A chronic kidney disease risk screening modeling method and system

Abstract

The present invention relates to the technical field of neural network models, and specifically to a chronic kidney disease risk screening modeling method and system, the method comprising: obtaining the target age of a patient to be screened and the target detection index under each preset detection factor; obtaining the feature weight of the target community to which the target age belongs under each preset detection factor; screening through a chronic kidney disease risk screening network model according to the target feature weight and the target detection index, and obtaining the risk screening factor corresponding to the patient to be screened, wherein during the screening process, the target feature weight under the preset detection factor is used as the initialization weight of the chronic kidney disease risk screening network model for the preset detection factor. The present invention analyzes age and target detection indexes, and adaptively sets the initialization weight of each preset detection factor in the chronic kidney disease risk screening network model, thereby improving the rationality of the initialization weight setting of the neural network model and the efficiency of risk screening.

Description

Technical Field

[0001] The present invention relates to the technical field of neural network models, and in particular to a chronic kidney disease risk screening modeling method and system. Background Art

[0002] With the development of science and technology, the application of neural network models is becoming more and more extensive. For example, it can be applied to the risk screening of chronic kidney disease (CKD). Among them, the neural network model is a relatively common modeling model. In the application process of the neural network model, if multiple factor features are input into the neural network model, since the importance of each factor feature is often different, it is often necessary to set an initialization weight for each factor feature. At present, in the application process of the neural network model, multiple random numbers are often generated and used as the initialization weight of each factor in the neural network model in turn.

[0003] However, if the neural network model used for chronic kidney disease risk screening uses multiple random numbers generated during application as the initialization weights of each detection factor in the neural network model, the following technical problems often occur:

[0004] Since the generated random numbers often cannot represent the true importance of different detection factors for the chronic kidney disease risk screening results, using random numbers as the initialization weights of the detection factors in the neural network model may lead to a poor convergence speed of the neural network model, thereby resulting in poor efficiency of the neural network model screening. Summary of the invention

[0005] In order to solve the technical problem of poor risk screening efficiency caused by poor rationality of initialization weight setting of a neural network model, the present invention proposes a chronic kidney disease risk screening modeling method and system.

[0006] In a first aspect, the present invention provides a chronic kidney disease risk screening modeling method, the method comprising:

[0007] Obtain the target age of patients to be screened and their target detection index under each preset detection factor;

[0008] Obtaining a target community to which the target age belongs, wherein the method for obtaining the target community includes: dividing all historical patients into communities according to the ages of all historical patients to obtain initial communities, and updating all initial communities based on historical detection indicators of historical patients of different ages under different preset detection factors to obtain the target community;

[0009] The characteristic weight of the target community to which the target age belongs under each preset detection factor is obtained as the target characteristic weight under each preset detection factor, wherein the method for obtaining the characteristic weight of each target community under each preset detection factor includes: determining the characteristic obviousness of each target community under each preset detection factor according to the historical detection indicators of all historical patients in each target community under each preset detection factor; determining the characteristic weight of each target community under each preset detection factor according to the characteristic obviousness of each target community under each preset detection factor and the difference between the characteristic obviousness of each target community and other target communities under each preset detection factor;

[0010] According to the target feature weights and target detection indicators under all preset detection factors, screening is performed through the pre-trained chronic kidney disease risk screening network model to obtain the risk screening factors corresponding to the patients to be screened. During the screening process, the target feature weights under the preset detection factors are used as the initialization weights of the chronic kidney disease risk screening network model for the preset detection factors.

[0011] In combination with the first aspect above, in a possible implementation, dividing all historical patients into communities according to their ages to obtain initial communities includes:

[0012] The maximum and minimum ages of all historical patients constitute the target age range;

[0013] Dividing the target age interval into equal parts to obtain age sub-intervals;

[0014] Historical patients whose ages belong to the same age sub-range are divided into the same initial community.

[0015] In combination with the first aspect above, in a possible implementation, all initial communities are updated based on the historical detection indicators of historical patients of different ages under different preset detection factors to obtain the target community, including:

[0016] The mean of the historical detection indicators of all historical patients at each age in each initial community under each preset detection factor is determined as the detection representative indicator corresponding to each preset detection factor at each age in each initial community;

[0017] Fitting the detection representative index corresponding to each preset detection factor at all ages in each initial community to obtain a fitting straight line for each initial community under each preset detection factor;

[0018] The slope of the fitting line of each initial community under each preset detection factor is determined as the change trend factor of each initial community under each preset detection factor;

[0019] According to the difference between the historical detection indicators of each two historical patients under the same preset detection factors, and the difference between the change trend factors of the initial communities to which each two historical patients belong under the same preset detection factors, the target similarity between each two historical patients is determined;

[0020] According to the target similarity between historical patients of different ages, all historical patients are re-divided to obtain the target community.

[0021] In combination with the first aspect above, in a possible implementation, the formula corresponding to the target similarity between two historical patients is:

[0022] ;

[0023] ;

[0024] ;in, is the target similarity between the ath historical patient and the bth historical patient; a and b are the serial numbers of different historical patients; Characterize the overall difference between the change trend factors of the initial communities to which the a-th historical patient and the b-th historical patient belong under the same preset detection factors; Characterize the overall difference between the historical test indicators of the a-th historical patient and the b-th historical patient under the same preset test factors; is a natural exponential function; i is the serial number of the preset detection factor; m is the number of the preset detection factors; It is the absolute value function; is the change trend factor of the initial community to which the a-th historical patient belongs under the i-th preset detection factor; is the change trend factor of the initial community to which the bth historical patient belongs under the i-th preset detection factor; is the historical test index of the a-th historical patient under the i-th preset test factor; It is the historical detection index of the bth historical patient under the i-th preset detection factor.

[0025] In combination with the first aspect above, in a possible implementation, all historical patients are re-divided according to the target similarity between historical patients of different ages to obtain the target community, including:

[0026] Determine the age of the historical patient as the reference age, determine any one of all the reference ages as the marked age, and determine the reference age in the preset neighborhood age range corresponding to the marked age as the candidate age;

[0027] Determine the target similarity between each historical patient at the marked age and each historical patient at each candidate age as a reference similarity, and obtain a reference similarity set between the marked age and each candidate age;

[0028] Determine the average of all reference similarities in the reference similarity set between the marked age and each candidate age as a similarity representative factor between the marked age and each candidate age;

[0029] Selecting a candidate age with the largest similarity representative factor with the marked age from all candidate ages as the target similar age corresponding to the marked age;

[0030] The historical patients with the marked age and the corresponding target similar age are divided into the same target community.

[0031] In combination with the first aspect above, in a possible implementation, determining the characteristic prominence of each target community under each preset detection factor according to the historical detection indicators of all historical patients in each target community under each preset detection factor includes:

[0032] The range of the historical test indicators of all historical patients in each target community under each preset test factor is determined as the indicator distribution difference of each target community under each preset test factor;

[0033] The variance of the historical detection indicators of all historical patients in each target community under each preset detection factor is determined as the indicator discrete factor of each target community under each preset detection factor;

[0034] According to the indicator distribution difference and indicator discrete factor of each target community under each preset detection factor, the characteristic obviousness of each target community under each preset detection factor is determined, among which the indicator distribution difference and indicator discrete factor are positively correlated with the characteristic obviousness.

[0035] In combination with the first aspect, in a possible implementation, determining the feature weight of each target community under each preset detection factor according to the feature prominence of each target community under each preset detection factor and the difference between the feature prominence of each target community and other target communities under each preset detection factor includes:

[0036] Determine the prominence of each target community under each preset detection factor based on the degree of prominence of the characteristics of each target community under each preset detection factor and the absolute value of the difference between the degree of prominence of the characteristics of each target community and other target communities under each preset detection factor;

[0037] The feature weight of each target community under each preset detection factor is determined according to the factor prominence and feature obviousness of each target community under each preset detection factor, wherein the factor prominence and feature obviousness are positively correlated with the feature weight.

[0038] In combination with the first aspect above, in a possible implementation, the formula corresponding to the factor prominence degree of the target community under the preset detection factor is:

[0039] ;in, is the factor prominence of the t-th target community under the i-th preset detection factor; t and h are the serial numbers of different target communities; N is the number of target communities; i is the serial number of the preset detection factor; is the normalization function; is the characteristic prominence of the t-th target community under the i-th preset detection factor; It is the absolute value function; It is the characteristic prominence of the h-th target community under the i-th preset detection factor.

[0040] In combination with the first aspect above, in a possible implementation, determining the feature weight of each target community under each preset detection factor according to the factor prominence degree and feature obviousness degree of each target community under each preset detection factor includes:

[0041] The product of the factor prominence and feature obviousness of each target community under each preset detection factor is normalized to obtain the feature weight of each target community under each preset detection factor.

[0042] In a second aspect, the present invention provides a chronic kidney disease risk screening modeling system, the system comprising:

[0043] A detection index acquisition module is used to obtain the target age of the patient to be screened and the target detection index under each preset detection factor;

[0044] A community acquisition module is used to obtain the target community to which the target age belongs;

[0045] A feature weight acquisition module is used to acquire the feature weight of the target community to which the target age belongs under each preset detection factor as the target feature weight under each preset detection factor;

[0046] The risk screening module is used to perform screening based on the target feature weights and target detection indicators under all preset detection factors through a pre-trained chronic kidney disease risk screening network model to obtain the risk screening factors corresponding to the patients to be screened.

[0047] In a third aspect, a server is provided, comprising a memory and a processor. The memory is used to store executable program code, and the processor is used to call and run the executable program code from the memory, so that the device executes the method in the first aspect or any possible implementation of the first aspect.

[0048] In a fourth aspect, a computer program product is provided, comprising: a computer program code, which, when executed on a computer, enables the computer to execute the method in the first aspect or any possible implementation of the first aspect.

[0049] In a fifth aspect, a computer-readable storage medium is provided, which stores a computer program code. When the computer program code runs on a computer, the computer executes the method in the above-mentioned first aspect or any possible implementation manner of the first aspect.

[0050] The present invention has the following beneficial effects:

[0051] A chronic kidney disease risk screening modeling method of the present invention analyzes age and target detection indicators, and adaptively sets the initialization weight of each preset detection factor in the chronic kidney disease risk screening network model, thereby solving the technical problem of poor risk screening efficiency caused by poor rationality of the initialization weight setting of the neural network model, improving the rationality of the initialization weight setting of the neural network model, and thus improving the risk screening efficiency. Compared with using random numbers as the initialization weights of the preset detection factors in the chronic kidney disease risk screening network model, the present invention comprehensively considers multiple features related to the importance of the screening results, such as age and feature prominence, thereby quantifying the feature weights that characterize the importance of the preset detection factors, thereby improving the rationality of the initialization weight setting of the preset detection factors in the chronic kidney disease risk screening network model, making the set initialization weights closer to the actual required weights, and to a certain extent improving the convergence speed of the chronic kidney disease risk screening network model, thereby improving the efficiency of risk screening through the chronic kidney disease risk screening network model, and to a certain extent avoiding the overfitting of the chronic kidney disease risk screening network model, thereby improving the accuracy of risk screening through the chronic kidney disease risk screening network model. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] In order to more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings required for use in the embodiments or the prior art descriptions are 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.

[0053] Figure 1 A flowchart of a chronic kidney disease risk screening modeling method of the present invention;

[0054] Figure 2 A schematic diagram of the structure of a chronic kidney disease risk screening modeling system of the present invention;

[0055] Figure 3 The figure is a schematic diagram of the structure of a computer device of the present invention. DETAILED DESCRIPTION

[0056] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the specific implementation methods, structures, features and effects of the technical solutions proposed by the present invention are described in detail below in conjunction with the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" does not necessarily refer to the same embodiment. In addition, specific features, structures or characteristics in one or more embodiments may be combined in any suitable form.

[0057] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

[0058] refer to Figure 1 , showing the process of some embodiments of a chronic kidney disease risk screening modeling method of the present invention. The chronic kidney disease risk screening modeling method comprises the following steps:

[0059] Step S1, obtaining the target age of the patient to be screened and the target detection index under each preset detection factor.

[0060] Among them, the patient to be screened may be a patient to be screened for chronic kidney disease risk. The target age may be the age of the patient to be screened. The preset detection factor may be a pre-set factor related to chronic kidney disease detection. For example, the preset detection factor may be, but is not limited to: a glomerular filtration rate factor, a baseline blood pressure factor, a hemoglobin concentration factor, a blood urea nitrogen concentration factor, a blood creatinine concentration factor, an albumin concentration factor, a blood lipid concentration factor, a urine protein concentration factor, a complete parathyroid hormone concentration factor, an electrolyte concentration factor, and a fasting blood glucose concentration factor. The number of preset detection factors may be a preset number, for example, the number of preset detection factors may be 10. The target detection index of the patient to be screened under the preset detection factor may be an index detected by the patient to be screened under the preset detection factor. For example, if a preset detection factor is a glomerular filtration rate factor, the target detection index of the patient to be screened under the preset detection factor may be the glomerular filtration rate of the patient to be screened detected.

[0061] It should be noted that chronic kidney disease (CKD) is a disease caused by kidney damage or decreased renal function due to chronic nephritis, diabetes, hypertension and other reasons. Patients of different ages may be at risk of chronic kidney disease, and for middle-aged and elderly people, they may face higher risks due to physiological changes. If chronic kidney disease is not diagnosed and treated in a timely manner in the early stages, it may seriously affect multiple functional systems of the body in the later stages, and in more serious cases it may develop into uremia. Therefore, it is often necessary to conduct risk screening as early as possible based on various detection indicators of chronic kidney disease to determine the patient's current physical condition.

[0062] As an example, the age of the patient to be screened can be recorded as the target age. If the number of preset detection factors is 11, and these 11 preset detection factors are: glomerular filtration rate factor, baseline blood pressure factor, hemoglobin concentration factor, blood urea nitrogen concentration factor, blood creatinine concentration factor, albumin concentration factor, blood lipid concentration factor, urine protein concentration factor, intact parathyroid hormone concentration factor, electrolyte concentration factor and fasting blood glucose concentration factor, then the glomerular filtration rate, baseline blood pressure, hemoglobin concentration, blood urea nitrogen concentration, blood creatinine concentration, albumin concentration, blood lipid concentration, urine protein concentration, intact parathyroid hormone concentration, electrolyte concentration and fasting blood glucose concentration of the patient to be screened can be collected, respectively, as the target detection indicators under these 11 preset detection factors.

[0063] Step S2, obtaining the target community to which the target age belongs.

[0064] It should be noted that different target communities often contain historical patients of different age groups. If a target community contains historical patients whose age is equal to the target age, then the target community can be used as the target community for the target age. For example, if a target community includes historical patients aged 19 to 24, and the target age is 20, then the target community is the target community for the target age. The historical patients can be patients who have been tested for corresponding indicators under different preset test factors.

[0065] As an example, a method for obtaining a target community may include the following steps:

[0066] The first step is to divide all historical patients into communities according to their ages. The initial community can include the following sub-steps:

[0067] In the first sub-step, the maximum and minimum ages of all historical patients are combined to form the target age range.

[0068] The target age range can be expressed as [ , ] characterization. This is the maximum age of all historical patients. This is the minimum age of all historical patients.

[0069] In the second sub-step, the target age range is divided equally into age sub-ranges.

[0070] The number of age sub-intervals may be a preset number, for example, the number of age sub-intervals may be 4.

[0071] For example, if the target age range is [15, 55], and the number of pre-set age sub-ranges is 4, then the 4 age sub-ranges may be: [15, 25), [25, 35), [35, 45), and [45, 55], respectively.

[0072] The third sub-step is to divide historical patients whose ages belong to the same age sub-range into the same initial community.

[0073] For example, if a certain age sub-interval is [15, 25), historical patients whose ages belong to [15, 25) can form an initial community.

[0074] It should be noted that patients of different age groups often have different emphases on indicators for chronic kidney disease risk assessment. For example, the renal function of elderly patients tends to naturally decrease with age, so the glomerular filtration rate (eGFR) is particularly important for elderly patients. Young patients tend to pay more attention to early indicators of kidney damage, and their urine microalbumin test indicators tend to change more prominently than those of middle-aged and elderly patients. Therefore, in order to facilitate the subsequent quantification of the importance of different test indicators for patients of different age groups, patients of similar ages can be divided into the same initial community.

[0075] In the second step, based on the historical detection indicators of historical patients of different ages under different preset detection factors, all initial communities are updated to obtain the target community.

[0076] The historical detection index of a historical patient under a preset detection factor may be an index detected by the historical patient under the preset detection factor. Secondly, the method for obtaining the historical detection index may be the same as the method for obtaining the target detection index, which will not be described in detail here.

[0077] For example, identifying a target community may include the following sub-steps:

[0078] In the first sub-step, the mean of the historical detection indicators of all historical patients at each age in each initial community under each preset detection factor is determined as the detection representative indicator corresponding to each preset detection factor at each age in each initial community.

[0079] For example, if a preset detection factor is a glomerular filtration rate factor, the age sub-interval corresponding to a certain initial community is [15, 25), and there are historical patients aged 22 in the initial community, then the detection representative indicator corresponding to the glomerular filtration rate factor under 22 years old in the initial community can be equal to: the mean glomerular filtration rate of all historical patients aged 22 in the initial community.

[0080] In the second sub-step, the detection representative index corresponding to each preset detection factor at all ages in each initial community is fitted to obtain a fitting straight line for each initial community under each preset detection factor.

[0081] For example, if a preset detection factor is the glomerular filtration rate factor, the age sub-interval corresponding to a certain initial community is [20, 23), and the ages of historical patients in the initial community include 20, 21, and 22 years old, then the coordinates (20, ), coordinates (21, ) and coordinates (22, ), and the coordinates (20, ), coordinates (21, ) and coordinates (22, ) is used for straight line fitting. The fitted straight line obtained at this time is the fitted straight line of the initial community under the glomerular filtration rate factor. Among them, It is the representative indicator for the detection of glomerular filtration rate factors under 20 years old in this initial community. It is the representative indicator for the detection of glomerular filtration rate factors under 21 years old in this initial community. It is the representative indicator for the detection of glomerular filtration rate factors under 22 years old in this initial community.

[0082] In the third sub-step, the slope of the fitting line of each initial community under each preset detection factor is determined as the change trend factor of each initial community under each preset detection factor.

[0083] The fourth sub-step is to determine the target similarity between each two historical patients based on the difference between the historical detection indicators of each two historical patients under the same preset detection factors, and the difference between the change trend factors of the initial communities to which each two historical patients belong under the same preset detection factors.

[0084] For example, the formula for determining the target similarity between two historical patients can be:

[0085] ;

[0086] ;

[0087] ;in, is the target similarity between the ath historical patient and the bth historical patient. a and b are the serial numbers of different historical patients. Characterizes the overall difference between the change trend factors of the initial communities to which the ath historical patient and the bth historical patient belong under the same preset detection factors. Characterizes the overall difference between the historical detection indicators of the a-th historical patient and the b-th historical patient under the same preset detection factors. is a natural exponential function. i is the serial number of the preset detection factor. m is the number of preset detection factors. It is the absolute value function. It is the change trend factor of the initial community to which the a-th historical patient belongs under the i-th preset detection factor. It is the change trend factor of the initial community to which the bth historical patient belongs under the i-th preset detection factor. It is the historical detection index of the a-th historical patient under the i-th preset detection factor. It is the historical detection index of the bth historical patient under the i-th preset detection factor.

[0088] It should be noted that when The smaller it is, the smaller the difference between the change trend factors of the initial communities to which the a-th historical patient and the b-th historical patient belong under the i-th preset detection factor is, which often indicates that the importance of the i-th preset detection factor for the a-th historical patient and the b-th historical patient is more similar. The smaller it is, the smaller the difference between the historical test indicators of the a-th historical patient and the b-th historical patient under the i-th preset test factor is, which often means that the impact of the i-th preset test factor on the a-th historical patient and the b-th historical patient is more similar. When it is larger, it often means that the effects of different preset detection factors on the a-th historical patient and the b-th historical patient are more similar, which often means that the a-th historical patient and the b-th historical patient can set the same initialization weights for the same preset detection factors when undergoing chronic kidney disease risk screening.

[0089] The fifth sub-step, based on the target similarity between historical patients of different ages, re-dividing all historical patients to obtain the target community may include the following steps:

[0090] First, the age of the historical patient is determined as the reference age, any reference age among all the reference ages is determined as the marked age, and the reference age in the preset neighborhood age range corresponding to the marked age is determined as the candidate age.

[0091] The preset neighborhood age range may be a relatively small age range that is preset. For example, the preset neighborhood age range corresponding to the marked age may be equal to the marked age. If the preset neighborhood age range corresponding to the marked age is equal to the marked age years old, and the marked age is 24 years old, then the preset neighborhood age range corresponding to the marked age can be [22, 26].

[0092] Next, the target similarity between each historical patient at the marked age and each historical patient at each candidate age is determined as the reference similarity, and a reference similarity set between the marked age and each candidate age is obtained.

[0093] The reference similarity set between the marked age and the candidate age may include: the target similarity between each historical patient at the marked age and each historical patient at the candidate age.

[0094] For example, if historical patients whose age is equal to the marked age include: a first historical patient and a second historical patient, a candidate age is determined as the first candidate age, and historical patients whose age is equal to the first candidate age include: a third historical patient and a fourth historical patient, then at this time the reference similarity set between the marked age and the first candidate age may include: the target similarity between the first historical patient and the third historical patient, the target similarity between the first historical patient and the fourth historical patient, the target similarity between the second historical patient and the third historical patient, and the target similarity between the second historical patient and the fourth historical patient.

[0095] Then, the average of all reference similarities in the reference similarity set between the marked age and each candidate age is determined as the similarity representative factor between the marked age and each candidate age.

[0096] Next, the candidate age with the largest similarity representative factor with the above-mentioned marked age is selected from all candidate ages as the target similar age corresponding to the above-mentioned marked age.

[0097] Finally, historical patients with the above marked age and their corresponding target similar age are divided into the same target community.

[0098] For example, if the target similar age corresponding to 22 years old is 21 years old, the target similar age corresponding to 21 years old is 20 years old, the target similar age corresponding to 20 years old is 19 years old, and the target similar age corresponding to 19 years old is 21 years old, then historical patients aged 19 to 22 can be divided into the same target community.

[0099] It should be noted that patients in the same target community can often set the same initialization weights for the same preset detection factors when undergoing chronic kidney disease risk screening.

[0100] Step S3, obtaining the feature weight of the target community to which the target age belongs under each preset detection factor as the target feature weight under each preset detection factor.

[0101] As an example, a method for obtaining the feature weight of each target community under each preset detection factor may include the following steps:

[0102] The first step is to determine the degree of characteristic prominence of each target community under each preset detection factor based on the historical detection indicators of all historical patients in each target community under each preset detection factor, which may include the following sub-steps:

[0103] In the first sub-step, the extreme values ​​of the historical detection indicators of all historical patients in each target community under each preset detection factor are determined as the indicator distribution differences of each target community under each preset detection factor.

[0104] In the second sub-step, the variance of the historical detection indicators of all historical patients in each target community under each preset detection factor is determined as the indicator discrete factor of each target community under each preset detection factor.

[0105] In the third sub-step, the characteristic prominence of each target community under each preset detection factor is determined according to the indicator distribution difference and indicator discrete factor of each target community under each preset detection factor.

[0106] Among them, the indicator distribution difference and indicator discrete factor can be positively correlated with the degree of feature obviousness.

[0107] For example, the formula for determining the characteristic prominence of the target community under the preset detection factors can be:

[0108] ;in, is the characteristic prominence of the t-th target community under the i-th preset detection factor. t is the serial number of the target community. i is the serial number of the preset detection factor. It is the maximum value of the historical detection indicators of all historical patients in the tth target community under the i-th preset detection factor. It is the minimum value of the historical detection indicators of all historical patients in the t-th target community under the i-th preset detection factor. is the difference in indicator distribution of the t-th target community under the i-th preset detection factor. It is the indicator discrete factor of the t-th target community under the i-th preset detection factor, that is, the variance of the historical detection indicators of all historical patients in the t-th target community under the i-th preset detection factor.

[0109] It should be noted that when When is larger, it often indicates that the degree of change between the historical detection indicators of the historical patients in the tth target community under the i-th preset detection factor is greater, and it often indicates that the i-th preset detection factor has a greater impact on the historical patients in the tth target community, and it often indicates that the i-th preset detection factor is more important to the historical patients in the tth target community. When is larger, it often means that the distribution of historical detection indicators of historical patients in the tth target community under the i-th preset detection factor is more discrete, it often means that the i-th preset detection factor changes more significantly in the tth target community, and it often means that the i-th preset detection factor is more important to the historical patients in the tth target community. The larger it is, the more important the i-th preset detection factor is for the historical patients in the t-th target community.

[0110] In the second step, according to the characteristic prominence of each target community under each preset detection factor and the difference between the characteristic prominence of each target community and other target communities under each preset detection factor, determining the characteristic weight of each target community under each preset detection factor may include the following sub-steps:

[0111] In the first sub-step, the factor prominence of each target community under each preset detection factor is determined according to the feature prominence of each target community under each preset detection factor and the absolute value of the difference between the feature prominence of each target community and other target communities under each preset detection factor.

[0112] For example, the formula for determining the factor prominence of the target community under the preset detection factors can be:

[0113] ;in, is the factor prominence of the t-th target community under the i-th preset detection factor. t and h are the serial numbers of different target communities. N is the number of target communities. i is the serial number of the preset detection factor. It is a normalization function, and its function value range can be [0, 1]. It is the characteristic prominence of the t-th target community under the i-th preset detection factor. It is the absolute value function. It is the characteristic prominence of the h-th target community under the i-th preset detection factor.

[0114] It should be noted that when When the value is larger, it often indicates that the i-th preset detection factor is more important to the historical patients in the t-th target community. When is larger, it often means that the indicator change under the i-th preset detection factor in the t-th target community is more likely to be different from the indicator change under the i-th preset detection factor in other target communities, and it often means that the i-th preset detection factor is more likely to represent the factors in the t-th target community. When is larger, it often indicates that the i-th preset detection factor is more important to the historical patients in the t-th target community, and the indicator changes under the i-th preset detection factor in the t-th target community and other target communities are more dissimilar, which often indicates that for the historical patients in the t-th target community, it is more necessary to set a larger initialization weight for the i-th preset detection factor.

[0115] In the second sub-step, the feature weight of each target community under each preset detection factor is determined according to the factor prominence and feature obviousness of each target community under each preset detection factor.

[0116] Among them, the prominence of factors and the obviousness of features can be positively correlated with the feature weight.

[0117] For example, the product of the factor prominence and feature obviousness of each target community under each preset detection factor may be normalized to obtain the feature weight of each target community under each preset detection factor.

[0118] For example, the formula for determining the feature weight of the target community under the preset detection factors can be:

[0119] ;in, is the feature weight of the t-th target community under the i-th preset detection factor. t is the serial number of the target community. i is the serial number of the preset detection factor. is a normalization function. It is the characteristic prominence of the t-th target community under the i-th preset detection factor. It is the factor prominence of the t-th target community under the i-th preset detection factor.

[0120] It should be noted that when When is larger, it often indicates that the i-th preset detection factor is more important to the historical patients in the t-th target community, and it often indicates that for the historical patients in the t-th target community, it is more necessary to set a larger initialization weight for the i-th preset detection factor. When is larger, it often indicates that the i-th preset detection factor is more important to the historical patients in the t-th target community, and the changes in indicators under the i-th preset detection factor in the t-th target community and other target communities are more dissimilar, which often indicates that for the historical patients in the t-th target community, it is more necessary to set a larger initialization weight for the i-th preset detection factor. The larger it is, the more it often means that for the historical patients in the t-th target community, it is necessary to set a larger initialization weight for the ith preset detection factor.

[0121] Step S4, based on the target feature weights and target detection indicators under all preset detection factors, screening is performed through the pre-trained chronic kidney disease risk screening network model to obtain the risk screening factors corresponding to the patients to be screened.

[0122] Among them, the target feature weight under the preset detection factor can characterize the importance of the preset detection factor for the chronic kidney disease risk screening of the patient to be screened. The larger the value, the greater the influence of the preset detection factor on the chronic kidney disease risk screening result of the patient to be screened. During the screening process, the target feature weight under the preset detection factor can be used as the initialization weight of the chronic kidney disease risk screening network model for the preset detection factor. The chronic kidney disease risk screening network model can be used for chronic kidney disease risk screening. For example, the chronic kidney disease risk screening network model can be a BP (Back Propagation Neural Network) model. The risk screening factor can characterize the risk level of a patient with chronic kidney disease. The larger the value, the higher the risk of a patient with chronic kidney disease.

[0123] Optionally, the training process of the chronic kidney disease risk screening network model may include the following steps:

[0124] The first step is to construct a BP model as a chronic kidney disease risk screening network model before training.

[0125] In the second step, the feature weight of the target community to which each historical patient belongs under each preset detection factor is used as the historical feature weight of each historical patient under each preset detection factor.

[0126] The third step is to assess the chronic kidney disease risk level of each historical patient through manual experience.

[0127] In the fourth step, the historical detection indicators and historical feature weights of historical patients under all preset detection factors are used as training samples; the chronic kidney disease risk level of historical patients is used as the training label, and the historical feature weights under the preset detection factors are used as the initialization weights of the preset detection factors during the training process of the chronic kidney disease risk screening network model, and the chronic kidney disease risk screening network model is trained to obtain a trained chronic kidney disease risk screening network model.

[0128] As an example, screening can be performed using a pre-trained chronic kidney disease risk screening network model based on the target feature weights and target detection indicators under all preset detection factors. The chronic kidney disease risk level obtained at this time is the risk screening factor corresponding to the patient to be screened.

[0129] refer to Figure 2 Based on the same inventive concept as the above method embodiment, the present invention provides a chronic kidney disease risk screening modeling system, the system comprising a memory, a processor, and a computer program stored in the memory and executable on the processor. When the above computer program is executed by the processor, the steps of a chronic kidney disease risk screening modeling method may specifically include:

[0130] The detection index acquisition module 201 is used to obtain the target age of the patient to be screened and the target detection index under each preset detection factor;

[0131] The community acquisition module 202 is used to acquire the target community to which the target age belongs;

[0132] The feature weight acquisition module 203 is used to acquire the feature weight of the target community to which the target age belongs under each preset detection factor as the target feature weight under each preset detection factor;

[0133] The risk screening module 204 is used to perform screening based on the target feature weights and target detection indicators under all preset detection factors through a pre-trained chronic kidney disease risk screening network model to obtain the risk screening factors corresponding to the patients to be screened.

[0134] Figure 3 is a schematic diagram of the structure of a computer device provided by an embodiment of the present invention. Figure 3 As shown, the computer device 300 includes: a memory 301, a processor 302, and a computer program 303 stored in the memory 301 and running on the processor 302, wherein when the processor 302 executes the computer program 303, the computer device can execute any one of the chronic kidney disease risk screening modeling methods introduced above.

[0135] Based on the same inventive concept as the above method embodiment, the present invention provides a server, including a memory and a processor. The memory is used to store executable program code, and the processor is used to call and run the executable program code from the memory, so that the device executes any of the above chronic kidney disease risk screening modeling methods.

[0136] Based on the same inventive concept as the above-mentioned method embodiment, the present invention provides a computer program product, which includes: computer program code, when the computer program code runs on a computer, enables the computer to execute any one of the above-mentioned chronic kidney disease risk screening modeling methods.

[0137] Based on the same inventive concept as the above-mentioned method embodiment, the present invention provides a computer-readable storage medium, which stores a computer program code. When the computer program code runs on a computer, the computer executes any one of the above-mentioned chronic kidney disease risk screening modeling methods.

[0138] In summary, compared with using random numbers as the initialization weights of preset detection factors in the chronic kidney disease risk screening network model, the present invention comprehensively considers multiple features related to the importance of the screening results, such as age and feature significance, etc., thereby quantifying the feature weights that characterize the importance of the preset detection factors, thereby improving the rationality of the initialization weight setting of the preset detection factors in the chronic kidney disease risk screening network model, making the set initialization weights closer to the actual required weights, and to a certain extent improving the convergence speed of the chronic kidney disease risk screening network model, thereby improving the efficiency of risk screening through the chronic kidney disease risk screening network model, and to a certain extent avoiding overfitting of the chronic kidney disease risk screening network model, thereby improving the accuracy of risk screening through the chronic kidney disease risk screening network model.

[0139] 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, a person skilled in the art should understand that the technical solutions described in the aforementioned embodiments can still be modified, or some of the technical features can be replaced by equivalents. These modifications or replacements do not deviate the essence of the corresponding technical solutions from the scope of the technical solutions of the embodiments of the present invention, and should all be included in the protection scope of the present invention.

Claims

1. A chronic kidney disease risk screening modeling method, characterized in that: The following steps are involved: Obtain the target age of the patients to be screened and their target detection index under each preset detection factor; Obtaining a target community to which the target age belongs, wherein the method for obtaining the target community includes: dividing all historical patients into communities according to the ages of all historical patients to obtain initial communities, and updating all initial communities based on historical detection indicators of historical patients of different ages under different preset detection factors to obtain the target community; The characteristic weight of the target community to which the target age belongs under each preset detection factor is obtained as the target characteristic weight under each preset detection factor, wherein the method for obtaining the characteristic weight of each target community under each preset detection factor includes: determining the characteristic obviousness of each target community under each preset detection factor according to the historical detection indicators of all historical patients in each target community under each preset detection factor; determining the characteristic weight of each target community under each preset detection factor according to the characteristic obviousness of each target community under each preset detection factor and the difference between the characteristic obviousness of each target community and other target communities under each preset detection factor; According to the target feature weights and target detection indicators under all preset detection factors, screening is performed through the pre-trained chronic kidney disease risk screening network model to obtain the risk screening factors corresponding to the patients to be screened, wherein the target feature weights under the preset detection factors are used as the initialization weights of the chronic kidney disease risk screening network model for the preset detection factors during the screening process, wherein the chronic kidney disease risk screening network model is a BP model; The method of determining the degree of characteristic prominence of each target community under each preset detection factor based on the historical detection indicators of all historical patients in each target community under each preset detection factor includes: The range of the historical test indicators of all historical patients in each target community under each preset test factor is determined as the indicator distribution difference of each target community under each preset test factor; The variance of the historical detection indicators of all historical patients in each target community under each preset detection factor is determined as the indicator discrete factor of each target community under each preset detection factor; According to the indicator distribution difference and indicator dispersion factor of each target community under each preset detection factor, the characteristic obviousness of each target community under each preset detection factor is determined, wherein the indicator distribution difference and indicator dispersion factor are both positively correlated with the characteristic obviousness; The step of determining the feature weight of each target community under each preset detection factor according to the feature prominence of each target community under each preset detection factor and the difference between the feature prominence of each target community and other target communities under each preset detection factor includes: Determine the prominence of each target community under each preset detection factor based on the prominence of each target community's characteristics under each preset detection factor and the absolute value of the difference between the prominence of each target community's characteristics under each preset detection factor and that of other target communities under each preset detection factor; The feature weight of each target community under each preset detection factor is determined according to the factor prominence and feature obviousness of each target community under each preset detection factor, wherein the factor prominence and feature obviousness are positively correlated with the feature weight.

2. A chronic kidney disease risk screening modeling method according to claim 1, characterized in that: According to the ages of all historical patients, all historical patients are divided into communities to obtain initial communities, including: The maximum and minimum ages of all historical patients constitute the target age range; Dividing the target age interval into equal parts to obtain age sub-intervals; Historical patients whose ages belong to the same age sub-range are divided into the same initial community.

3. A chronic kidney disease risk screening modeling method according to claim 1, characterized in that: Based on the historical detection indicators of historical patients of different ages under different preset detection factors, all initial communities are updated to obtain target communities, including: The mean of the historical detection indicators of all historical patients at each age in each initial community under each preset detection factor is determined as the detection representative indicator corresponding to each preset detection factor at each age in each initial community; Fitting the detection representative index corresponding to each preset detection factor at all ages in each initial community to obtain a fitting straight line for each initial community under each preset detection factor; The slope of the fitting line of each initial community under each preset detection factor is determined as the change trend factor of each initial community under each preset detection factor; According to the difference between the historical detection indicators of each two historical patients under the same preset detection factors, and the difference between the change trend factors of the initial communities to which each two historical patients belong under the same preset detection factors, the target similarity between each two historical patients is determined; According to the target similarity between historical patients of different ages, all historical patients are re-divided to obtain the target community.

4. A chronic kidney disease risk screening modeling method according to claim 3, characterized in that: The formula corresponding to the target similarity between two historical patients is: ； ； ;in, It is a The historical patients and b The target similarity between historical patients; a and b It is the serial number of different historical patients; Characterization a The historical patients and b The overall difference between the change trend factors of the initial communities to which the historical patients belong under the same preset detection factors; Characterization a The historical patients and b The overall difference between the historical test indicators of the historical patients under the same preset test factors; is a natural exponential function; i is the serial number of the preset detection factor; m is the number of preset detection factors; It is the absolute value function; It is a The initial community to which the historical patient belongs is in i The change trend factor under the preset detection factors; It is b The initial community to which the historical patient belongs is in i The change trend factor under the preset detection factors; It is a The historical patients i Historical detection indicators under preset detection factors; It is b The historical patients i Historical detection indicators under a preset detection factor.

5. A chronic kidney disease risk screening modeling method according to claim 3, characterized in that: According to the target similarity between historical patients of different ages, all historical patients are re-divided to obtain target communities, including: Determine the age of the historical patient as the reference age, determine any one of all the reference ages as the marked age, and determine the reference age in the preset neighborhood age range corresponding to the marked age as the candidate age; Determine the target similarity between each historical patient at the marked age and each historical patient at each candidate age as a reference similarity, and obtain a reference similarity set between the marked age and each candidate age; Determine the average of all reference similarities in the reference similarity set between the marked age and each candidate age as a similarity representative factor between the marked age and each candidate age; Selecting a candidate age with the largest similarity representative factor with the marked age from all candidate ages as the target similar age corresponding to the marked age; The historical patients with the marked age and the corresponding target similar age are divided into the same target community.

6. A chronic kidney disease risk screening modeling method according to claim 1, characterized in that: The formula corresponding to the factor prominence of the target community under the preset detection factors is: ;in, is the factor prominence of the t-th target community under the i-th preset detection factor; t and h are the serial numbers of different target communities; N is the number of target communities; i is the serial number of the preset detection factor; is the normalization function; is the characteristic prominence of the t-th target community under the i-th preset detection factor; It is the absolute value function; It is the characteristic prominence of the h-th target community under the i-th preset detection factor.

7. A chronic kidney disease risk screening modeling method according to claim 1, characterized in that: Determining the feature weight of each target community under each preset detection factor according to the factor prominence and feature obviousness of each target community under each preset detection factor includes: The product of the factor prominence and feature obviousness of each target community under each preset detection factor is normalized to obtain the feature weight of each target community under each preset detection factor.

8. A chronic kidney disease risk screening modeling system, characterized in that: It comprises a processor and a memory, wherein the processor is used to process instructions stored in the memory to implement a chronic kidney disease risk screening modeling method according to any one of claims 1 to 7.

Images