Method and system for predicting diabetic nephropathy based on white eye and retinal vessel characteristics

Through a machine learning model based on white eye and retinal vascular characteristics, the problem of insufficient accuracy in early recognition of diabetic nephropathy in the prior art was solved, and non-invasive high-accuracy prediction was achieved, reducing the risk of patients' trauma.

CN119007987APending Publication Date: 2024-11-22THE FIRST MEDICAL CENT CHINESE PLA GENERAL HOSPITAL
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411207376.X
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-30
Publication Date
2024-11-22

AI Technical Summary

Technical Problem

The prior art is difficult to accurately identify diabetic nephropathy in the early stage. The traditional diagnostic methods are invasive operations and poor accuracy, resulting in a high misdiagnosis rate.

Method used

Based on the characteristics of white eyes and retinal blood vessels, a prediction model is constructed through machine learning methods, and the white eyes and retinal blood vessel parameters such as mounds, spots, spots, fog, and moon halos are used to predict diabetic nephropathy.

Benefits of technology

It improves the prediction accuracy of diabetic nephropathy, reduces the risk of patients receiving unnecessary trauma, and provides a non-invasive diagnostic method.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119007987A_ABST
    Figure CN119007987A_ABST
Patent Text Reader

Abstract

The invention discloses a method and system for predicting diabetic nephropathy based on white eye and retinal vessel characteristics, and belongs to the technical field of image recognition, the method comprises the following steps: collecting patient data, the patient data comprising white eye characteristics or retinal vessel characteristics; and based on a prediction model, predicting the patient data to obtain the probability of diabetic nephropathy. The diabetic nephropathy is predicted based on white light features or retinal vessel features, so that the prediction accuracy is improved; and unnecessary wounds on patients can be reduced.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of image recognition technology, and in particular to a method and system for predicting diabetic nephropathy based on sclera and retinal blood vessel characteristics. Background Art

[0002] The global prevalence of diabetes is high, and the age of diabetes onset is getting younger and younger. Diabetic microvascular complications seriously affect the quality of life and bring heavy economic and life burdens. Therefore, the prevention and treatment of diabetic microvascular complications are becoming increasingly severe. Diabetes is one of the main causes of chronic kidney disease, and type 2 diabetes is the largest factor causing the burden of diabetes-related chronic kidney disease. Diabetic nephropathy (DN) is a common microvascular complication of diabetes. DN not only significantly increases the risk of cardiovascular events in patients, but is also one of the main causes of end-stage renal disease.

[0003] Diabetes can cause eye and kidney damage. Diabetic retinopathy (Dr) and DN are both common microvascular complications of diabetes, and there is a close pathophysiological relationship between the two. Microvascular changes can be directly observed through the retina. Predicting diabetic nephropathy based on ocular vascular characteristics has become an important research direction.

[0004] Early DN is not easy to be taken seriously. There are significant differences in the treatment and prognosis of DN and non-diabetic renal disease (NDRD). Therefore, early identification of DN is of great significance. Renal puncture biopsy is the gold standard for diagnosing kidney disease and judging prognosis, but it is an invasive operation, has certain contraindications, requires certain medical conditions to support, and cannot be used as a routine examination. In clinical nephrology, general laboratory tests such as glomerular filtration rate, creatinine, urine protein, and glycated hemoglobin are often combined to identify DN, but they are still invasive operations, and the accuracy is poor, which can easily lead to misdiagnosis. Summary of the invention

[0005] In response to the above-mentioned technical problems existing in the prior art, the present invention provides a method and system for predicting diabetic nephropathy based on sclera and retinal vascular features, which extracts sclera features and retinal vascular features from eye images, predicts diabetic nephropathy, improves prediction accuracy, and reduces unnecessary trauma to patients.

[0006] The present invention discloses a method for predicting diabetic nephropathy based on sclera and retinal vascular features, comprising the following steps: collecting patient data, wherein the patient data includes sclera features or retinal vascular features; based on a prediction model, predicting the patient data to obtain the probability of diabetic nephropathy; wherein the method for obtaining the prediction model comprises: collecting a training set, wherein the indicators of the training set include any of the following sclera features or a combination thereof: the number of hillocks, spots, dots, fog and halo, and the color of sclera blood vessels, and the indicators of the training set also include the number of sclera features of the liver, heart, spleen, lung, kidney, gallbladder, small intestine, stomach, large intestine and bladder, and retinal vascular features; and training the training set based on a machine learning method to obtain a prediction model.

[0007] Preferably, the blood vein color characteristics of the white eye include: pale blood vein, dark blood vein, dark pink blood vein, dark red blood vein and pink blood vein; The retinal vascular characteristics include any of the following indicators or a combination thereof: total vascular diameter, arterial diameter, venous diameter, arteriovenous diameter ratio, total vascular fractal dimension, arterial fractal dimension, venous fractal dimension, total vascular tortuosity, arterial tortuosity, venous tortuosity, total vascular tortuosity density, arterial tortuosity density, venous tortuosity density and vascular branching angle.

[0008] Preferably, the method for obtaining the prediction model specifically includes: A training set was collected, and the patient data of the training set was labeled as: diabetic nephropathy and non-diabetic nephropathy; Based on the methods of inter-group comparison and collinearity test, pre-selected indicators are selected from the indicators of the training set; Based on the method of single factor analysis, the modeling indicators are selected from the pre-selected indicators; Based on the binary logistic regression method, the modeling indicators of the training set are trained to obtain the prediction model.

[0009] Preferably, the preselected indicators include: total number of hills, dark pink color of blood vessels, white clear feature number of kidney area, white clear feature number of lung area, white clear feature number of gallbladder area, venous fractal dimension FDv and arterial curvature density TDa; The modeling indicators include: total number of hillocks, dark pink color of blood vessels, number of white clear features in kidney area, number of white clear features in lung area and fractal dimension of veins.

[0010] Preferably, the prediction model is expressed as:

[0011] in, P DN It is expressed as the probability of predicting diabetic nephropathy, A1 is a constant, and A2-A6 are coefficients; Q represents the total number of hills, Xa represents the dark pink color of blood vessels, Sq represents the white clear characteristic number of kidney area, Fq represents the white clear characteristic number of lung area, and FDv represents the fractal dimension of veins.

[0012] Preferably, the preselected indicators include: duration of diabetes, diabetic retinopathy Dr, smoking history, BMI, total number of hillocks, dark pink blood vessel color, white clear feature number of kidney area, white clear feature number of lung area, white clear feature number of gallbladder area and venous fractal dimension; The modeling indicators include: diabetes course, diabetic retinopathy, smoking history, kidney area leukoplakia characteristic number and vein fractal dimension.

[0013] Preferably, the prediction model is expressed as:

[0014] in, P DN It represents the probability of predicting diabetic nephropathy, A1 is a constant, A2-A6 are coefficients respectively; Tb represents the history of diabetes in months, Dr represents diabetic retinopathy, Xy represents the smoking history, Sq represents the characteristic number of white spots in the renal area, and FDv represents the fractal dimension of veins.

[0015] Preferably, determining whether the probability of diabetic nephropathy is greater than a threshold; If the threshold is exceeded, diabetic nephropathy is predicted; If it is less than the threshold, it is predicted to be non-diabetic nephropathy.

[0016] The present invention also provides a system for implementing the above method, comprising a collection module and a prediction module. The acquisition module is used to collect patient data; The prediction module is used to predict the patient data based on the prediction model to obtain the probability of diabetic nephropathy.

[0017] Preferably, the system further comprises a training module, the training module is used to collect a training set, the indicators of the training set include any of the following white-eye characteristics or a combination thereof: the number of hillocks, spots, dots, fog and moon halo, and the color of white-eye blood vessels, the indicators of the training set also include the number of white-eye characteristics of the liver area, heart area, spleen area, lung area, kidney area, gallbladder area, small intestine area, stomach area, large intestine area and bladder area, and retinal blood vessel characteristics; Based on a machine learning method, the training set is trained to obtain a prediction model.

[0018] Compared with the prior art, the beneficial effects of the present invention are: predicting diabetic nephropathy based on leukoplakia characteristics or retinal vascular characteristics to improve the accuracy of prediction; and reducing unnecessary trauma to patients. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 is a flow chart of a method for predicting diabetic nephropathy based on white eye and retinal vascular characteristics of the present invention; Figure 2 This is a schematic diagram of the location of the internal organs of the white eye characteristics; Figure 3 is the distribution map of the optic disc and concentric circle areas; Figure 4 is the ROC curve diagram of the first prediction model; Figure 5 is the ROC curve graph of the second prediction model; Figure 6 It is a system logic block diagram of the present invention. DETAILED DESCRIPTION

[0020] 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.

[0021] The present invention is further described in detail below in conjunction with the accompanying drawings: The first aspect of the present invention provides a method for predicting diabetic nephropathy based on white eye and retinal vascular characteristics, such as Figure 1 , including the following steps: Step 101: Collect training sets.

[0022] The training set indicators include any of the following clear sky features or their combination: the number of hillocks, spots, dots, mist and moon halo (morphological features), and the blood vessel color of the clear sky (blood vessel color features). The training set indicators also include the number of clear sky features of the liver, heart, spleen, lung, kidney, gallbladder, small intestine, stomach, large intestine and bladder (viscera location features), and retinal vascular features.

[0023] Step 102: Based on a machine learning method, the training set is trained to obtain a prediction model.

[0024] The machine learning methods that can be used are: logistic regression, random forest, XGBoost, Adaboost, Lightboost, ResNet152, etc.

[0025] Step 103: Collect patient data, including iris characteristics, retinal vascular characteristics, and clinical characteristics.

[0026] Step 104: Based on the prediction model, predict the patient data to obtain the probability of diabetic nephropathy.

[0027] Step 105: Determine whether the probability of diabetic nephropathy is greater than a threshold.

[0028] If the threshold is exceeded, step 106 is executed: predicting diabetic nephropathy.

[0029] If it is less than the threshold, step 107 is executed: predicting non-diabetic nephropathy.

[0030] Diabetic nephropathy can be predicted based on leukoplakia characteristics or retinal vascular characteristics to improve the accuracy of the prediction; if diabetic nephropathy is predicted and further diagnosis is required, the patient may be advised to undergo a renal puncture biopsy to reduce unnecessary trauma to the patient.

[0031] In step 103, general data of the patient data are collected including gender, age, duration of diabetes, duration of hypertension, smoking history, drinking history, Dr, systolic blood pressure, diastolic blood pressure, and body mass index (BMI).

[0032] The information collection of white eye characteristics is collected by experienced technicians using the white eye shadowless camera (MyEyeD-10) developed by Biobio Biotech to collect the white eye characteristics of the subjects, automatically identify the location of the internal organs on the surface of the white eye, and subdivide them according to the area where the internal organs are located. The subjects gently put their eyes close to the instrument mask, keep their eyes straight, look straight ahead, and open their eyes as much as possible to ensure that the white part of the eye is fully exposed. Collect 5 images of the patient's left and right eyes, first take the left eye (L) and then the right eye (R), in the orientations of frontal, left, right, upward and downward, respectively, to collect the white eye characteristics of the patient in different locations. Automatically identify and extract the white eye characteristics, including morphological characteristics, color characteristics and the location of the internal organs where the characteristics are located, and quantify the number of times each white eye feature appears. Each white eye feature appears once, which is recorded as 1 point, and the scores of various features are accumulated in sequence. The accumulated scores of various features are used for the final statistical analysis.

[0033] Based on the characteristics and changes of the white eye summarized in Wang Jinjue's book "Diagnosis of Syndrome Differentiation by Looking at the Eyes". Hill: refers to a large, round, oval or irregular opaque protrusion that is higher than the surface of the white eye, with a diameter of more than 0.2cm, which is mainly caused by damp phlegm accumulation syndrome, or damp phlegm and blood stasis syndrome. Spot: refers to a round, oval or various irregular colored patches that are not raised on the surface of the white eye, which is mainly caused by phlegm dampness syndrome or blood stasis, or mainly caused by combined diseases, such as heat, cold, dampness, etc. Dot: refers to a round colored spot that does not protrude from the surface of the white eye, which is mainly caused by qi stagnation and blood stasis syndrome, and can be caused by cold evil, damp evil, heat evil, etc. Fog: refers to the color infiltration state that is not raised on the surface of the white eye and spreads in a foggy manner to the entire white eye, which is mainly caused by internal and external wind syndrome. Moon halo: There is a "dot" in the center and "spots" around the "dot", like a halo of a full moon. This feature is not prominent on the surface of the white eye. It mainly indicates stagnation of phlegm and qi with mild wind syndrome. It may be accompanied by dampness, cold, heat, etc. when the causes and pathogenesis are different.

[0034] Blood vein color characteristics on the white of the eye: Blood veins are the tiny collaterals on the white of the eye. Under normal circumstances, they exist in the dome. When the internal organs are suffering from diseases, clinical manifestations such as "visible shape and color" will appear in the corresponding parts; different colors have different corresponding clinical significances. Blood vein color characteristics on the white of the eye include: light blood veins, dark blood veins, dark pink blood veins, dark red blood veins and pink blood veins. Light blood veins: refers to light blood veins, which are mainly Qi deficiency syndrome. Dark blood veins: refers to the gray-black color in the red blood veins, which are mainly blood stasis syndrome, and often accompanied by cold syndrome. Dark pink blood veins: refers to the blood veins that are dark and pink in the white of the eye, especially the dark gray blood veins, which are mainly blood deficiency and blood stasis syndrome, and can be seen with cold syndrome, or pain syndrome. Dark red blood veins: refers to the blood veins that are dark and red in the white of the eye, which are mainly blood stasis and heat syndrome. Pink blood veins: refers to the pink blood veins in the white of the eye, which are mainly blood deficiency and fever syndrome.

[0035] The location characteristics of the white eye organs: L is the left eye, R is the right eye, such as Figure 2, the score of the white eye viscera location feature is the total number of white eye features identified in this area. Liver area (L12+L14, R12+R14): This area is distributed with the foot Jueyin liver meridian. When the "liver" in traditional Chinese medicine theory becomes ill, the corresponding white eye blood vessel characteristics and white eye characteristics appear in the white eye area corresponding to the liver and gallbladder. Heart area (L5, R5): This area is distributed with the hand Shaoyin heart meridian. Clinical practice shows that when the "heart" in traditional Chinese medicine theory becomes ill, the corresponding white eye blood vessel characteristics and white eye characteristics appear in the inner canthus area. Spleen area (L2+L15, R2+R15): This area is distributed with the foot Taiyin spleen meridian. Clinical practice shows that since the spleen and stomach are the exterior and interior, after the spleen is sick, the corresponding blood vessel characteristics and white eye characteristics often appear next to the stomach. Lung area (L7, R7): This area is distributed with the hand Taiyin lung meridian. However, when the chest and lungs are sick, it can be reflected in the area shown in the white eye. Clinical practice shows that when the lung and its main parts are diseased, the white eye characteristics and white eye blood characteristics can be seen in this area. Kidney area (L8+L10, R8+R10): Clinical practice shows that because the kidney and bladder are exterior and interior, tissue anatomy shows that the kidney and bladder are connected through the ureter, and after kidney disease, the corresponding white eye characteristics and white eye blood characteristics appear on both sides of the bladder. Based on this, it is inferred that this part is distributed with the foot Shaoyin kidney meridian. Gallbladder area (L13, R13): There are foot Shaoyang gallbladder meridian, hand Shaoyang triple burner meridian, hand Taiyang small intestine meridian, foot Jueyin liver meridian, yang Jiao pulse, and yin Jiao pulse. After observation, the distribution of bladder, stomach, and gallbladder organs in the white eye described in the scriptures is consistent with clinical practice. Clinical practice shows that because the gallbladder and liver are exterior and interior, tissue anatomy shows that the gallbladder and liver are closely connected, and after gallbladder disease, the corresponding white eye characteristics and white eye blood characteristics appear in the outer canthus of the white eye. Small intestine area (L4, R4): The hand-sun small intestine meridian is distributed. Clinical practice shows that after the small intestine is sick, and after the heart transfers heat to the small intestine, the corresponding white eye characteristics and white eye blood vessel characteristics often appear in the inner canthus. Stomach area (L1, R1): The foot-yangming stomach meridian is distributed. "Lingshu·On Disease Diagnosis" says: "Diagnosing eye pain, red veins...from bottom to top is Yangming disease." After observation, the distribution of stomach organs in the white eye described in the text is consistent with clinical practice. Large intestine area (L3, R3): After observation, the distribution of large intestine organs in the white eye described in the text is consistent with clinical practice, and the hand-yangming large intestine meridian is distributed in this area. Bladder area (L9, R9): The foot-sun bladder meridian is distributed. "Lingshu·On Disease Diagnosis" says: "Diagnosing eye pain, red veins from top to bottom is Taiyang disease." After observation, the distribution of bladder organs in the white eye described in the text is consistent with clinical practice.

[0036] Retinal vascular geometric parameters were collected using a KOWA retinal camera (VX-20, nonmyd 7) to capture retinal images of the patient's left and right eyes with a 45° field of view. The fundus images were required to include the optic disc and macular area. The right eye retinal image was preferred for measurement. If the right eye image could not be obtained or the retinal image was unclear, the left eye image was selected; cases where both eyes had unclear images were excluded. A fully automatic computer program written in python was used to measure retinal vessels. It combines traditional image processing methods with machine learning algorithms to automatically find the optic disc position, segment the blood vessels, and use the resNet152 neural network algorithm to distinguish between single and multiple blood vessels and arteries and veins. The retinal vessel diameter, fractal dimension, tortuosity, bending density, and branching angle were calculated, as shown in Table 1. Figure 3 , with the midpoint of the optic disc as the center, the area around the optic disc is divided into three concentric circles, namely, A, B, and C, which are concentric circles between 0-0.5, 0.5-1, and 1-2 disc diameters (DD) from the edge of the optic disc, respectively. Parameters of specific areas are measured according to clinical needs.

[0037] Table 1

[0038] Data were analyzed using SPSS statistical software (version 26.0). Continuous variables that met normal distribution were described by mean ± standard deviation, and continuous variables that did not meet normal distribution were described by median (interquartile range). Categorical variables were described by composition ratio or percentage (%). Normal distribution and variance homogeneity analysis were first performed for measurement data. Two independent sample t-tests were used to compare continuous variables that met normal distribution and variance homogeneity between the two groups, and nonparametric tests were used for unequal variance. Count data were analyzed using χ2 test. 2 Test. Retinal vascular geometric parameters and white eye characteristics were analyzed as continuous variables. Variables were selected according to the results of univariate and multivariate analysis and clinical experience, and a differential diagnosis model was established by stepwise logistic regression with DN as the dependent variable. P < 0.05 was considered statistically significant, and the OR values ​​were given with 95% CI, and were considered statistically significant if they did not cross 1.

[0039] The present invention included patients with type 2 diabetes and chronic kidney disease who underwent renal puncture biopsy in the Department of Nephrology, First Medical Center, General Hospital of the Chinese People's Liberation Army from May 2017 to September 2022. A total of 142 cases with complete case data were included, 38 cases (26.8%) of DN combined with NDRD were excluded, and 104 patients with complete parameter data were finally included. Among them, there were 36 patients with DN and 68 patients with NDRD. Among the cases included in the study, 36 cases (34.6%) of DN and 68 cases (65.4%) of NDRD. Among the pathological types of patients in the NDRD group, the highest proportion was membranous nephropathy in 35 cases (51.5%), followed by IgA nephropathy in 16 cases (23.5%).

[0040] Comparison of clinical indicators between the DN group and the NDRD group: As shown in Table 2, both the DN group and the NDRD group were mainly male patients. Compared with the NDRD group, the DN group had a longer course of diabetes, more patients with a history of smoking and Dr, and a lower BMI level, and the difference was statistically significant (P < 0.05). There was no statistical difference in other clinical indicators (P > 0.05).

[0041] Table 2

[0042] Comparison of white eye characteristics between the DN group and the NDRD group: The comparison of the white eye characteristics scores between the DN group and the NDRD group is shown in Table 3. There is no statistical difference in the white eye morphology and blood color characteristics between the DN group and the NDRD group ( P >0.05).

[0043] Table 3

[0044] The comparison of the viscera location scores of the white eye characteristics between the DN group and the NDRD group is shown in Table 4. There was no statistical difference in the viscera location scores of the white eye characteristics between the DN group and the NDRD group ( P >0.05).

[0045] Table 4

[0046] The comparison of retinal vascular geometric parameters between the DN group and the NDRD group is shown in Table 5. Compared with the NDRD group, the retinal vascular FD and FDv levels in the DN group were lower, and the difference was statistically significant (P<0.05), while there were no statistical differences in other retinal vascular geometric parameters (P>0.05).

[0047] Table 5

[0048] Based on the above intergroup comparison, the related factors of the white eye characteristics of DN and NDRD were extracted. In the first specific training, only the white eye characteristics were extracted, and the general clinical characteristics were not extracted. In order to avoid missing important information, P <0.2 was used as the test level, and a collinearity test was performed. The white eye characteristics (hill, dark pink blood vessels, kidney area, lung area, gallbladder area) and retinal vascular geometric parameters (FDv, TDa) that passed the collinearity test were included in the equation. The renal pathology type was used as the dependent variable (DN=1, NDRD=0); after univariate analysis to screen the independent variables, the backward stepwise regression method based on maximum likelihood estimation was used, and the final logistic regression model was statistically significant. P <0.001, see Table 6.

[0049] The indicator that finally entered the model was Qiu ( P =0.010), dark pink blood ( P =0.018), kidney area ( P =0.003), lung area ( P =0.006)、FDv( P <0.001). This indicates that the characteristic scores of the hillock, dark pink blood vessels, kidney area, lung area, and FDv are correlated with DN in people with type 2 diabetes and chronic kidney disease.

[0050] Table 6

[0051] The first prediction model established is as follows: (1) in, P DN It represents the probability of predicting diabetic nephropathy, A1 is a constant, A2-A6 are coefficients respectively; Q represents the total number of hills, Xa represents the dark pink color of blood vessels, Sq represents the white clear characteristic number of kidney area, Fq represents the white clear characteristic number of lung area, and FDv represents the fractal dimension of veins.

[0052] More specifically, the equation of the first prediction model is expressed as: P DN = exp (12.902+0.438×Q+0.165×Xa+0.409×Sq-0.224×Fq-10.396×FDv) / [1+exp(12.902+0.438×Q+0.165×Xa+0.409×Sq-0.224×Fq-10.396×FDv)] (2) P DN ≥0.5 indicates that the patient has DN, PDN <0.5 indicates that the patient is NDRD; based on the binary logistic regression equation, the ROC curve was drawn, as shown in Figure 4 , the horizontal axis is the negative positive rate 1-Specificity, and the vertical axis is the true positive rate Sensitivity. AUC is 0.836, 95% CI (0.753-0.919), SE is 0.042, P The value is <0.001. It should be noted that the coefficients obtained are slightly different when different training methods are used.

[0053] In the second specific training, general clinical indicators, white eye characteristics and retinal vascular geometric parameters were used to establish a prediction model. According to the comparison between the groups, P < 0.2 was used as the test level, and a collinearity test was performed (because the three parameters of the retinal vascular fractal dimension were collinear, FDv with the highest diagnostic efficiency was selected to enter the model). Combined with the comparison between the groups and clinical experience, clinical indicators that passed the collinearity test (diabetes course, Dr, smoking history, BMI), white eye characteristics (hill, dark pink blood vessels, kidney area, lung area, gallbladder area), and retinal vascular geometric parameters (FDv) were included in the equation. With the renal pathology type as the dependent variable (DN = 1, NDRD = 0), the independent variables were screened through univariate analysis, and the backward stepwise regression method based on maximum likelihood estimation was used. The final logistic regression model was statistically significant (P < 0.001), see Table 7.

[0054] The results showed that the duration of diabetes (OR=1.009; P=0.059), Dr (OR=40.842; P<0.001), smoking history (OR=3.528; P=0.092), renal characteristics (OR=1.449; P=0.047), FDv (OR=0.001; P=0.032) were correlated with DN. In the population with type 2 diabetes and chronic kidney disease, the duration of diabetes, renal characteristics, Dr, smoking history were positively correlated with DN, and FDv was negatively correlated with DN.

[0055] Table 7

[0056] The second prediction model established is as follows: (3) in, P DN It is expressed as the probability of predicting diabetic nephropathy, A1 is a constant, and A2-A6 are coefficients; Tb represents the history of diabetes, in months, Dr represents diabetic retinopathy, Dr takes the value of 1 when there is disease and 0 when there is no disease; Xy represents the smoking history, Sq represents the characteristic number of the white part of the kidney area, and FDv represents the fractal dimension of the vein.

[0057] A more specific second prediction model formula is: P DN =exp(5.359+0.009×Tb+3.710×Dr+1.261×Xy+0.371×Sq-7.241×FDv) / [1 +exp(5.359+0.009×Tb+3.710×Dr+1.261×Xy+0.371×Sq-7.241×FDv)] (4) P DN ≥0.5 indicates that the patient has DN, P DN <0.5 indicates that the patient is NDRD. Based on the binary logistic regression equation, the ROC curve was drawn, as shown in Figure 5 , AUC is 0.958, 95% CI (0.921-0.996), SE is 0.019, P Value < 0.001.

[0058] The image acquisition of eye features in the present invention is simple, and the above verification shows that the eye collateral features are of great significance for the non-invasive diagnosis of DN by combining traditional Chinese and Western medicine, and the prediction model including the white eye collateral features and the retinal vascular fractal dimension has high diagnostic efficiency. The present invention extracts features from white eye images and retinal vascular images for the first time to achieve non-invasive prediction of diabetic nephropathy.

[0059] In a specific prediction, the “hill” feature score of patient A is 1, the “blood vein dark pink” feature score is 4, the “kidney area” feature score is 2, the “lung area” feature score is 7, and the FDv is 1.628. Substituting into formula 2, P DN = exp (12.902+0.438×1+0.165×4+0.409×2-0.224×7-10.396×1.628) / [1+ exp(12.902 + 0.438×1 + 0.165×4+0.409×2 - 0.224×7 - 10.396×1.628)] =0.025; P DN <0.5 indicates that patient A is predicted to be NDRD.

[0060] In another prediction, patient B has a diabetes duration of 180 months, no smoking history, diabetic retinopathy, a "renal area" feature score of 3, and a FDv value of 1.435. Substituting into formula 4, P DN = exp (5.359+0.009×180+3.710×0+1.261×1+0.371×3 - 7.241×1.435) / [1 + exp (5.359+0.009×180+3.710×0+1.261×1+0.371×3 - 7.241×1.435)] = 0.809; P DN ≥0.5 indicates that the patient is verified as DN.

[0061] A second aspect of the present invention provides a system for implementing the above method, such as Figure 6 , including a collection module 1 and a prediction module 2, wherein the collection module 1 is used to collect patient data; the prediction module 2 is used to predict the patient data based on a prediction model to obtain the probability of diabetic nephropathy.

[0062] The system also includes a training module 3, which is used to collect a training set. The indicators of the training set include any of the following white sky features or a combination of them: the number of hills, spots, dots, fog and halo, and the color of white sky blood vessels. The indicators of the training set also include the number of white sky features of the liver, heart, spleen, lung, kidney, gallbladder, small intestine, stomach, large intestine and bladder areas, and retinal vascular features; and based on a machine learning method, the training set is trained to obtain a prediction model.

[0063] The prediction module 2 is also used to determine whether it is diabetic nephropathy based on the probability and threshold.

[0064] 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 method for predicting diabetic nephropathy based on the characteristics of sclera and retinal blood vessels. It is characterized in that The following steps are involved: Collecting patient data, wherein the patient data includes white spot characteristics or retinal blood vessel characteristics; Based on the prediction model, predicting the patient data to obtain the probability of diabetic nephropathy; The method of obtaining the prediction model includes: A training set is collected, wherein the indicators of the training set include any of the following clear sky features or their combination: the number of hills, spots, dots, fog and moon halos, and the color of the clear sky blood vessels, The indicators of the training set also include the number of white spots in the liver, heart, spleen, lung, kidney, gallbladder, small intestine, stomach, large intestine and bladder regions, as well as retinal vascular features; Based on a machine learning method, the training set is trained to obtain a prediction model.

2. The method according to claim 1, It is characterized in that The blood vein color characteristics of the white eye include: pale blood vein, dark blood vein, dark blood vein pink, dark blood vein red and blood vein pink; The retinal vascular characteristics include any of the following indicators or a combination thereof: total vascular diameter, arterial diameter, venous diameter, arteriovenous diameter ratio, total vascular fractal dimension, arterial fractal dimension, venous fractal dimension, total vascular tortuosity, arterial tortuosity, venous tortuosity, total vascular tortuosity density, arterial tortuosity density, venous tortuosity density and vascular branching angle.

3. The method according to claim 1, It is characterized in that The method of obtaining the prediction model specifically includes: A training set was collected, and the patient data of the training set was labeled as: diabetic nephropathy and non-diabetic nephropathy; Based on the methods of inter-group comparison and collinearity test, pre-selected indicators are selected from the indicators of the training set; Based on the method of single factor analysis, the modeling indicators are selected from the pre-selected indicators; Based on the binary logistic regression method, the modeling indicators of the training set are trained to obtain the prediction model.

4. The method according to claim 3, It is characterized in that The pre-selected indicators include: total number of hills, dark pink color of blood vessels, number of white clear features in kidney area, number of white clear features in lung area, number of white clear features in gallbladder area, fractal dimension of veins and density of arterial curvature; The modeling indicators include: total number of hillocks, dark pink color of blood vessels, number of white clear features in kidney area, number of white clear features in lung area and fractal dimension of veins.

5. The method according to claim 4, It is characterized in that The prediction model is expressed as: ; in, P DN A is the probability of predicting diabetic nephropathy. 1 is a constant, A 2 -A 6 are the coefficients respectively; Q represents the total number of hills, Xa represents the dark pink color of blood vessels, Sq represents the white clear characteristic number of kidney area, Fq represents the white clear characteristic number of lung area, and FDv represents the fractal dimension of veins.

6. The method according to claim 3, It is characterized in that The pre-selected indicators include: duration of diabetes, diabetic retinopathy, smoking history, body mass index, total number of verrucae, dark pink blood vessels, number of white features in the kidney area, number of white features in the lung area, number of white features in the gallbladder area, and venous fractal dimension; The modeling indicators include: diabetes course, diabetic retinopathy, smoking history, kidney area leukoplakia characteristic number and vein fractal dimension.

7. The method according to claim 6, It is characterized in that The prediction model is expressed as: ; in, P DN A is the probability of predicting diabetic nephropathy. 1 is a constant, A 2 -A 6 are coefficients respectively; Tb represents the history of diabetes, in months, Dr represents diabetic retinopathy, Xy represents the smoking history, Sq represents the characteristic number of the white part of the kidney area, and FDv represents the fractal dimension of the vein.

8. The method according to claim 1, It is characterized in that Determining whether the probability of diabetic nephropathy is greater than a threshold; If the threshold is exceeded, diabetic nephropathy is predicted; If it is less than the threshold, it is predicted to be non-diabetic nephropathy.

9. A system, It is characterized in that For implementing the method according to any one of claims 1 to 8, the system comprises a collection module and a prediction module, The acquisition module is used to collect patient data; The prediction module is used to predict the patient data based on the prediction model to obtain the probability of diabetic nephropathy.

10. The system according to claim 9, It is characterized in that Also includes training modules, The training module is used to collect a training set, and the indicators of the training set include any of the following clear sky features or their combination: the number of hills, spots, dots, fog and moon halos, and the color of the clear sky blood vessels, The indicators of the training set also include the number of white spots in the liver, heart, spleen, lung, kidney, gallbladder, small intestine, stomach, large intestine and bladder regions, as well as retinal vascular features; Based on a machine learning method, the training set is trained to obtain a prediction model.