Mixing index-based diabetic nephropathy prediction system and prediction device
By using weighted quantile regression models to construct a mixed index in the prediction system of diabetic nephropathy, combined with clinical characteristics and retinal vascular geometric parameters, and using machine learning methods to predict, the problem of collinearity and mixed effects in the existing technology is solved, and the accuracy of prediction and the opportunity for early intervention is improved.
Patent Information
- Application Number
- CN202411874097.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-19
- Publication Date
- 2025-04-15
AI Technical Summary
The prior art has shortcomings in the early diagnosis and prediction of diabetic nephropathy, and the failure to effectively consider the collinearity and mixed effects between retinal vascular geometric parameters, resulting in insufficient diagnostic accuracy and specificity.
A mixed index based on weighted quantile regression model was used to construct a mixed index through retinal vascular geometric parameters, and combined with clinical characteristics, a predictive model was constructed using machine learning methods to predict diabetic nephropathy.
By fully considering the differences and internal correlations between individuals, effectively integrating retinal vascular geometric parameters has improved the prediction accuracy of diabetic nephropathy and provided an opportunity for earlier intervention.
Smart Images
Figure CN119833155A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of computer systems based on specific calculation models, and in particular to a diabetic nephropathy prediction system and prediction device based on a mixed index. Background Art
[0002] Diabetic nephropathy (DN) is the leading cause of end-stage renal disease worldwide. Early diagnosis and prognosis prediction are crucial to prevent DN from developing into uremia. However, renal biopsy diagnosis not only increases the risk of bleeding and infection, but is also expensive and has poor accessibility in remote areas. Both DN and diabetic retinopathy are microvascular complications of diabetes. Retinal microvessels are microvessels that can be directly observed and measured in the body.
[0003] Advances in computer methods have made it possible to accurately evaluate retinal microvasculature, such as quantitative measurement of retinal vascular geometric parameters such as vessel diameter, arteriovenous ratio, curvature, fractal dimension, and vessel density. These parameters have been widely used in studies of hypertension, stroke, and coronary artery disease, showing consistent associations. In addition, studies have found that changes in retinal vascular geometric parameters are associated with systemic diabetic complications, such as diabetic retinopathy and cardiovascular disease. Recent studies have shown that there is a correlation between retinal vascular geometric parameters and kidney damage. Changes in retinal microvascular structure may reflect changes in systemic microvasculature in diabetic patients and may serve as an early indicator of diabetic nephropathy. However, current studies analyzed retinal microvascular geometric parameters independently without considering the collinearity and mixed effects between the parameters. Therefore, it is necessary to study the association between the collinearity and mixed effects between the parameters and the prediction of diabetic nephropathy.
[0004] At present, there are several non-invasive DN diagnostic models based on clinical parameters at home and abroad, which usually only include some common indicators, such as blood sugar and blood pressure, but ignore other possible factors. They cannot fully reflect the situation of diabetic nephropathy, and their accuracy and specificity are worrying. Summary of the invention
[0005] In view of the above technical problems existing in the prior art, the present invention provides a diabetic nephropathy prediction system and prediction device based on a mixed index, which reflects the mixed effect of retinal vascular geometric parameters through the mixed index and predicts diabetic nephropathy based on the mixed index.
[0006] The present invention discloses a diabetic nephropathy prediction system based on a hybrid index, comprising an acquisition module, a weighting module and a prediction module, wherein the acquisition module is used to acquire clinical characteristics and retinal vascular geometric parameters of patients; the weighting module is used to construct a hybrid index through the retinal vascular geometric parameters based on a weighted quantile regression model; the prediction module is used to analyze the clinical characteristics and the hybrid index based on a prediction model to predict diabetic nephropathy or non-diabetic nephropathy.
[0007] Preferably, the retinal vascular geometric parameters are selected from the following features:
[0008] Blood vessel fractal dimension variable b, arterial fractal dimension variable k, venous tortuosity, arterial tortuosity, venous fractal dimension variable k, venous fractal dimension variable b, arterial diameter, arterial tortuosity, blood vessel tortuosity, fractal dimension variable k, blood vessel diameter, venous diameter, tortuosity density, branching angle, venous tortuosity density, and arterial fractal dimension variable b.
[0009] Preferably, the mixed index is calculated as follows:
[0010] WQS-diagnosis=B0+Fb*B1+Fka*B2+Fbv*B3+Dv*B4+TDa*B5+DMa*B6+TD*B7+DMv*B8+B*B9+Da*B 10 +DM*B 11 +TDv*B 12 +D*B 13 +Fkv*B 14 +Fba*B 15 +Fk*B 16
[0011] Among them, WQS-diagnosis is expressed as a mixed parameter, B0 is a constant, B1-B 16 is the weight coefficient, Fb is the vascular fractal dimension variable b, Fka is the arterial fractal dimension variable k, Fbv is the venous fractal dimension variable b, Dv is the venous diameter, TDa is the arterial curvature density, DMa is the arterial curvature, TD is the curvature density, DMv is the venous curvature, B is the branching angle, Da is the arterial diameter, DM is the vascular curvature, TDv is the venous curvature density, D is the vascular diameter, Fkv is the venous fractal dimension variable k, Fba is the arterial fractal dimension variable b, and Fk is the fractal dimension variable k.
[0012] Preferably, the specific calculation method of the mixing index is:
[0013] WQS-diagnosis=Fb*0.343+Fka*0.213+Fbv*0.179+Dv*0.069+TDa*0.052+DMa*0.049+T D*0.032+DMv*0.030+B*0.013+Da*0.011+DM*0.004+TDv*0.003+D*0.0006+Fkv*0.0003.
[0014] Preferably, the clinical characteristics are selected from the group consisting of: low-density lipoprotein, duration of diabetes, age, systolic blood pressure, glycosylated hemoglobin, diabetic retinopathy, and glomerular filtration rate.
[0015] Preferably, the system further comprises a training module, and the training module is used to train the prediction model.
[0016] Preferably, the training method of the prediction model includes:
[0017] Screen and collect retinal vascular geometric parameters;
[0018] Screening and collection of clinical characteristics;
[0019] Construct a training set;
[0020] The hybrid index is constructed using retinal vascular geometric parameters and training sets;
[0021] Based on the machine learning method, the training set is trained to obtain the prediction model.
[0022] Preferably, the machine learning method is selected from: SVM, random forest, Lasso, Elasticnet, Xgboost and Lightboost.
[0023] Preferably, the prediction model is validated using a test set and a multi-center validation set.
[0024] The present invention also provides a prediction device for diabetic nephropathy, comprising an input module, a processing module and an output module.
[0025] The input module is used to obtain the patient's clinical indicators and retinal vascular geometric parameters;
[0026] The processing module is used to construct a mixed index through the retinal vascular geometric parameters based on a weighted quantile regression model; and to analyze the clinical characteristics and the mixed index based on a prediction model, classify the patient's kidney disease, and predict whether it is diabetic nephropathy or non-diabetic nephropathy;
[0027] The output module is used to output the prediction results.
[0028] Compared with the prior art, the beneficial effects of the present invention are: a weighted quantile regression model is used to calculate mixed effects, which fully considers the differences between individuals and internal correlations and effectively integrates retinal vascular geometric parameters; the prediction model can more accurately identify diabetic nephropathy, thereby providing an earlier opportunity for intervention. BRIEF DESCRIPTION OF THE DRAWINGS
[0029] Figure 1A It is a significance comparison diagram of fractal dimension variable k;
[0030] Figure 1B It is the significance comparison diagram of the vascular fractal dimension variable b;
[0031] Figure 1C It is the significance comparison diagram of arterial fractal dimension variable k;
[0032] Figure 1D It is the significance comparison diagram of arterial fractal dimension variable b;
[0033] Figure 1E It is the significance comparison diagram of the venous fractal dimension variable k;
[0034] Figure 1F It is the significance comparison diagram of the venous fractal dimension variable b;
[0035] Figure 2 It is the significant level analysis diagram of each parameter in the mixed index;
[0036] Figure 3 is a significant comparison chart of the mixed index;
[0037] Figure 4A A is the AUC curve of the logistic regression model verified on the multi-center validation set;
[0038] Figure 4B is the AUC curve graph of the prediction model using only clinical parameters;
[0039] Figure 4C is the AUC curve of the prediction model using only retinal vascular geometric parameters;
[0040] Figure 5 It is a logic block diagram of the diabetic nephropathy prediction system of the present invention. DETAILED DESCRIPTION
[0041] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0042] The present invention is further described in detail below in conjunction with the accompanying drawings:
[0043] The first aspect of the present invention provides a method for predicting diabetic nephropathy based on a hybrid index, comprising the following steps:
[0044] Step 101: Collect the patient's clinical characteristics and retinal vascular geometric parameters.
[0045] Step 102: constructing a hybrid index using the retinal vascular geometric parameters based on a weighted quantile regression model (WQS).
[0046] Step 103: Based on the prediction model, the clinical characteristics and the mixed index are analyzed to classify the patient's kidney disease and predict it as diabetic nephropathy or non-diabetic nephropathy.
[0047] A weighted quantile regression model was used to calculate mixed effects, which fully considered inter-individual differences and internal correlations and effectively integrated retinal vascular geometric parameters; this enabled the prediction model to more accurately identify diabetic nephropathy, thereby providing an earlier opportunity for intervention.
[0048] The method for training the prediction model includes the following steps:
[0049] Step 201: Screening and collecting retinal blood vessel geometric parameters.
[0050] Among them, the retinal vascular geometric parameters are selected from: vascular diameter (D, average vascular diameter), arterial diameter (Da), venous diameter (Dv), fractal dimension variable k (Fk), fractal dimension variable b (Fb), arterial fractal dimension variable k (Fka), arterial fractal dimension variable b (Fba), venous fractal dimension variable k (Fkv), venous fractal dimension variable b (Fbv), vascular tortuosity (DM), arterial tortuosity (DMa), venous tortuosity (DMv), curvature density (TD), arterial curvature density (TDa), venous curvature density (TDv), and branching angle (B).
[0051] Among them, the blood vessel diameter, fractal dimension, curvature, curvature density, and branch angle are calculated based on the features extracted from the eye image, and most of them are calculated using averages. The curvature is the ratio of the arc length of the blood vessel contour fitting curve divided by the chord length. The calculation and acquisition methods of the above geometric parameters are prior art and will not be repeated in the present invention.
[0052] Step 202: Screen and collect clinical characteristics.
[0053] In terms of clinical characteristics (parameters), seven variables with significant impact on the diagnosis of DN were selected based on Lasso regression analysis, including LDL (low-density lipoprotein), duration of diabetes, age, SBP (systolic blood pressure), HbA1c (glycosylated hemoglobin), DR (diabetic retinopathy), and eGFR (estimated glomerular filtration rate).
[0054] Step 203: Construct a training set and a test set.
[0055] The training and testing datasets included data from 374 DKD patients who underwent renal biopsy, fundus imaging, and blood tests in a hospital, which were approved by the patients and the medical ethics committee. The feature distribution of the dataset is shown in Table 1. The following eight features were selected for the development of the prediction model: LDL, duration of diabetes, age, systolic blood pressure, HbA1c, DR, eGFR, and WQS.
[0056] Table 1
[0057]
[0058]
[0059] Step 204: construct a hybrid index using the retinal vascular geometric parameters and the training set.
[0060] Due to the existence of multicollinearity among retinal vascular geometry parameters, a mixed index was constructed using weighted quantile sum regression (WQS) to explore the overall effect of retinal vascular geometry parameters on the identification of diabetic nephropathy, which performed well in characterizing mixed effects. The mixed index can be calculated using the R package gWQS version 3.0.5, which is an empirically based weighted sum of individual parameters. The mixed index represents the level of mixed effects of the parameters, and the parameters of interest are determined by non-negligible weights.
[0061] like Figures 1A-1F There were statistically significant differences in Fk, Fb, Fka, Fba, Fkv and Fbv between the diabetic nephropathy DN group and the non-diabetic nephropathy NDRD group. Figure 2 The weights of various parameters are shown, such as Figure 3, there is a significant difference in the mixing index between DN and NDRD. The calculation formula of the mixing index is expressed as:
[0062] WQS-diagnosis=B0+Fb*B1+Fka*B2+Fbv*B3+Dv*B4+TDa*B5+DMa*B6+TD*B7+DMv*B8+B*B9+Da*B 10 +DM*B 11 +TDv*B 12 +D*B 13 +Fkv*B 14 +Fba*B 15 +Fk*B 16
[0063] Among them, WQS-diagnosis is expressed as a mixed parameter, B0 is a constant, B1-B 16 is the weight coefficient, Fb is the vascular fractal dimension variable b, Fka is the arterial fractal dimension variable k, Fbv is the venous fractal dimension variable b, Dv is the venous diameter, TDa is the arterial curvature density, DMa is the arterial curvature, TD is the curvature density, DMv is the venous curvature, B is the branching angle, Da is the arterial diameter, DM is the vascular curvature, TDv is the venous curvature density, D is the vascular diameter, Fkv is the venous fractal dimension variable k, Fba is the arterial fractal dimension variable b, and Fk is the fractal dimension variable k.
[0064] The specific calculation formula of the mixed index is:
[0065] WQS-diagnosis=Fb*0.343+Fka*0.213+Fbv*0.179+Dv*0.069+TDa*0.052+DMa*0.049+TD*0.0 32+DMv*0.030+B*0.013+Da*0.011+DM*0.004+TDv*0.003+D*0.0006+Fkv*0.0003+Fba*0+Fk*0
[0066] It should be noted that the weight coefficients are slightly different for different training sets, and the parameters with weight coefficients less than the threshold can be ignored and not included in the calculation formula, such as less than 0.001. The calculation formula of the mixed index can be simplified as:
[0067] WQS-diagnosis=Fb*0.343+Fka*0.213+Fbv*0.179+Dv*0.069+TDa*0.052+DMa*0.049+TD*0.032+DMv*0.030+B*0.013+Da*0.011+DM*0.004+TDv*0.003
[0068] Step 205: Based on the machine learning method, the training set is trained to obtain a prediction model.
[0069] A non-invasive prediction model for diabetic nephropathy was constructed by combining clinical parameters and retinal vascular geometry parameters using a variety of machine learning methods. Compared with prediction models such as SVM, random forest, Lasso, Elasticnet, Xgboost, and Lightboost, the logistic regression model performed best on the test set, with an AUC of 0.98. Among them, the AUC of the SVM model was 0.94, the AUC of the random forest was 0.95, the AUC of the Lasso model was 0.94, the AUC of the Elasticnet model was 0.95, the AUC of the Xgboost model was 0.97, and the AUC of the Lightboost model was 0.94. The AUCs of each model were good. The other four indicators of the logistic regression model also performed well: accuracy was 0.92, precision was 0.92, recall was 0.85, and F1 score was 0.88. Furthermore, the performance of the logistic regression model was verified on a multi-center validation set, and an accuracy of 0.91, precision of 0.90, recall of 0.90, F1 score of 0.90, and AUC of 0.98 were obtained. Figure 4A This shows that the model has good generalization ability. See Table 2 for specific test data.
[0070] Table 2
[0071]
[0072] Referring to the dissolution experiment, after considering the mixed effects and specific characteristics of retinal vascular parameters, the WQS-diagnosis mixed effect parameters were constructed, and after integrating with multidimensional data such as clinical characteristics, the performance of the constructed prediction model for identifying DN and NDRD was better than the prediction model using only clinical parameters ( Figure 4B ) or prediction models that only use retinal vascular geometric parameters ( Figure 4C ).
[0073] The second aspect of the present invention also provides a diabetic nephropathy prediction system for implementing the above method, such as Figure 5 As shown, it includes a collection module 1, a weighting module 2 and a prediction module 3.
[0074] The acquisition module 1 is used to acquire the patient's clinical characteristics and retinal vascular geometric parameters;
[0075] The weighting module 2 is used to construct a hybrid index through the retinal vascular geometric parameters based on a weighted quantile regression model;
[0076] The prediction module 3 is used to analyze the clinical characteristics and the mixed index based on the prediction model, classify the patient's kidney disease, and predict whether it is diabetic nephropathy or non-diabetic nephropathy.
[0077] The diabetic nephropathy prediction system of the present invention further comprises a training module 4, and the training module 4 is used to train the prediction model.
[0078] The third aspect of the present invention also provides a prediction device for implementing the above method, comprising an input module, a processing module and an output module.
[0079] The input module is used to obtain the patient's clinical indicators and retinal vascular geometric parameters;
[0080] The processing module is used to construct a mixed index through the retinal vascular geometric parameters based on a weighted quantile regression model; based on the prediction model, the clinical characteristics and the mixed index are analyzed to classify the patient's kidney disease and predict whether it is diabetic nephropathy or non-diabetic nephropathy;
[0081] The output module is used to output the prediction results.
[0082] The prediction device also includes a model management module, which is used for training, self-training, or loading the prediction model.
[0083] The present invention adopts a weighted quantile regression model to calculate mixed effects, fully considering inter-individual differences and internal correlations, and effectively integrating parameters such as retinal microvessel diameter, arteriovenous ratio, curvature, fractal dimension and vascular density.
[0084] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.
Claims
1. A diabetic nephropathy prediction system based on a hybrid index, characterized in that: Including acquisition module, weighting module and prediction module, The acquisition module is used to acquire clinical characteristics and retinal vascular geometric parameters of the patient; The weighting module is used to construct a hybrid index through the retinal vascular geometric parameters based on a weighted quantile regression model; The prediction module is used to analyze the clinical characteristics and the mixed index based on the prediction model to predict diabetic nephropathy or non-diabetic nephropathy.
2. The diabetic nephropathy prediction system according to claim 1, characterized in that: Retinal vascular geometry parameters are selected from: Blood vessel fractal dimension variable b, arterial fractal dimension variable k, venous tortuosity, arterial tortuosity, venous fractal dimension variable k, venous fractal dimension variable b, arterial diameter, arterial tortuosity, blood vessel tortuosity, fractal dimension variable k, blood vessel diameter, venous diameter, tortuosity density, branching angle, venous tortuosity density, and arterial fractal dimension variable b.
3. The diabetic nephropathy prediction system according to claim 2, characterized in that: The hybrid index is calculated as: WQS-diagnosis=B0+Fb*B1+Fka*B2+Fbv*B3+Dv*B4+TDa*B5+DMa*B6+TD*B7+DMv *B8+B*B9+Da*B 10 +DM*B 11 +TDv*B 12 +D*B 13 +Fkv*B 14 +Fba*B 15 +Fk*B 16 Among them, WQS-diagnosis is expressed as a mixed parameter, B0 is a constant, B1-B 16 is the weight coefficient, Fb is the vascular fractal dimension variable b, Fka is the arterial fractal dimension variable k, Fbv is the venous fractal dimension variable b, Dv is the venous diameter, TDa is the arterial curvature density, DMa is the arterial curvature, TD is the curvature density, DMv is the venous curvature, B is the branching angle, Da is the arterial diameter, DM is the vascular curvature, TDv is the venous curvature density, D is the vascular diameter, Fkv is the venous fractal dimension variable k, Fba is the arterial fractal dimension variable b, and Fk is the fractal dimension variable k.
4. The diabetic nephropathy prediction system according to claim 3, characterized in that: The specific calculation method of the mixed index is: WQS-diagnosis=Fb*0.343+Fka*0.213+Fbv*0.179+Dv*0.069+TDa*0.052+DMa*0.049+T D*0.032+DMv*0.030+B*0.013+Da*0.011+DM*0.004+TDv*0.003+D*0.0006+Fkv*0.0003.
5. The diabetic nephropathy prediction system according to claim 1, characterized in that: Clinical characteristics were selected from: low-density lipoprotein, duration of diabetes, age, systolic blood pressure, glycated hemoglobin, diabetic retinopathy, and glomerular filtration rate.
6. The diabetic nephropathy prediction system according to claim 1, characterized in that: It also includes a training module, which is used to train the prediction model.
7. The diabetic nephropathy prediction system according to claim 6, characterized in that: The training method of the prediction model includes: Screen and collect retinal vascular geometric parameters; Screening and collection of clinical characteristics; Construct a training set; The hybrid index is constructed using retinal vascular geometric parameters and training sets; Based on the machine learning method, the training set is trained to obtain the prediction model.
8. The diabetic nephropathy prediction system according to claim 7, characterized in that: The machine learning method is selected from: SVM, random forest, Lasso, Elasticnet, Xgboost and Lightboost.
9. The diabetic nephropathy prediction system according to claim 7, characterized in that: The prediction model was validated using a test set and a multi-center validation set.
10. A prediction device for diabetic nephropathy, characterized in that: It includes input module, processing module and output module. The input module is used to obtain the patient's clinical indicators and retinal vascular geometric parameters; The processing module is used to construct a mixed index through the retinal vascular geometric parameters based on a weighted quantile regression model; and to analyze the clinical characteristics and the mixed index based on a prediction model to predict diabetic nephropathy or non-diabetic nephropathy; The output module is used to output the prediction results.