters are introduced to avoid the negative impact of sample sizes of different data sets to obtain patient graph structure data.

[0044] Step 106: Input the patient graph structure data into the pre-trained graph neural network for clustering analysis to obtain clustering results.

[0045] Step 107: Determine a graph reconstruction learning target corresponding to the patient graph structure data based on the clustering result.

[0046] In an embodiment of the present invention, a graph reconstruction learning objective is designed, and the patient graph structure data is input into a graph neural network for representation learning and cluster analysis to generate clustering results of the samples, thereby obtaining their reference intervals.

[0047] Specifically, the graph reconstruction learning goal is to input the patient graph structure data into the graph neural network for representation learning and clustering analysis to generate the clustering results of the samples. The following formulas (1) to (3) can be referred to:

[0048] (1)

[0049] in, Represents the graph reconstruction learning objective, where and Represents the normalized adjacency matrix and reconstruct the adjacency matrix The first ( , ) elements, represents the total number of samples, represents the Frobenius norm.

[0050] (2)

[0051] (3)

[0052] in, Representation clustering The feature transformation or mapping function, is a binary variable (assignment variable) representing a data point Is it a cluster? , represents the cluster center, Indicates that when the value is =1, The samples belong to clusters, and vice versa =0, it does not belong to the target cluster. Number of clusters =3, iteratively update the variables of cluster centers and cluster labels.

[0053] Among them, Graph Neural Network (GNN) refers to the general term for algorithms that use neural networks to learn graph structured data, extract and discover features and patterns in graph structured data, and meet the needs of graph learning tasks such as clustering, classification, prediction, segmentation, and generation.

[0054] In an embodiment of the present invention, the patient graph structure data is input into a graph neural network (e.g., a graph convolutional network, a graph attention network, or a graph autoencoder); a loss function is designed (e.g., cross entropy loss for classification tasks or reconstruction loss for autoencoder tasks), and a GNN model is trained using an optimization algorithm such as gradient descent; a low-dimensional embedding representation of each node is learned through the GNN model, and these representations will be used for subsequent clustering analysis; a suitable clustering algorithm is selected according to the characteristics and requirements of the data, such as K-means, hierarchical clustering, DBSCAN, or a density-based clustering algorithm, and the node embedding (low-dimensional embedding representation) learned by the GNN is used as input to execute the clustering algorithm to group the patients; based on the clustering results, a reference interval is defined for each cluster group. These intervals can be based on common features or statistical properties of patients within the cluster (e.g., mean, median, standard deviation, etc.).

[0055] Through the embodiments of the present invention, representation learning and cluster analysis can be performed through a graph neural network, and finally the clustering results of the samples are generated and their reference intervals are obtained.

[0056] Step 108, jointly train the graph reconstruction learning objective and the relevance learning objective, and output the reference interval result of the objective verification indicator when the preset joint learning objective is met.

[0057] In the embodiment of the present invention, an optimization algorithm is designed to optimize and solve the joint learning objective until the convergence criterion is met, and the final reference interval result is output. The joint learning objective can refer to the following formula (4):

[0058] (4)

[0059] in, represents the joint learning goal, represents the relevance learning objective (main loss), is a hyperparameter used to adjust the weight of the regularization loss in the total loss. Represents the graph reconstruction learning objective (regularization loss).

[0060] In an embodiment of the present invention, in machine learning or deep learning, multiple learning objectives (also called tasks) are jointly trained, which is generally referred to as multi-task learning (MTL). Joint training aims to improve the overall performance and generalization ability of the model by learning multiple related tasks at the same time. Each task has its own loss function and learning objective, but some parameters of the model are shared, which helps to share information and promote each other between tasks.

[0061] In the embodiment of the present invention, all learning objectives are jointly trained until the convergence criterion is met, and the final reference interval result is output.

[0062] Through the above steps of the embodiment of the present invention, multiple test indicator data are obtained, wherein the multiple test indicator data are multiple clinical test indicator data for children; the multiple test indicator data are preprocessed to obtain a sample data set of multiple test indicator data; the sample data set of multiple test indicator data is input into a pre-trained weight learning network to obtain the correlation between test indicators output by the pre-trained weight learning network; based on the correlation learning target, the correlation of other test indicators in the correlation between test indicators to the target test indicator is iteratively fine-tuned; each multiple test indicator in the sample data set of multiple test indicator data is fine-tuned based on the correlation between test indicators according to a preset similarity function. The test indicator data samples are weightedly fused to obtain the patient graph structure data; the patient graph structure data is input into the pre-trained graph neural network for clustering analysis to obtain the clustering results; based on the clustering results, the graph reconstruction learning objectives corresponding to the patient graph structure data are determined; the graph reconstruction learning objectives and the correlation learning objectives are jointly trained, and the reference interval results of the target test indicators are output when the preset joint learning objectives are met. As a result, it can be effectively used in multiple test indicator data scenarios without label information, thereby solving the technical problem that the estimation method of the reference interval of children's test indicators in the related technology is difficult to cope with multiple test indicator data scenarios.

[0063] According to an indirect estimation method for a reference interval of a child test index provided by the present invention, when weighted fusion is performed on each of the multiple test index data samples in the multiple test index data sample data set based on the correlation between the test indexes according to a preset similarity function, the method further includes:

[0064] A hyperparameter is determined, where the hyperparameter is used to adjust the weight of each multi-test indicator data sample in the multi-test indicator data sample dataset.

[0065] In an embodiment of the present invention, a patient similarity function is designed to perform weighted fusion of multiple indicators of the sample according to the correlation, and a hyperparameter is introduced to avoid the negative impact of the sample size of different data sets to obtain patient graph structure data; internal evaluation indicators are designed to analyze the model performance under different hyperparameter settings, and then the hyperparameter settings for optimal performance are obtained.

[0066] According to an indirect estimation method for a reference interval of a child test index provided by the present invention, after determining the hyperparameter, the method further includes:

[0067] Determine the proportionality coefficient based on the ratio of the data size of the multi-test indicator data sample data set to the hyperparameter;

[0068] Determine a correlation weight matrix based on the correlation of other test indicators to the target test indicator;

[0069] Determine a first vector composed of all test indicators of the i-th sample in the multiple test indicator data sample data set, and a second vector composed of all test indicators of the j-th sample in the multiple test indicator data sample data set;

[0070] The similarity between the i-th sample and the j-th sample is determined based on the proportionality coefficient, the correlation weight matrix, the first vector, and the second vector.

[0071] In the embodiment of the present invention, the patient similarity function aims to weight and fuse multiple indicators of the sample according to the correlation, and introduces hyperparameters to avoid the negative impact of the sample size of different data sets. The following formula (5) can be used as a reference:

[0072] (5)

[0073] In the formula, the proportionality coefficient By data size and hyperparameters Composition, the correlation between other test indicators and target test indicators constitutes the weight , Indicates The samples and The similarity between samples, Indicates The vector representation of all test indicators of samples (i.e. the first vector), Indicates The vector representation of all test indicators of samples (i.e. the second vector), represents the transposed matrix, Represents the base of natural logarithms.

[0074] The above internal evaluation indicators analyze the model performance under different hyperparameter settings, and then obtain the hyperparameter settings for optimal performance. For details, please refer to the following formula (6):

[0075] (6)

[0076] In the formula, Represents the healthy population Percentile, which is used to describe a certain percentage of observations in a data set that is lower than or equal to a certain value; according to the outlier detection strategy, the hyperparameter ( ) are set to 50, 100, 200, 300, 400 and 500 respectively, and the internal evaluation index of the output is selected The hyperparameter closest to 1 As the optimal model hyperparameters, Indicates the length of the interval from the 2.5th percentile to the 97.5th percentile; Indicates the length of the interval from the minimum to the maximum value; the 0th percentile ( ) usually represents the minimum value in a data set, while the 100th percentile ( ) represents the maximum value in the data set.

[0077] Here, outliers refer to extremely large and extremely small values ​​in a time series that are far away from the general level of the series. For example, in a normal distribution, about 99.7% of the data points fall within 3 standard deviations of the mean. Therefore, data points outside this range can be considered outliers.

[0078] According to an indirect estimation method for a reference interval of a child test index provided by the present invention, a multi-test index data sample data set is input into a pre-trained weight learning network to obtain the correlation between the test indexes output by the pre-trained weight learning network, including:

[0079] Input the multi-test index data sample data set into the pre-trained weight learning network;

[0080] Determine the distribution characteristics of the class centers of the multi-test index data sample data set through the pre-trained weight learning network;

[0081] Determine the feature similarity between the target inspection index and other inspection indexes through the weight learning module based on the attention mechanism in the pre-trained weight learning network to construct a weight matrix;

[0082] The weight matrix is ​​exponentially enhanced and normalized through the weight sharpening module of the pre-trained weight learning network to obtain the correlation matrix between the test indicators.

[0083] In an embodiment of the present invention, a multi-test index data sample data set is input into a weight learning network, and the input data is first converted into distribution features about the class center, and then the weight learning module based on the attention mechanism calculates the similarity between the target test index feature and other test index features to form a weight matrix. The weight sharpening module exponentially enhances the weight matrix and normalizes it to obtain the correlation matrix between the test indicators. For details, please refer to the following formulas (7) to (10):

[0084] (7)

[0085] in, Indicates In the sample The distribution characteristics of clinical indicators Quantity, Indicates In the category The central value of the test index, Indicates In the sample Clinical index values, represents a positive real parameter that controls the width or "degrees of freedom" of the distribution, where is an index variable used to iterate over all clusters during the summation in the denominator , for each data point and clustering , the denominator will calculate the data points to all cluster centers distance.

[0086] (8)

[0087] Among them, the clinical test indicators as target analytes (target test indicators), represents the weight matrix calculated using the attention mechanism, represents the normalized exponential function, represents the query vector consisting of the distribution feature vectors of non-target test indicators, Indicates In the sample The distribution characteristics of the test indicators are Represents the transposed matrix.

[0088] (9)

[0089] in, Indicates In the sample The weighted latent embedding of the test indicators, represents the query vector consisting of the distribution feature vectors of non-target test indicators, represents the transposed matrix, represents a query vector consisting of the distribution feature vectors of non-target test indicators.

[0090] (10)

[0091] in, It is to weight the correlation of all samples to obtain the correlation between test indicators. represents the normalization function, represents the sharpening function, represents the weight matrix calculated using the attention mechanism, Represents the total number of samples.

[0092] According to an indirect estimation method for a reference interval of a child test indicator provided by the present invention, before iteratively fine-tuning the correlation of other test indicators in the correlation between test indicators to the target test indicator based on the correlation learning target, the method further includes:

[0093] Determine the distribution characteristics of the target test indicator of the i-th sample in the multi-test indicator data sample data set, and the weighted potential embedding representation of the target test indicator of the i-th sample;

[0094] Based on the sample quantity, distribution characteristics and weighted potential embedding representation of the multi-test indicator data sample data set, the relevance learning objective of the target test indicator is determined.

[0095] In this embodiment of the present invention, the relevance learning objective can be expressed by the following formula (11):

[0096] (11)

[0097] in, Represents the relevant learning goal, Clinical test indicators are used as target test indicators. Represents the number of samples in the multi-test indicator data sample data set, Indicates In the sample The weighted latent embedding of the test indicators, Indicates In the sample The distribution characteristics of the test indicators.

[0098] According to an indirect estimation method for a reference interval of a child test index provided by the present invention, multiple test index data are preprocessed to obtain a sample data set of multiple test index data, including:

[0099] Deleting the multiple test index data with missing test indexes to obtain filtered multiple test index data;

[0100] The multi-test index data are filtered based on a preset age range to obtain the multi-test index data for children;

[0101] The children's multiple test index data are grouped according to age and gender to obtain a sample data set of multiple test index data.

[0102] In the embodiment of the present invention, a variety of children's test indicator data are collected; the children's test indicator data are cleaned to construct a multi-test indicator data sample data set.

[0103] Among them, data cleaning included deleting missing indicator data, excluding data whose age was not within the study range and grouping the data by age and gender. Individuals with two or more records in one year were represented by the earliest record.

[0104] refer to Figure 2 , Figure 2 It is a flow chart of the indirect estimation method of the reference interval for children based on multiple clinical test indicators and the auxiliary diagnosis system provided by the present invention, and the specific steps are as follows.

[0105] Step T1, collecting data on various children's test indicators.

[0106] Step T2, cleaning the children's test index data.

[0107] Step T3, construct a multi-test indicator data sample data set.

[0108] Step T4, learning and testing the correlation between indicators.

[0109] Step T5, constructing children graph structure data.

[0110] Step T6, performing cluster analysis on the nodes in the graph.

[0111] Step T7, optimize hyperparameters using internal evaluation indicators.

[0112] Step T8, output the optimal reference interval of the target inspection index.

[0113] refer to Figure 3 , Figure 3 It is a structural schematic diagram of the main steps from T3 to T8 provided by the present invention, which specifically includes the following formulas.

[0114] The correlation learning between the inspection indicators can refer to the following formula (12):

[0115] (12)

[0116] In the formula, clinical test indicators as target analytes (target test indicators), represents the weight matrix calculated using the attention mechanism, represents the normalized exponential function, represents the query vector consisting of the distribution feature vectors of non-target test indicators, Indicates In the sample The distribution characteristics of the test indicators are Represents the transposed matrix.

[0117] The graph structure generation can refer to the following formula (13):

[0118] (13)

[0119] In the formula, the proportionality coefficient By data size and hyperparameters Composition, the correlation between other test indicators and target test indicators constitutes the weight , Indicates The samples and The similarity between samples, Indicates The vector representation of all test indicators of samples (i.e. the first vector), Indicates The vector representation of all test indicators of samples (i.e. the second vector), represents the transposed matrix, Represents the base of natural logarithms.

[0120] Among them, the hyperparameter optimization can refer to the following formula (14):

[0121] (14)

[0122] In the formula, Represents the healthy population Percentile, which is used to describe a certain percentage of observations in a data set that are less than or equal to a certain value; according to the outlier detection strategy, Set them to 50, 100, 200, 300, 400 and 500 respectively, and select the internal evaluation index of the output The hyperparameter closest to 1 As the optimal model hyperparameters, Indicates the length of the interval from the 2.5th percentile to the 97.5th percentile; Indicates the length of the interval from the minimum to the maximum value; the 0th percentile ( ) usually represents the minimum value in a data set, while the 100th percentile ( ) represents the maximum value in the data set.

[0123] The joint learning objective can refer to the following formula (15):

[0124] (15)

[0125] in, represents the joint learning objective (main loss), represents the relevance learning objective (main loss), is a hyperparameter used to adjust the weight of the regularization loss in the total loss. Represents the graph reconstruction learning objective (regularization loss).

[0126] Through the indirect estimation method of the reference interval of children's test indicators provided by the present invention, the embodiment of the present invention collects a variety of children's test indicator data; cleans the children's test indicator data to construct a multi-test indicator data sample data set; inputs the multi-test indicator data sample data set into a weight learning network to obtain the correlation between the test indicators; designs a correlation learning target, and iteratively fine-tunes the correlation of each test indicator of the sample with respect to the target indicator; designs a patient similarity function to weightedly fuse multiple indicators of the sample according to the correlation, and introduces hyperparameters to avoid the negative impact of the sample size of different data sets, and obtains patient graph structure data; designs internal evaluation indicators, analyzes the model performance under different hyperparameter settings, and then obtains the hyperparameter settings under the optimal performance; designs a graph reconstruction learning target, inputs the patient graph structure data into the graph neural network for representation learning and cluster analysis, generates the clustering results of the sample, and then obtains its reference interval; jointly trains all learning targets until the convergence criterion is met, and outputs the final reference interval result; proposes an indirect estimation method for children's reference interval based on multiple clinical test indicators, which can be effectively used in multiple test indicator data scenarios without label information, and can be integrated into an auxiliary diagnosis system.

[0127] The indirect estimation system for the reference interval of children's test indicators provided by the present invention is described below. The indirect estimation system for the reference interval of children's test indicators described below and the indirect estimation method for the reference interval of children's test indicators described above can be referenced to each other.

[0128] refer to Figure 4 , Figure 4It is a structural diagram of the indirect estimation system of the reference interval of the child test index provided by the present invention, which includes: an acquisition module 401, a preprocessing module 402, a correlation module 403, a fine-tuning module 404, a fusion module 405, a clustering module 406, a determination module 407 and an output module 408.

[0129] An acquisition module 401 is used to acquire multiple test index data, wherein the multiple test index data are multiple clinical test index data for children;

[0130] A preprocessing module 402 is used to preprocess the multiple test index data to obtain a multiple test index data sample data set;

[0131] The correlation module 403 is used to input multiple test index data sample data sets into the pre-trained weight learning network to obtain the correlation between the test indexes output by the pre-trained weight learning network;

[0132] A fine-tuning module 404 is used to iteratively fine-tune the relevance of other test indicators in the relevance between test indicators to the target test indicator based on the relevance learning target;

[0133] A fusion module 405 is used to perform weighted fusion on each of the multiple test indicator data samples in the multiple test indicator data sample data set based on the correlation between the test indicators according to a preset similarity function to obtain patient graph structure data;

[0134] The clustering module 406 is used to input the patient graph structure data into the pre-trained graph neural network for clustering analysis to obtain clustering results;

[0135] A determination module 407, for determining a graph reconstruction learning target corresponding to the patient graph structure data based on the clustering result;

[0136] The output module 408 is used to jointly train the graph reconstruction learning objective and the relevance learning objective, and output the reference interval result of the objective verification indicator when the preset joint learning objective is met.

[0137] Specifically, the indirect estimation system for the reference interval of the above-mentioned children's test indicators provided by the present invention can implement all the method steps implemented in the above-mentioned indirect estimation method embodiment for the reference interval of the children's test indicators, and can achieve the same technical effect. The parts and beneficial effects of this embodiment that are the same as the method embodiment will not be described in detail here.

[0138] refer to Figure 5 , Figure 5 is a schematic diagram of the physical structure of the electronic device provided by the present invention, such as Figure 5As shown, the electronic device may include: a processor (processor) 510, a communication interface (Communications Interface) 520, a memory (memory) 530 and a communication bus 540, wherein the processor 510, the communication interface 520, and the memory 530 communicate with each other through the communication bus 540. The processor 510 can call the logic instructions in the memory 530 to execute an indirect estimation method for the reference interval of children's test indicators, which method includes: obtaining multiple test indicator data, wherein the multiple test indicator data are multiple clinical test indicator data for children; preprocessing the multiple test indicator data to obtain a multiple test indicator data sample data set; inputting the multiple test indicator data sample data set into a pre-trained weight learning network to obtain the correlation between test indicators output by the pre-trained weight learning network; based on the correlation learning target, iteratively fine-tuning the correlation of other test indicators in the correlation between test indicators to the target test indicator; weightedly fusion of each multiple test indicator data sample in the multiple test indicator data sample data set based on the correlation between test indicators according to a preset similarity function to obtain patient graph structure data; inputting the patient graph structure data into a pre-trained graph neural network for clustering analysis to obtain clustering results; based on the clustering results, determining a graph reconstruction learning target corresponding to the patient graph structure data; jointly training the graph reconstruction learning target and the correlation learning target, and outputting the reference interval result of the target test indicator when the preset joint learning target is met.

[0139] In addition, the logic instructions in the above-mentioned memory 530 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.

[0140] On the other hand, the present invention also provides a computer program product, which includes a computer program, which can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the indirect estimation method of the reference interval of children's test indicators provided by the above-mentioned methods, and the method includes: obtaining multiple test indicator data, wherein the multiple test indicator data are multiple clinical test indicator data for children; preprocessing the multiple test indicator data to obtain a sample data set of multiple test indicator data; inputting the sample data set of multiple test indicator data into a pre-trained weight learning network to obtain the test indicators output by the pre-trained weight learning network. correlation between standards; based on the correlation learning objective, iteratively fine-tune the correlation of other test indicators in the correlation between test indicators to the target test indicator; according to the preset similarity function, weighted fusion is performed on each multi-test indicator data sample in the multi-test indicator data sample data set based on the correlation between test indicators to obtain patient graph structure data; the patient graph structure data is input into the pre-trained graph neural network for clustering analysis to obtain clustering results; based on the clustering results, the graph reconstruction learning objective corresponding to the patient graph structure data is determined; the graph reconstruction learning objective and the correlation learning objective are jointly trained, and the reference interval result of the target test indicator is output when the preset joint learning objective is met.

[0141] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements an indirect estimation method for the reference interval of a child's test index provided by the above-mentioned methods, the method comprising: obtaining multiple test index data, wherein the multiple test index data are multiple clinical test index data for children; preprocessing the multiple test index data to obtain a sample data set of multiple test index data; inputting the sample data set of multiple test index data into a pre-trained weight learning network to obtain the correlation between the test indexes output by the pre-trained weight learning network; based on the correlation learning target , iteratively fine-tune the correlation of other test indicators in the correlation between test indicators to the target test indicator; weightedly fuse each multi-test indicator data sample in the multi-test indicator data sample data set based on the correlation between test indicators according to the preset similarity function to obtain patient graph structure data; input the patient graph structure data into the pre-trained graph neural network for clustering analysis to obtain clustering results; based on the clustering results, determine the graph reconstruction learning target corresponding to the patient graph structure data; jointly train the graph reconstruction learning target and the correlation learning target, and output the reference interval result of the target test indicator when the preset joint learning target is met.

[0142] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0143] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0144] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these 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.

Claims

1. An indirect estimation method for reference intervals of children's test indicators, characterized in that: include: Acquiring multiple test index data, wherein the multiple test index data are multiple clinical test index data for children; Preprocessing the multiple test index data to obtain a multiple test index data sample data set; Inputting the multiple test index data sample data sets into a pre-trained weight learning network to obtain the correlation between the test indexes output by the pre-trained weight learning network; Based on the correlation learning objective, iteratively fine-tuning the correlations of other test indicators in the correlations between the test indicators to the target test indicator; Performing weighted fusion on each of the multiple test indicator data samples in the multiple test indicator data sample data set based on the correlation between the test indicators according to a preset similarity function to obtain patient graph structure data; Inputting the patient graph structure data into a pre-trained graph neural network for clustering analysis to obtain clustering results; Based on the clustering result, determining a graph reconstruction learning objective corresponding to the patient graph structure data; The graph reconstruction learning objective and the relevance learning objective are jointly trained, and a reference interval result of the objective verification indicator is output when a preset joint learning objective is met.

2. The indirect estimation method of the reference interval of children's test indicators according to claim 1 is characterized in that: When weighted fusion is performed on each of the multiple test indicator data samples in the multiple test indicator data sample data set based on the correlation between the test indicators according to the preset similarity function, the method further includes: Determine a hyperparameter, wherein the hyperparameter is used to adjust the weight of each multi-test indicator data sample in the multi-test indicator data sample dataset.

3. The indirect estimation method of the reference interval of children's test indicators according to claim 2 is characterized in that: After determining the hyperparameters, the method further includes: Determining a proportionality coefficient based on a ratio of the data size of the multiple test index data sample data set to the hyperparameter; Determining a correlation weight matrix based on the correlation of the other inspection indicators to the target inspection indicator; Determine a first vector composed of all test indicators of the i-th sample in the multiple test indicator data sample data set, and a second vector composed of all test indicators of the j-th sample in the multiple test indicator data sample data set; Based on the proportionality coefficient, the correlation weight matrix, the first vector, and the second vector, a similarity between the i-th sample and the j-th sample is determined.

4. The indirect estimation method of the reference interval of children's test indicators according to claim 1 is characterized in that: The step of inputting the multiple test index data sample data sets into a pre-trained weight learning network to obtain the correlation between the test indexes output by the pre-trained weight learning network comprises: Inputting the multiple test index data sample data set into a pre-trained weight learning network; Determining the distribution characteristics of the class centers of the multi-test index data sample data set through the pre-trained weight learning network; Determine the feature similarity between the target inspection index and other inspection indexes through the weight learning module based on the attention mechanism in the pre-trained weight learning network to construct a weight matrix; The weight matrix is ​​exponentially enhanced and normalized through the weight sharpening module of the pre-trained weight learning network to obtain a correlation matrix between test indicators.

5. The indirect estimation method of the reference interval of children's test indicators according to claim 1 is characterized in that: Before iteratively fine-tuning the relevance of other inspection indicators in the inter-indicator relevance to the target inspection indicator based on the relevance learning target, the method further includes: Determining a distribution characteristic of a target test indicator of an i-th sample in the multi-test indicator data sample data set, and a weighted potential embedding representation of the target test indicator of the i-th sample; Based on the number of samples of the multiple test indicator data sample data set, the distribution characteristics and the weighted potential embedding representation, a relevance learning target of the target test indicator is determined.

6. The indirect estimation method of the reference interval of children's test indicators according to claim 1 is characterized in that: The preprocessing of the multiple test index data to obtain a multiple test index data sample data set includes: Deleting the multiple test index data with missing test indexes to obtain filtered multiple test index data; Filtering the filtered multiple test index data based on a preset age range to obtain multiple test index data for children; The children's multiple test index data are grouped according to age and gender respectively to obtain a multiple test index data sample data set.

7. An indirect estimation system for reference intervals of children's test indicators, characterized in that: include: An acquisition module, used for acquiring multiple test index data, wherein the multiple test index data are multiple clinical test index data for children; A preprocessing module, used to preprocess the multiple test index data to obtain a multiple test index data sample data set; A correlation module, used for inputting the multiple test index data sample data sets into a pre-trained weight learning network to obtain the correlation between the test indexes output by the pre-trained weight learning network; A fine-tuning module, configured to iteratively fine-tune the relevance of other test indicators in the relevance between the test indicators to the target test indicator based on the relevance learning target; A fusion module, configured to perform weighted fusion on each of the multiple test indicator data samples in the multiple test indicator data sample data set based on the correlation between the test indicators according to a preset similarity function to obtain patient graph structure data; A clustering module, used for inputting the patient graph structure data into a pre-trained graph neural network for clustering analysis to obtain clustering results; A determination module, configured to determine a graph reconstruction learning target corresponding to the patient graph structure data based on the clustering result; The output module is used to jointly train the graph reconstruction learning objective and the correlation learning objective, and output the reference interval result of the objective inspection indicator when the preset joint learning objective is met.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the indirect estimation method of the reference interval of the child test index as described in any one of claims 1 to 6 is implemented.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, it implements the indirect estimation method of the reference interval of the child test index as described in any one of claims 1 to 6.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, it implements the indirect estimation method of the reference interval of the child test index as described in any one of claims 1 to 6.

Images