IgA nephropathy prediction model and construction method thereof

Abstract

The invention relates to the technical field of disease prediction, in particular to an IgA nephropathy prediction model and a construction method thereof, and the construction steps comprise data acquisition, data preprocessing, data mapping and combination, model training based on an OPNs framework, and model prediction. In the data acquisition step, clinical examination data and pathological examination data of a patient sample are acquired; the data preprocessing is used for reducing noise, improving data quality and the like; the data mapping and combination convert the data into a condition which meets the condition of being input into a model constructed based on an OPNs theoretical framework; the model training step based on the OPNs framework is used for training an IgA nephropathy prediction model; the model prediction step is used for predicting IgA pathological data of a patient; according to the method, the orderly standard real number pair theory and the linear regression are combined to predict the pathological data of the patient, and a nonlinear operation rule is added in the level of an operation basic unit, so that the nonlinear relationship between the data can be better mined, and the prediction capability is improved.

Description

Technical Field

[0001] The present invention relates to the field of disease prediction models, and in particular to an IgA nephropathy prediction model and a construction method thereof. Background Art

[0002] With the rapid development of science and technology, machine learning technology is becoming more and more mature and its application areas are becoming more and more extensive. Prediction technology based on machine learning is gaining more and more attention. This technology can help predict the probability of events and provide a more accurate basis for people's decision-making. In the field of disease prediction, machine learning prediction technology can use the patient's clinical characteristics and biomarker data to predict the patient's risk of disease, which helps to identify high-risk patients early and take appropriate intervention measures.

[0003] The diagnosis of IgA nephropathy is a prominent problem in the field of kidney disease. Early diagnosis has important clinical significance for delaying the progression of the disease. At present, the gold standard for diagnosing IgA nephropathy is renal biopsy. Renal biopsy is an invasive examination method, which is prone to complications such as bleeding and aggravates the patient's condition. Clinicians cannot judge treatment based solely on the data at the time of puncture, but must also combine clinical indicators. IgA nephropathy data is relatively complex, with more clinical characteristics and biomarker data. When building a prediction model for IgA nephropathy, it is still a challenge to select the most predictive clinical characteristics and biomarkers that are most meaningful for prediction. At the same time, due to the large number of nonlinear relationships between some data and data, how to use machine learning algorithms to mine and process the relationship between data and improve the prediction ability of machine learning models for IgA nephropathy is one of the important technical topics in this field. Studies have established non-invasive artificial neural network models for predicting the risk of IgA nephropathy based on urine indicators, but this can only determine the presence or absence of IgA nephropathy and does not involve renal pathology. Moreover, in the existing models for predicting diseases, the general practice is to process and calculate the data set in the form of real numbers (or real vectors, real matrices), and it is impossible to handle the relationship between the indicators in the prediction process. Therefore, establishing a prediction model that can predict the Oxford classification of IgA nephropathy with non-invasive indicators and handle the relationship between the indicators has great clinical significance for the judgment of IgA nephropathy and the prediction of pathology. Summary of the invention

[0004] In order to solve the above problems, based on the existing disease prediction model constructed by machine learning technology, the present invention provides an IgA nephropathy prediction model and a construction method thereof.

[0005] In one aspect, the present invention provides a method for constructing an IgA nephropathy prediction model, the construction steps comprising: data collection, data preprocessing, data mapping and data combination, model training based on the OPNs framework, and model prediction, wherein:

[0006] Data collection includes collecting clinical examination data and pathological examination data of patient samples;

[0007] Data preprocessing includes data cleaning, feature selection, data scaling, and data set partitioning;

[0008] Data mapping and combination is to transform the IgA nephropathy data in the real number range into OPNs theoretical canonical real number pairs;

[0009] Model training based on the OPNs framework is to input the data that has been preprocessed, mapped and combined into the algorithm model, and use the matrix in the form of OPNs and label value vector The model parameters are trained to obtain an IgA nephropathy prediction model;

[0010] Model prediction is to use the IgA nephropathy prediction model to predict the patient's IgA pathology data.

[0011] Furthermore, the indicators of clinical examination data include serum creatinine, glomerular filtration rate, blood pressure, and uric acid.

[0012] Furthermore, the pathological examination data is data related to the kidney disease obtained by biopsy of the patient's kidney slices, including five indicators: M, E, S, T, and C, among which:

[0013] M, mesangial hypercellularity, indicates mesangial cell hyperplasia. If more than 50% of the glomeruli have mesangial cell hyperplasia, it is M1, otherwise it is M0.

[0014] E, Endocapillary hypercellularity, indicates the proliferation of capillary endothelial cells: if there is hyperplasia of capillary endothelial cells, it is E1, otherwise it is E0;

[0015] S, segmental glomerulosclerosis, indicates segmental glomerular sclerosis. If there is segmental glomerular sclerosis or adhesion, it is S1, otherwise it is S0;

[0016] T, Tubular atrophy / interstitial fibrosis, indicates tubular atrophy or interstitial fibrosis. T0 indicates that the proportion of tubular atrophy or interstitial fibrosis is less than 25%, T1 indicates that the proportion of tubular atrophy or interstitial fibrosis is greater than 25% but less than 50%, and T2 indicates that the proportion of tubular atrophy or interstitial fibrosis is greater than 50%.

[0017] C, cellular fibrocellular crescents, indicates cellular or fibrocellular crescents, C0 indicates the absence of cellular or fibrocellular crescents, C1 indicates the presence of cellular or fibrocellular crescents in less than 25% of glomeruli, and C2 indicates the presence of cellular or fibrocellular crescents in more than 25% of glomeruli.

[0018] Furthermore, data cleaning, feature selection, data scaling, and data set partitioning in data preprocessing are specifically as follows: eliminating samples with missing data, selecting important feature dimensions with strong explanatory power, performing data enhancement, and obtaining data mapping and combined clinical examination data feature and pathological examination data feature representations; feature selection is performed by combining feature selection algorithms in machine learning with medical practice experience, including using chi-square validation and random forests to score the importance of each feature indicator, and selecting features with higher scores, different coverage, and lower correlation.

[0019] The clinical examination data features are represented as X = [X_1, X_2, ..., X_n]. Among them, X_i represents the i-th feature, n represents a total of n features, and the pathological examination data features are used as the label value vector Y;

[0020] The dataset is divided into training set and test set in a ratio of 7:3 or 8:2.

[0021] Furthermore, data mapping is to map data within the real number range to real number pairs between 0 and 1 according to the forward mapping method in the OPNs theory, and data combination is to combine the data in pairs to form a basic operation unit form that conforms to the OPNs theory.

[0022] Furthermore, the mapping function of the forward mapping method is:

[0023]

[0024] The forward mapping function maps any real number to a value between 0 and 1, and is not equal to 0 or 1.

[0025] Furthermore, the algorithm model is a linear regression model based on the OPNs framework, and its construction process is: using the logical structure of the linear regression algorithm, the basic operation units and their operation methods in the linear regression algorithm are replaced with the binary basic operation units and their operation methods in the OPNs theory, and the loss function used in the algorithm model is constructed using the least squares method, as shown in the following formula:

[0026]

[0027] In the above formula, All appear in the form of OPNs matrix, where are model parameters, is the sample feature matrix, is the true category label vector of the sample, according to Find the partial derivative to get the parameters to be learned:

[0028]

[0029] The operations of elements in the OPNs matrix are not conventional real number operations. They are tuples composed of two numbers between 0 and 1. The basic operations of each number in the tuple, namely scalar multiplication, addition, and multiplication formulas are as follows:

[0030]

[0031] In the above formula, c represents any real number, x and y represent one of the ordered canonical real number pairs between 0 and 1. The function is the above mapping function, is its inverse function. The following basic operation rules exist between OPNs:

[0032] cα=(c⊙μ α ,c⊙ν α ),

[0033]

[0034] In the above formula, the basic operation formulas for OPNs are respectively the scalar multiplication of real numbers and OPNs, and the addition and multiplication between OPNs. In the process of constructing a new model, these formulas are used to replace the basic operation rules in the conventional linear regression algorithm; other subtraction, division, trigonometric function, power function, exponential function, and logarithmic function operations also use the relevant operations in the OPNs theory. By replacing the rules in this way, the operation rules of the conventional model can be converted into the operation rules under the OPNs theoretical framework. Because most of the operation rules of OPNs have nonlinear characteristics, even the most conventional linear regression can have nonlinear operations after replacing the basic operation units and operation rules. At the same time, the basic operation units OPNs in OPNs are paired and combined in pairs, and the possibility of combination increases with the increase of sample characteristics of the data set, so that the number of model parameters, input feature dimensions, and combination methods increase. Since the operation in the OPNs theory is closed, that is, the calculation result obtained by the operation of OPNs is still OPNs, the operation rules of the new model constructed by the method of the present invention are all completed under the OPNs theoretical framework.

[0035] Furthermore, the matrix in the form of OPNs is a matrix composed of OPNs as elements, and each element x in the matrix ij =(a ij ,b ij), and a ij ,b ij ∈(0,1), the specific matrix form is as follows:

[0036]

[0037] Each element in the matrix is ​​calculated by the following mathematical formula:

[0038]

[0039] In the above formula, is an m×n matrix, is an n×p matrix, The product of the sum It is an m×p matrix, i=1,2,…,m; j=1,2,…,p; k=1,2,…,n. is a matrix The element in row i and column k in is a matrix The element at row k and column j in .

[0040] Furthermore, the model prediction process includes:

[0041] (1) Input the data obtained in the data mapping and data combination steps into the trained model, and output the data to be predicted in the form of a binary that conforms to the OPNs theory;

[0042] (2) Split the data to be predicted in binary form into single values, and map the values ​​in the range of 0 to 1 to data in the range of real numbers according to the reverse mapping method in OPNs theory to output the predicted values ​​of IgA nephropathy data;

[0043] The mapping function of the inverse mapping method is the mapping function of the forward mapping method The inverse function of The mapping function is as follows:

[0044]

[0045] The inverse mapping function maps any number between 0 and 1 that is not equal to 0 or 1 to the real number range to obtain the predicted data;

[0046] (3) The predicted data are classified based on the Euclidean distance, and the distance between the predicted value and the label value is determined. The predicted value is classified into the category with the smallest distance, and the prediction of the five indicators M, E, S, T, and C of the patient is completed to obtain the final classification prediction result.

[0047] On the other hand, the present invention also provides an IgA nephropathy prediction model constructed by the above-mentioned construction method.

[0048] Compared with the prior art, the present invention has the following beneficial effects:

[0049] 1. The present invention uses the ordered normal pair (OPNs) theory to construct a machine learning framework, proposes a method for constructing an IgA nephropathy prediction model based on the OPNs linear regression algorithm, adds nonlinear operation rules at the operation basic unit level, and combines the ordered normal pair theory with linear regression, which can greatly improve the prediction ability. When predicting the pathological data of IgA nephropathy patients, the nonlinear relationship between the data can be better mined, which solves the problem of IgA nephropathy prediction with complex data and high precision, establishes a new machine learning data model for IgA nephropathy, provides decision support for the clinical practice of IgA nephropathy, and provides new ideas and methods for research in related fields of IgA nephropathy.

[0050] 2. The IgA nephropathy prediction model constructed by the method for constructing an IgA nephropathy prediction model provided by the present invention can not only provide IgA nephropathy prediction, but also predict the patient's five indicators of M, E, S, T, and C, and finally obtain classified prediction results. From a pathological perspective, it conducts in-depth and detailed prediction analysis on the patient's current status, and can be used as an auxiliary tool to provide prediction results, predict the progression rate and severity of the disease, so that doctors can make a final diagnosis based on professional knowledge and prediction results, help doctors formulate personalized treatment plans, and provide better clinical management. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, 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 work.

[0052] Figure 1 is the feature importance of embodiment 2 of the present invention;

[0053] Figure 2 This is the feature pairing method of Example 5 of the present invention;

[0054] Figure 3 A schematic diagram of a process for constructing an IgA nephropathy prediction model according to Example 5 of the present invention;

[0055] Figure 4 This is a schematic diagram of IgA nephropathy data processing according to Example 5 of the present invention;

[0056] Figure 5 This is a schematic diagram of the process of training the IgA nephropathy prediction model according to Example 5 of the present invention;

[0057] Figure 6 Schematic diagram of the prediction process of the IgA nephropathy prediction algorithm of Example 5 of the present invention. DETAILED DESCRIPTION

[0058] The technical solution of the present invention will be clearly and completely described below in conjunction with specific embodiments. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work belong to the protection scope of the present invention.

[0059] Example 1 Data Collection

[0060] (I) Collection of basic information

[0061] Basic information such as patient's name, gender, date of birth, age, hospitalization number, etc.

[0062] 2. Clinical data collection

[0063] The cause of the disease, age of onset, time of renal biopsy, first clinical symptoms (gross hematuria, foamy urine, edema, hypertension, physical examination findings), past medical history, clinical classification, systolic blood pressure (SBP), diastolic blood pressure (DBP), and mean arterial pressure (diastolic blood pressure + 1 / 3 pulse pressure difference) before the patients were admitted to the hospital for renal puncture were recorded.

[0064] 3. Laboratory Data Collection

[0065] Serum creatinine, blood uric acid, blood lipids, serum total protein, electrolytes, serum albumin, serum immunoglobulin A, complement C3, C-reactive protein, estimated glomerular filtration rate (calculated by CKD-EPI equation), urine red blood cell count, urine red blood cell teratogenicity ratio, 24-hour urine protein, etc.

[0066] 4. Pathological data collection

[0067] Pathologists were asked to evaluate the pathological type of each included case's renal tissue specimen according to the Oxford classification (MEST-C) score. Observe and record renal pathological changes under light microscopy: glomerular sclerosis, crescent size (large crescents are crescents that account for 50% or more of the renal capsule and small crescents are crescents that account for 50% or less of the renal capsule) and composition (cellular crescents, cellular fibrous crescents, fibrous crescents). At the same time, the proportion of the number of glomeruli with crescents in the total number of glomeruli was divided, and the cellular / fibrous cell / fibrous crescents were evaluated according to the relative ratio. Tubular atrophy, interstitial fibrosis, vascular lesions, basement membrane thickening, etc. were observed. Immunofluorescence observation recorded immunoglobulin types (IgA, IgG, IgM, C1q, C3), deposition sites and intensity, etc.

[0068] The data collected in (i), (ii), and (iii) are used as feature variables (predictors) in the data set, and the data collected in (iv) are used as sample label values.

[0069] Example 2 Data Preprocessing

[0070] Data preprocessing mainly includes: data cleaning, feature selection, feature encoding, data scaling, data set division, etc. In this step, samples with missing data are eliminated, and the feature selection algorithm in machine learning and medical practice experience are combined to select data, perform data enhancement, and improve data quality. In the process of data cleaning, the main processing is missing values, removing outliers, and checking data consistency. For the collected IgAN patient data, first process the features with more missing values. For example, if the missing values ​​of a column exceed 5% to 10%, then it is necessary to consider removing the features of this column. For features that can be filled, multiple interpolation methods are mainly used for data filling. Outlier processing mainly involves deleting duplicate data and outliers. The consistency check mainly checks whether the columns in the data set have the same data type, whether they have a reasonable data range, whether they have consistent units, etc.

[0071] In the feature selection step, random forest is mainly used to score the importance of features, select features with higher importance, and combine medical experience to select some features with different coverage and lower correlation among these features with higher importance. Figure 1 As shown in the figure, the features that can be considered are estimated glomerular filtration rate (eGFR), 24-hour urine protein quantification, serum creatinine, urea nitrogen, urine red blood cell count, serum albumin, serum total protein, hemoglobin, uric acid, low-density lipoprotein, high-density lipoprotein, etc. In feature encoding, non-numerical features need to be encoded, such as gender and the presence or absence of various pathological lesions in the Oxford classification. For these two categories of features or label values, they only need to be converted into 0 or 1.

[0072] Example 3 Data Mapping and Combination

[0073] Data mapping and combination is to map the data in the real number range to between 0 and 1 according to the mapping function in the OPNs theory. Data combination is to combine the data in pairs to form a basic operation unit form that conforms to the OPNs theory, so that the model generated by the model building step can process the data formed by the data processing step.

[0074] For example, the low-density lipoprotein of a patient A is 3.27, and the high-density lipoprotein is 0.91. Function mapping, we can get When these two data are combined into an OPNs, we get (0.827, 0.591). This is an OPNs feature composed of two clinical indicators (characteristic variables) of patient a. Such a combination always remains intact during model calculation. For the entire data set, such mapping and combination are performed on two features in all samples.

[0075] For example, among the collected data, there is a set of standardized data, as shown in Table 1 below:

[0076] Table 1. Standardized data sample

[0077]

[0078] Mapping and combining them, the results are shown in Table 2 below:

[0079] Table 2 Data examples of mapping and combination

[0080]

[0081] This completes the mapping and combination of the data set. In actual operation, feature pairing is not completed in one go. It is also a training process. Different combinations of features will have different effects on the output of the model. In other words, each feature can be paired with other features, and then these combinations are re-input into the model for calculation until the end of the loop is reached.

[0082] Example 4 Model training and prediction

[0083] During the model construction process, the algorithm used can be a polynomial regression of any basis function, such as x+x 2 +…+x n Or x+sinx+cosx+…+cotx, etc. Now x+x 2 Take the OPNs framework as an example to illustrate that the algorithm model is written as in It is an OPNs matrix composed of data similar to Table 2. Taking the data in Table 2 as an example, the OPNs matrix it constitutes is as follows:

[0084]

[0085] Here and This is the model parameter to be calculated. With the parameter, we can find it according to this formula That is, the predicted value of the model. The training process is to calculate these parameters in the model. For ease of representation, and Combine into one vector Using the formula Get the final parameters This formula gives It is an OPNs label vector composed of the label value vector Y in the dataset and a column of dummy labels. For example, in the dataset, the label value vector of the pathological T score of 5 samples is Y = [0, 2, 1, 0, 1] T , a single column cannot constitute OPNs, so add a column of dummy labels, that is, a column of zero values ​​[0,0,0,0,0] T , so that a column of OPNs label vectors can be formed without adding any information to the dataset This gives us a label vector in the form of OPNs This method of adding dummy labels also works when there are only an odd number of features.

[0087] With the feature matrix in the form of OPNs and label vector According to OPNs matrix operation and basic operation rules, the parameters can be calculated But it requires the optimal or suboptimal parameters It is necessary to repeatedly obtain evaluation indicators such as the accuracy of the model under different feature pairing combinations until these indicators can reach the global or local optimum.

[0088] From the above, we can see that the final parameter is the OPNs vector. Assuming the final parameters Then the model can be expressed as Output the predicted value. Of course, the model output value here is also OPNs. Assume that the predicted value output by the model is: The effective prediction here is the first term, because during training, the label vector Of course, if Y and 0 are swapped, then the effective prediction value is the second term. Now we need to extract the first term, using The function maps it to the real number range and obtains the final predicted value Y * = After that, we can use the distance to classify them into corresponding categories. Assume that the label value has three categories, 0, 1, and 2. Then the predicted value Y * After classification, it becomes [0,2,0,0,1] T , and the true value is [0,2,1,0,1]T , then we can calculate that the accuracy is 0.8, the precision is 0.87, the recall is 0.8, and the F1 value is 0.79.

[0089] Example 5 Prediction of three-category pathological T score:

[0090] In the following specific data, for the sake of convenience, some data processing steps and data mapping steps are simplified when only data input and output are observed.

[0091] The data set style used is shown in Table 3 below:

[0092] Table 3 Dataset style

[0093]

[0094] The data set consists of 813 samples, 17 feature variables, and out-T is the label value. It is divided into a test set and a training set. The number of each category is shown in Table 4 below:

[0095] Table 4 Quantity of each category

[0096]

[0097]

[0098] The feature pairing method obtained after model training is as follows Figure 2 This is the best pairing method in the 1000 training processes and is the local optimum of these 1000 matching methods. This pairing method is also needed during the test process. The model parameters output after the training are shown in Table 5 below:

[0099] Table 5 Model parameters output after training

[0100]

[0101]

[0102] Some of the prediction results output by the model are shown in Table 6 below:

[0103] Table 6 Partial prediction results of the model output

[0104]

[0105]

[0106] In each row of binary data shown in Table 6, the first item is the effective prediction value that needs to be extracted, and then it is classified into the actual category [0, 1, 2] according to the distance, and the model output data is converted into the actual category, as shown in Table 7 below:

[0107] Table 7 Actual categories of model output data conversion

[0108]

[0109]

[0110] Finally, the weighted accuracy, precision, recall, specificity, and F1 value were calculated based on the above results to obtain the prediction performance of the IgA nephropathy prediction model constructed in this embodiment, as shown in Table 8 below:

[0111] Table 8 Prediction performance of the IgA nephropathy prediction model constructed in this example

[0112] Accuracy: 0.7423312883435583 Accuracy: 0.7763214009172672 Recall: 0.7423312883435583 Specificity: 0.7819409911011349 F1 value: 0.7509610533838629

[0113] The overall processing flow is as follows Figure 3-6 As shown. Figure 3 It can be seen that the IgA nephropathy prediction model constructed in this embodiment combines the ordered normalized real pair theory and linear regression to predict the patient's pathological data, adds nonlinear operation rules at the operation basic unit level, and better mines the nonlinear relationship between data. The final constructed model has a prediction accuracy of approximately 0.7423312883435583, a precision of approximately 0.7763214009172672, a recall rate of approximately 0.7423312883435583, a specificity of approximately 0.7819409911011349, and an F1 value of approximately 0.7509610533838629.

[0114] like Figure 5As shown, since the entire framework is based on the OPNs theory, and the calculation process under this theory is closed, that is, the calculation results between OPNs are still OPNs, so the model parameters obtained after training are also in the form of binary value pairs, so that the general linear regression algorithm model including the basic operation units and operation rules therein can be converted into a linear regression algorithm model based on the OPNs theoretical framework. The operation process of this step meets the calculation rules of the OPNs theory. In the operation process based on this, the calculation will become more complicated, the nonlinear operation will increase, and the nonlinear relationship in the data can be effectively processed. It should be noted that if the label value actually contains only one column, and the data input to the model for training is a binary value pair, that is, the binary data in the form of OPNs requires two columns of label values, then a column of 0 values ​​can be added. This method can meet the conditions for combining into a binary value pair, and at the same time does not add any valid data to the data set. For example, in the IgA nephropathy data set, there are four feature columns A, B, C, and D and a label value column Y. A and C are combined, B and D are combined, and the label value Y and 0 are combined to form binary data that conforms to the OPNs theory. The combined data is input into the model for training, and the trained model is output. The parameter ω in the model i =(ω i1 ,ω i2 ) is also a binary data composed of two numbers. As for whether the true label value Y should be in the first or second item of the binary, you need to consider whether the true prediction value you want to obtain is in the first or second item. If you want the true prediction value to be in the first item, put Y in the first item and 0 in the second item.

[0115] like Figure 6As shown, since the model parameters obtained above are all OPNs, in the prediction process of the test set, it is still necessary to process the IgA nephropathy data into a format that meets the model input requirements, that is, the data should be in the form of a binary. First, the IgA test set data is mapped and combined in the above steps to obtain binary data that conforms to the OPNs theory, and then the data is input into the trained model. After the model is operated, the model prediction data in the form of a binary will be output, and these binary data will be split to obtain a single independent value. After obtaining the data in the form of a binary, if the true prediction value contains only one column, it is specifically located in the first item or the second item of the binary, which depends on the processing of the label value in the above training process. If during the training process, the true label value is located in the first item of the binary value pair in the form of OPNs, then the true prediction data is also located in the first item. After splitting the value pair, only the single value previously located in the first item needs to be processed. After splitting into a single value, the target value needs to be mapped from the value in the range of 0 to 1 to the data in the range of real numbers according to the mapping method in the OPNs theory, so as to output the predicted value of the real IgA nephropathy data. After obtaining a single predicted value, it is subjected to corresponding classification processing. Using the linear regression algorithm based on the OPNs theory, the output result is the regression fitting data, not the probability value of each category after classification prediction. Therefore, the classification processing based on the Euclidean distance is adopted, that is, the distance between the predicted value and the label value is judged, and it is classified into the category with the smallest distance. From then on, the final classification result is obtained, and the prediction of the five indicators of M, E, S, T, and C of the patient is completed.

[0116] The IgA nephropathy prediction model constructed by the IgA nephropathy prediction model construction method provided by the present invention has a high prediction accuracy and can be used as an auxiliary tool to provide prediction results, predict the progression rate and severity of the disease, so that doctors can make a final diagnosis based on professional knowledge and prediction results, help doctors develop personalized treatment plans, and provide better clinical management.

[0117] The present invention is further described above with the aid of specific embodiments. However, it should be understood that the specific description here should not be construed as limiting the essence and scope of the present invention. Various modifications made to the above embodiments by ordinary technicians in the field after reading this specification are all within the scope of protection of the present invention.

Claims

1. A method for constructing an IgA nephropathy prediction model, characterized in that the steps include: Data collection, data preprocessing, data mapping and data combination, model training based on the OPNs framework, and model prediction; The data collection includes collecting clinical examination data and pathological examination data of patient samples; The data preprocessing includes data cleaning, feature selection, data scaling, and data set partitioning; The data mapping and combination is to convert the IgA nephropathy data within the real number range into OPNs theoretical standard real number pairs; The model training based on the OPNs framework is to input the data that has been preprocessed, mapped and combined into the algorithm model, and use the matrix in the form of OPNs and label value vector The model parameters are trained to obtain an IgA nephropathy prediction model; The model prediction is to use the IgA nephropathy prediction model to predict the patient's IgA pathology data.

2. The method for constructing an IgA nephropathy prediction model according to claim 1, characterized in that: The indicators of the clinical examination data include blood creatinine, glomerular filtration rate, blood pressure, and uric acid.

3. The method for constructing an IgA nephropathy prediction model according to claim 1, characterized in that: The pathological examination data is data related to the kidney disease obtained by biopsy of the patient's kidney slices, including five indicators: M, E, S, T, and C, among which: M, mesangial hypercellularity, indicates mesangial cell hyperplasia. If more than 50% of the glomeruli have mesangial cell hyperplasia, it is M1, otherwise it is M0. E, Endocapillary hypercellularity, indicates the proliferation of capillary endothelial cells: if there is hyperplasia of capillary endothelial cells, it is E1, otherwise it is E0; S, segmental glomerulosclerosis, indicates segmental glomerular sclerosis. If there is segmental glomerular sclerosis or adhesion, it is S1, otherwise it is S0; T, Tubular atrophy / interstitial fibrosis, indicates tubular atrophy or interstitial fibrosis. T0 indicates that the proportion of tubular atrophy or interstitial fibrosis is less than 25%, T1 indicates that the proportion of tubular atrophy or interstitial fibrosis is greater than 25% but less than 50%, and T2 indicates that the proportion of tubular atrophy or interstitial fibrosis is greater than 50%. C, cellular fibrocellular crescents, indicates cellular or fibrocellular crescents, C0 indicates the absence of cellular or fibrocellular crescents, C1 indicates the presence of cellular or fibrocellular crescents in less than 25% of glomeruli, and C2 indicates the presence of cellular or fibrocellular crescents in more than 25% of glomeruli.

4. The method for constructing an IgA nephropathy prediction model according to claim 1, characterized in that: The data cleaning, feature selection, data scaling, and data set division in the data preprocessing are specifically as follows: removing samples with missing data, selecting important feature dimensions with strong explanatory power, performing data enhancement, and obtaining data mapping and combined clinical examination data feature and pathological examination data feature representation; The clinical examination data feature is represented by X=[X_1, X_2,…, X_n], where X_i represents the ith feature, n represents a total of n features, and the pathological examination data feature is used as a label value vector Y; The dataset is divided into training set and test set in a ratio of 7:3 or 8:

2.

5. The method for constructing an IgA nephropathy prediction model according to claim 1, characterized in that: The data mapping is to map the data within the real number range to a real number pair between 0 and 1 according to the forward mapping method in the OPNs theory, and the data combination is to combine the data in pairs to form a basic operation unit form that conforms to the OPNs theory.

6. The method for constructing an IgAN nephropathy prediction model according to claim 5, characterized in that: The mapping function of the forward mapping method is: The forward mapping function maps any real number to a value between 0 and 1, and is not equal to 0 or 1.

7. The method for constructing an IgA nephropathy prediction model according to claim 1, characterized in that: The algorithm model is a linear regression model based on the OPNs framework. The construction process is as follows: using the logical structure of the linear regression algorithm, the basic operation units and their operation methods in the linear regression algorithm are replaced with the binary basic operation units and their operation methods in the OPNs theory. The loss function used in the algorithm model is constructed using the least squares method, as shown in the following formula: In the above formula, All appear in the form of OPNs matrix, where is the model parameter, X is the sample feature matrix, is the true category label vector of the sample, according to Find the partial derivative to get the parameters to be learned 8. The method for constructing an IgA nephropathy prediction model according to claim 1, characterized in that: The OPNs matrix is ​​a matrix composed of OPNs as elements. ij =(a ij ,b ij ), and a ij ,b ij ∈(0,1), the specific matrix form is as follows: Each element in the matrix is ​​calculated by the following mathematical formula: In the above formula, is an m×n matrix, is an n×p matrix, and The product of It is an m×p matrix, i=1,2,…,m; j=1,2,…,p; k=1,2,…,n. is a matrix The element in row i and column k in is a matrix The element at row k and column j in .

9. The method for constructing an IgA nephropathy prediction model according to claim 1, characterized in that: The model prediction process includes: (1) inputting the data obtained in the data mapping and data combining steps into the trained model, and outputting the data to be predicted in the form of a binary that conforms to the OPNs theory; (2) splitting the data to be predicted in the binary form into single values, and mapping the numerical values ​​in the range of 0 to 1 to data in the range of real numbers according to the inverse mapping method in the OPNs theory, so as to output the value to be predicted of the IgA nephropathy data; The mapping function of the reverse mapping method is the mapping function of the forward mapping method. The inverse function of The mapping function is as follows: The inverse mapping function maps any number between 0 and 1 and not equal to 0 or 1 to a real number range to obtain predicted data; (3) The predicted data are classified based on the Euclidean distance, and the distance between the predicted value and the label value is determined. The predicted value is classified into the category with the smallest distance, and the prediction of the five indicators M, E, S, T, and C of the patient is completed to obtain the final classification prediction result.

10. A prediction model for IgA nephropathy, characterized in that: The method is constructed according to any one of claims 1 to 9.

Images