A method for constructing a non-invasive screening model for non-alcoholic fatty hepatitis, a prediction system, a device and a storage medium

By constructing a non-alcoholic steatohepatitis model based on serum, imaging and pathological data, the existing problems of low accuracy and cost of diagnosis of steatohepatitis are solved, and non-invasive and efficient steatohepatitis screening and monitoring are achieved, reducing the number of non-invasive and efficient steatohepatitis screening and monitoring. medical burden.

CN113160994BActive Publication Date: 2024-12-06NANJING DRUM TOWER HOSPITAL +1
View PDF 1 Cites 0 Cited by

Patent Information

Application Number
CN202011634864.0
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2020-12-31
Publication Date
2024-12-06
Estimated Expiration
2040-12-31

AI Technical Summary

Technical Problem

The existing diagnostic methods for steatohepatitis are low in accuracy and expensive, and there is a problem of insufficient non-invasive screening methods.

Method used

By collecting serum, imaging and pathological data of patients with steatohepatitis, using multiple logistic regression and support vector machine models, combining B-ultrasound feature extraction and screening, a non-invasive screening non-alcoholic steatohepatitis model was constructed to achieve non-alcoholic fattyness in non-alcoholic fattyness in non-alcoholic fattyness in combination with B-ultrasound feature extraction and screening. Predictions of hepatitis.

Benefits of technology

It improves the accuracy and efficiency of steatohepatitis diagnosis, reduces the financial burden on patients and the time cost of doctors, avoids the pain and complications of invasive examinations, and reduces the medical burden.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN113160994B_ABST
    Figure CN113160994B_ABST
Patent Text Reader

Abstract

A method for constructing a non-invasive screening model for non-alcoholic fatty liver disease, a prediction system, a device and a storage medium, belonging to the field of fatty liver detection. Existing methods for fatty liver disease have disadvantages such as low accuracy and high cost. A method for constructing a non-invasive screening model for non-alcoholic fatty liver disease, collecting serum, imaging and pathological data of patients with fatty liver disease; preprocessing serum feature data; statistical screening of serum features; extraction and screening of B-ultrasound features; using support vector machine SVM to integrate effective B-ultrasound features; combining the acquired serum index features with the R-NASH features of B-ultrasound, and using the multivariate logistic regression method to obtain a non-alcoholic fatty liver disease diagnosis model; through model comparison and performance analysis, the model threshold range for evaluating NASH is confirmed according to the maximum Youden index of the ROC curve, so as to obtain a final model with fixed parameters for prediction.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the construction of a model for screening fatty hepatitis, and in particular to a construction method, prediction system, device and storage medium for non-invasive screening of non-alcoholic fatty hepatitis model. Background Art

[0002] At present, the diagnosis of fatty liver hepatitis needs to be carried out through methods such as B-ultrasound, CT plain scan liver density or blood tests. B-ultrasound examination is an important and practical means of diagnosing fatty liver hepatitis, and its diagnostic accuracy is about 70%. CT plain scan liver density is generally reduced, and the liver / spleen CT plain scan density ratio ≤1 can confirm the diagnosis of fatty liver hepatitis. The degree of fatty liver hepatitis can be judged according to the liver / spleen CT plain scan density ratio. Its accuracy is slightly better than B-ultrasound, but it has disadvantages such as high cost and radioactivity. Summary of the invention

[0003] In order to solve the problems of low accuracy and high cost in existing methods for fatty liver disease, the present invention provides a method for constructing a non-invasive screening model for non-alcoholic fatty liver disease, a prediction system, a device and a storage medium.

[0004] A method for constructing a non-invasive screening model for non-alcoholic fatty liver disease, wherein the detection method is implemented by the following steps:

[0005] Step 1: Collect serum, imaging and pathological data of patients with steatohepatitis;

[0006] Collect data, including basic information of patients, NASH and non-NASH pathological biopsy data, liver B-ultrasound images and serum laboratory test data; and ultrasound images of the liver in BMP format of the corresponding patients;

[0007] in,

[0008] The Chinese meaning of NASH is non-alcoholic fatty hepatitis;

[0009] Basic information includes age, gender, height, weight, BMI calculation, drinking history, whether there is diabetes, and hepatitis history; BMI means body mass index, also known as body mass index, also known as weight, in English, Body Mass Index, referred to as BMI;

[0010] Serum laboratory test data included white blood cell count, platelet count, alanine aminotransferase, aspartate aminotransferase, glutamyl transpeptidase, total bilirubin, direct bilirubin, clotting time, alkaline phosphatase, albumin, blood cholesterol, international normalized ratio of coagulation, blood ferritin, fasting blood glucose, uric acid, blood lipids, fasting insulin, and HOMA index;

[0011] Step 2: Preprocessing of serum characteristic data;

[0012] Excluding NASH and non-NASH pathological biopsy data, the remaining serum variables were standardized as follows:

[0013] (1) Arrange each numerical variable from small to large, and consider those less than the 2.5th percentile or greater than the 97.5th percentile as outliers and change them to 2.5th and 97.5th percentiles, respectively;

[0014] (2) Normalize all numerical variables. The formula is: n normalized=(X n –X min ) / (X max –X min ), where X n represents any numeric variable, X n normalized represents a numeric variable X n The normalized value of X max Indicates the maximum value of a numeric variable, X min Represents the minimum value of a numeric variable;

[0015] Step 3: Statistical screening of serum characteristics;

[0016] A portion of the collected data is selected as a training group; then, in the training group, the indicators with significant correlation are screened out through Spearman correlation analysis; then, the serum indicators associated with NASH are statistically screened out through forward conditional multivariate logistic regression; wherein, the serum indicators are aspartate aminotransferase, platelet count, blood lipids, BMI calculation, and uric acid;

[0017] Step 4: B-ultrasound feature extraction and screening;

[0018] The B-ultrasound feature extraction refers to calculating and extracting the grayscale texture features of liver tissue; wherein the grayscale texture features include grayscale co-occurrence matrix, grayscale walking distance matrix, intensity histogram and invariant moment;

[0019] The screening operation includes the following three steps:

[0020] (1) The intra-group and inter-group observer consistency test exceeded 0.8;

[0021] (2) The screening results obtained in step (1) are screened as follows: a preliminary screening operation is performed using the variance threshold method, and the threshold involved is set to 1.0;

[0022] (3) The screening results obtained in step (2) are screened as follows: the lasso regression method is used for final screening; specifically:

[0023] First, the Lasso regression model is used to screen out effective B-ultrasound features. The cost function of the Lasso regression model is:

[0024]

[0025] Among them, x ij represents the jth B-ultrasound characteristic variable of the i-th patient data, y i is the response variable of the i-th patient data, q represents the total number of B-ultrasound feature variables, m is the total number of patient data, λ is the penalty function, For the best result;

[0026] Then, the non-zero coefficient variables screened by the lasso regression method are used as effective B-ultrasound features for the final modeling;

[0027] Step 5: Use support vector machine (SVM) to integrate effective B-ultrasound features;

[0028] The support vector machine (SVM) model is used to predict the features of the effective B-ultrasound image group to obtain the predicted value R-NASH; wherein the steps of predicting by the SVM model include the following steps:

[0029] (1) Select Gaussian kernel function: In space, x and z are any two points in space, and σ is the width parameter of the function;

[0030] (2) Constructing support vector machine model: where α i and b are the coefficients to be determined; α i It is obtained by solving the following dual problem:

[0031]

[0032]

[0033] 0≤α i ,i=1,2,K,N

[0034] y i For input x i Corresponding R-NASH value y i ;

[0035] i represents the i-th, j represents the j-th, a j 、a i For the jth, i-th sample, the required parameter x i 、x j represents the B-ultrasound input feature of the i-th and j-th samples; y j, yi represent the integrated B-ultrasound features R-NASH corresponding to the jth and ith samples, and N represents the number of samples; the optimal solution is obtained: Select α * A component of Satisfy the conditions Get the optimal:

[0036] (3) Using the obtained SVM model, the characteristics of the B-ultrasound image group are predicted to obtain the predicted value R-NASH:

[0037]

[0038] Among them, x is the vector composed of the features of the effective B-ultrasound image group;

[0039] Step 6: The acquired serum index characteristics and B-ultrasound R-NASH characteristics are combined, and a multivariate logistic regression method is used to obtain a non-alcoholic fatty liver disease diagnostic model, including the following steps:

[0040] (1) Assume x1,x2,K,x n is the predictive value of R-NASH and other serum index characteristics, g is the fibrosis index, and the regression model is as follows:

[0041]

[0042] z=h θ (x) = θ0 + θ1x1 + θ2x2 + Λ + θ n x n

[0043] Among them, θ0, θ1, θ2Λθ n is the model parameter of multivariate logistic regression, n is the number of variables, and the features used include: R-NASH, PLT, AST, BMI, Lipid, and Uric acid;

[0044] (2) Obtain the model parameters and obtain the diagnostic model as follows:

[0045] Y NASH =2.145×R-NASH–0.306×PLT+0.267×AST+1.152×BMI+1.185×Lipid+0.114×Uric acid–3.375;

[0046] Step 7: Model confirmation;

[0047] Through model comparison and performance analysis, the model threshold range for evaluating NASH was confirmed according to the maximum Youden index of the ROC curve, so as to obtain the final model with fixed parameters for prediction.

[0048] Beneficial effects:

[0049] The present invention is a model for screening non-alcoholic fatty hepatitis, which uses the serum, imaging and pathological data of multiple fatty hepatitis patients to form a non-alcoholic fatty hepatitis related data prediction model. The non-alcoholic fatty hepatitis screening model based on multiple clinical indicators not only complements the advantages and disadvantages of serology and imaging, improves the prediction speed and accuracy of the model, and can provide a reference for clinical liver fibrosis diagnosis. The initial screening of NASH can be completed quickly without liver puncture, so that the degree of liver lesions in patients can be monitored efficiently at any time and corresponding clinical intervention measures can be formulated. Non-invasive prediction not only reduces the economic burden of patients and the time cost of doctors, but also avoids the pain and complications caused by unnecessary patients requiring invasive examinations, thereby reducing the medical burden. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Attached Figure 1 is a flow chart of the method of the present invention;

[0051] Attached Figure 2 It is a schematic diagram of the comparison between the effect evaluation using the ROC curve and the design model of the present invention;

[0052] Attached Figure 3 This is a schematic diagram of circling liver tissue and calculating and extracting grayscale texture features;

[0053] Attached Figure 4 It is a diagram of the gray-level co-occurrence matrix calculation method;

[0054] Attached Figure 5 It is a diagram of the grayscale walking run matrix calculation method;

[0055] Attached Figure 6 It is a diagram of the calculation method of the intensity histogram related features;

[0056] Attached Figure 7 It is a diagram of the method of calculating invariant moments. DETAILED DESCRIPTION Specific implementation method one:

[0058] A method for constructing a non-invasive screening model for non-alcoholic fatty liver disease in this embodiment is as follows: Figure 1 As shown, the detection method is implemented by the following steps:

[0059] Step 1: Collect serum, imaging and pathological data of patients with steatohepatitis;

[0060] The data of the Hepatobiliary and Pancreatic Center of Nanjing Drum Tower Hospital are collected, including the basic information of patients, NASH (non-alcoholic steatohepatitis) and non-NASH pathological biopsy data, liver B-ultrasound images and serum laboratory test data; and the ultrasound image of the liver of the corresponding patient in BMP format. For example, the ultrasound image of the corresponding patient is copied in BMP format by the Ultrasound Department of Nanjing Drum Tower Hospital.

[0061] in,

[0062] The Chinese meaning of NASH is non-alcoholic fatty hepatitis;

[0063] Basic information includes age, gender, height, weight, BMI calculation, drinking history, whether there is diabetes, and hepatitis history; BMI means body mass index, also known as body mass index, also known as weight, in English, Body Mass Index, abbreviated as BMI;

[0064] Serum laboratory test data included white blood cell count, platelet count, alanine aminotransferase, aspartate aminotransferase, glutamyl transpeptidase, total bilirubin, direct bilirubin, clotting time, alkaline phosphatase, albumin, blood cholesterol, international normalized ratio of coagulation, blood ferritin, fasting blood glucose, uric acid, blood lipids, fasting insulin, and HOMA index;

[0065] Step 2: Preprocessing of serum characteristic data;

[0066] Except for NASH and non-NASH pathological biopsy data, which were not preprocessed, serum variables were uniformly standardized as follows:

[0067] (1) Arrange each numerical variable from small to large, and consider those less than the 2.5th percentile or greater than the 97.5th percentile as outliers and change them to 2.5th and 97.5th percentiles, respectively;

[0068] (2) Normalize all numerical variables. The formula is: n normalized=(X n –X min ) / (X max –X min ), where X n represents any numeric variable, X n normalized represents a numeric variable X n The normalized value of X max Indicates the maximum value of a numeric variable, X min Represents the minimum value of a numeric variable;

[0069] Step 3: Statistical screening of serum characteristics;

[0070] A part of the collected data is selected as the training group; for example, the cases of the Hepatobiliary and Pancreatic Center of Nanjing Drum Tower Hospital are used as the training group. Afterwards, the indicators with significant correlation are screened out in the training group through Spearman correlation analysis; then, the final meaningful serum indicators that are most related to NASH are obtained through statistical screening through forward conditional multivariate logistic regression; wherein, the serum indicators are aspartate aminotransferase, platelet count, blood lipids, BMI calculation, and uric acid; the serum indicators are aspartate aminotransferase referred to as AST, platelet count referred to as PLT, blood lipids referred to as lipid, BMI calculation, and uric acid referred to as Uric acid;

[0071] Step 4: B-ultrasound feature extraction and screening;

[0072] The B-ultrasound feature extraction refers to calculating and extracting the grayscale texture features of liver tissue; wherein the grayscale texture features include grayscale co-occurrence matrix, grayscale walking distance matrix, intensity histogram and invariant moment;

[0073] The screening operation includes the following three steps:

[0074] (1) The intra- and inter-observer consistency test (ICC) exceeded 0.8;

[0075] (2) The variance threshold method was used for initial screening, and the threshold involved was set to 1.0;

[0076] (3) The lasso regression method is used for final screening; specifically:

[0077] First, the Lasso regression model is used to screen out effective B-ultrasound features, thereby eliminating features with little effect. The cost function of the Lasso regression model is:

[0078]

[0079] Among them, x ij represents the jth B-ultrasound characteristic variable of the i-th patient data, y i is the response variable of the i-th patient data, q represents the total number of B-ultrasound feature variables, m is the total number of patient data, λ is the penalty function, For the best result;

[0080] Then, the non-zero coefficient variables screened by the lasso regression method are used as effective B-ultrasound features for the final modeling;

[0081] Step 5: Use support vector machine (SVM) to integrate effective B-ultrasound features;

[0082] The support vector machine (SVM) model is used to predict the features of the effective B-ultrasound image group to obtain the predicted value R-NASH; wherein the SVM model prediction step includes the following steps:

[0083] (1) Select Gaussian kernel function: In space, x and z are any two points in space, and σ is the width parameter of the function;

[0084] (2) Constructing support vector machine model: where α i and b are the coefficients to be determined; α i It is obtained by solving the following dual problem:

[0085]

[0086]

[0087] 0≤α i ,i=1,2,K,N

[0088] y i For input x i Corresponding R-NASH value y i ;

[0089] i represents the i-th, j represents the j-th, a j 、a i For the jth, i-th sample, the required parameter x i 、x j represents the B-ultrasound input feature of the i-th and j-th samples; y j , yi represents the integrated B-ultrasound feature R-NASH corresponding to the jth, i-th sample, and N represents the number of samples; the optimal solution is obtained: Select α * A component of Satisfy the conditions Get the optimal:

[0090] (3) Using the obtained SVM model, the characteristics of the B-ultrasound image group are predicted to obtain the predicted value R-NASH:

[0091]

[0092] Among them, x is the vector composed of the features of the effective B-ultrasound image group;

[0093] Step 6: The acquired serum index characteristics and the R-NASH characteristics of B-ultrasound are combined, and a multivariate logistic regression method is used to obtain a diagnostic model for non-alcoholic fatty liver disease; wherein the multivariate logistic regression method comprises the following steps:

[0094] (1) Assume x1,x2,K,x n is the predictive value of R-NASH and other serum index characteristics, g is the fibrosis index, and the regression model is as follows:

[0095]

[0096] z=h θ (x) = θ0 + θ1x1 + θ2x2 + Λ + θ n x n

[0097] Among them, θ0, θ1, θ2Λθ n is the model parameter of multivariate logistic regression, n is the number of variables, and the features used include: R-NASH, PLT, AST, BMI, Lipid, and Uric acid;

[0098] (2) Obtain the model parameters and obtain the diagnostic model as follows:

[0099] Y NASH =2.145×R-NASH–0.306×PLT+0.267×AST+1.152×BMI+1.185×Lipid+0.114×Uric acid–3.375.

[0100] Step 7: Model confirmation;

[0101] Through model comparison and performance analysis, the model threshold range for evaluating NASH was confirmed according to the maximum Youden index (sensitivity plus specificity - 1) of the ROC curve. The reasonable parameter range of the final model was obtained according to the determined threshold, thereby obtaining a model with fixed final parameters for prediction; and it was applied to clinical verification.

[0102] The NASH preliminary assessment model based on multiple clinical data can quickly complete the initial screening of NASH without liver puncture, so as to efficiently monitor the degree of liver lesions in patients at any time and formulate corresponding clinical intervention measures. Non-invasive diagnosis not only reduces the economic burden of patients and the time cost of doctors, but also avoids the pain and complications caused by invasive examinations for unnecessary patients, thus reducing the medical burden. Specific implementation method 2:

[0104] Different from the first specific embodiment, the method for constructing a non-invasive screening model for non-alcoholic fatty liver disease in this embodiment further adds a model verification and comparison step between step six and step seven, specifically:

[0105] The model of the present invention was compared with the existing model by ultrasound FLI test. ROC curve was used to evaluate the effect and compare the models. Figure 2 shown.

[0106] The horizontal axis of the ROC curve represents specificity, and the vertical axis represents sensitivity. Generally, the higher the curve, the better the model classification ability. The solid line curve is a newly created model, and the dotted line curve is an existing model. When the two are compared together, the superior and inferior abilities can be clearly distinguished. Specific implementation method three:

[0108] Different from the first or second specific embodiments, in the method for constructing a non-invasive screening model for non-alcoholic fatty liver disease in this embodiment, in step 4, the feature used in the step of B-ultrasound feature extraction is: (B-ultrasound feature in the raw material)

[0109] B-ultrasound feature extraction:

[0110] The B-ultrasound image was read by screening software (liverradiomics), and then the image was standardized through the grayscale stretching step, and then the liver grayscale texture features were calculated and extracted; the grayscale texture features include grayscale co-occurrence matrix, grayscale walk-length matrix, intensity histogram and invariant moment; Figure 3 As shown;

[0111] Among them, the algorithm for extracting features by the screening software is as follows:

[0112] The method for calculating the gray level co-occurrence matrix is ​​to describe the texture according to the spatial correlation characteristics of the gray level, which is a common method and includes:

[0113] Autocorrelation feature, the full name of which is Autocorrelation: refers to the correlation between the expected values ​​of random error terms, which is used to evaluate the clarity of the image;

[0114] Correlation: It measures the similarity of the elements of the spatial grayscale co-occurrence matrix in the row or column direction. Therefore, the size of the correlation value reflects the local grayscale correlation in the image. When the values ​​of the matrix elements are uniformly equal, the correlation value is large; on the contrary, if the matrix pixel values ​​are very different, the correlation value is small. If there is horizontal texture in the image, the COR of the horizontal matrix is ​​greater than the COR value of the remaining matrices; COR stands for Correlation, which means correlation in Chinese;

[0115] Cluster prominence: The prominence of an object in the image. The greater the contrast between the lines and patterns, the greater the prom value. Prom stands for Cluster prominence, which means cluster prominence in Chinese.

[0116] Cluster shadow, the full name of English is Clustershade: the size of the shadow (shade), which may be related to the quality of the picture, and more intuitively related to the degree of wrinkles in the clothing. The smoother the clothes are, the smaller the shade value is; it measures the skewness of the matrix and measures the concept of uniformity; when the value is high, the image is asymmetrical; the Chinese meaning of shade is shadow;

[0117] Dissimilarity: When calculating contrast, the weight grows exponentially with the distance between the matrix element and the diagonal. If it grows linearly instead, dissimilarity is obtained.

[0118] Energy is the sum of the squares of the grayscale co-occurrence matrix element values, so it is also called energy, which reflects the uniformity of the grayscale distribution and the coarseness of the texture. If all the values ​​of the co-occurrence matrix are equal, the ASM value is small; on the contrary, if some of the values ​​are large and other values ​​are small, the ASM value is large. When the elements in the co-occurrence matrix are concentrated, the ASM value is large. A large ASM value indicates a more uniform and regularly changing texture pattern.

[0119] Homogeneity: measures the local uniformity of an image, with lower values ​​for non-uniform images and higher values ​​for uniform images. In contrast to contrast or dissimilarity, the weight of homogeneity decreases exponentially as the element value moves away from the diagonal.

[0120] Maximum probability, the full English name is Maximumprobability: it represents the texture feature that appears most frequently in the image.

[0121] Calculation method: Figure 4 As shown in the figure, j represents the gray value of pixel y, i represents the gray value of pixel x, p represents probability, and M and N represent M×N matrices. Specific implementation method four:

[0123] Different from the third specific embodiment, the method for constructing a non-invasive screening model for non-alcoholic fatty liver disease in this embodiment is as follows: Figure 5As shown in the figure, the grayscale run matrix is ​​a matrix composed of the lengths of grayscale value runs. It is only a measure and statistics of image pixel information. In actual use, the generated grayscale run matrix is ​​calculated to obtain image feature information based on the grayscale co-occurrence matrix. Including short-term emphasis (SRE), long-term emphasis (LRE), run length non-uniformity (RLN), gray level non-uniformity (GLN), run percentage (RP), low gray level run emphasis (LGRE), high gray level run emphasis (HGRE), short-term low gray level emphasis (SRLGE), short-term high gray level emphasis (SRHGE), long-term low gray level emphasis (LRLGE), long-term high gray level emphasis (LRHGE).

[0124] Figure 5 In the equation, point (i, j) represents the number of consecutive occurrences of a pixel with gray level i in a certain direction of the image j times, p represents the probability, M represents the number of gray levels on the image; N represents the number of different runs on the image; n r Indicates the number of pixels on the image; Specific implementation method five:

[0126] Different from the third or fourth specific implementation, in the method for constructing a non-invasive screening model for non-alcoholic fatty liver disease in this implementation, the method for calculating the intensity histogram is as follows: Figure 6 As shown in the figure, the relevant features of the intensity histogram are the first-order features calculated based on the intensity histogram. The first-order features are the simplest first-order features, including Mean: the statistical mean of the intensity histogram, Variance: the statistical variance of the intensity histogram, Entropy: the statistical entropy value of the intensity histogram, Energy: the sum of the statistical energy of the intensity histogram; Kurtosis: the characteristic number that characterizes the height of the peak of the probability density distribution curve at the average value, describing whether the peak of the distribution composed of the sample is abrupt or flat. Intuitively, the kurtosis reflects the sharpness of the peak. The kurtosis of the sample is a statistic compared with the normal distribution. The kurtosis of the time series x is calculated. The kurtosis is used to measure the deviation of x from a certain distribution. The kurtosis of the normal distribution is 3. If the kurtosis is greater than 3, the shape of the peak is sharper and steeper than the normal distribution peak. Vice versa; in statistics, kurtosis measures the peak state of the probability distribution of a real random variable. A high kurtosis means that the increase in variance is caused by low-frequency extreme differences that are greater or less than the mean. Skewness: describes the symmetry of the distribution composed of samples.

[0127] Among them, the situation describing the symmetry of the distribution composed of Sample is to calculate the skewness of the time series x, and the skewness is used to measure the symmetry of x; if the skewness is negative, the dispersion on the left side of the mean of x is stronger than that on the right side; if the skewness is positive, the dispersion on the left side of the mean of x is weaker than that on the right side; for normal distribution (or strictly symmetric distribution), the skewness is equal to O. Specific implementation method six:

[0129] Different from the fifth specific embodiment, in the method for constructing a non-invasive screening model for non-alcoholic fatty liver disease in this embodiment, the method for calculating the invariant moment is as follows: Figure 7 As shown, moment is a concept in probability and statistics, and is a digital characteristic of a random variable. The first-order origin moment is the expectation, the first-order central moment μ1 is 0, and the second-order central moment μ2 is the variance Var(X) of X. In statistics, moments higher than the fourth order are rarely used. The third-order central moment μ3 is used to measure whether the distribution is biased, and the fourth-order central moment μ4 is used to measure how steep the distribution (or density) is around the mean.

[0130] Seven invariant moment groups (Φ1Φ7) are derived using the second-order and third-order canonical central moments. The invariant moment groups (Φ1Φ7) remain unchanged when the image is translated, rotated and scaled. Specific implementation method seven:

[0132] A prediction system for a method for constructing a non-invasive screening model for non-alcoholic fatty liver disease in this embodiment includes:

[0133] A data acquisition module is used to pre-process the serum, imaging and pathological data of patients with fatty liver disease;

[0134] The serum characteristic data preprocessing module removes NASH and non-NASH pathological biopsy data and performs the following standardized processing on the remaining serum variables; the specific processing method is as follows:

[0135] (1) Arrange each numerical variable from small to large, and consider those less than the 2.5th percentile or greater than the 97.5th percentile as outliers and change them to 2.5th and 97.5th percentiles, respectively;

[0136] (2) Normalize all numerical variables. The formula is: n normalized=(X n –X min ) / (X max –X min ), where X n represents any numeric variable, X n normalized represents a numeric variable X n The normalized value of X maxIndicates the maximum value of a numeric variable, X min Represents the minimum value of a numeric variable;

[0137] Serum characteristic statistical screening, used for serum characteristic statistical screening; specifically, a part of the collected data is selected as a training group; then, in the training group, the indicators with significant correlation are screened out through Spearman correlation analysis; then, the serum indicators associated with NASH are statistically screened out through forward conditional multivariate logistic regression; wherein, the serum indicators are aspartate aminotransferase, platelet count, blood lipids, BMI calculation, and uric acid;

[0138] The B-ultrasound feature extraction and screening module is used to extract and screen B-ultrasound features. The non-zero coefficient variables screened out by the lasso regression method are used as the effective B-ultrasound features for the final modeling, so as to calculate and extract the grayscale texture features of liver tissue; wherein, the Lasso regression model is used to screen out effective B-ultrasound features, and the cost function of the Lasso regression model is:

[0139]

[0140] Among them, x ij represents the jth B-ultrasound characteristic variable of the i-th patient data, y i is the response variable of the i-th patient data, q represents the total number of B-ultrasound feature variables, m is the total number of patient data, λ is the penalty function, For the best result;

[0141] A feature integration module is used to predict the features of the effective B-ultrasound image group using a support vector machine (SVM) model, obtain a predicted value R-NASH, and integrate effective B-ultrasound features;

[0142] The prediction model building module is used to combine the acquired serum index characteristics and the R-NASH characteristics of B-ultrasound, and use the multivariate logistic regression method to obtain a non-alcoholic fatty liver disease screening model;

[0143] The prediction model confirmation module is equivalent to confirming the model threshold range for evaluating NASH, so as to obtain a model with fixed parameters; wherein, the model threshold range for evaluating NASH is confirmed by model comparison and performance analysis based on the maximum Youden index of the ROC curve, and is ultimately used for prediction. Specific implementation eight:

[0145] A detection device for a method for constructing a non-invasive screening model of non-alcoholic fatty liver disease in this embodiment includes:

[0146] processor;

[0147] a memory storing executable instructions of the processor;

[0148] Wherein, the processor is configured to execute the steps of the method for constructing a non-invasive screening model for non-alcoholic fatty liver disease by executing the executable instructions. Specific implementation method nine:

[0150] A computer-readable storage medium of this embodiment is used to store a program, and when the program is executed, the steps of the method for constructing a non-invasive screening model for non-alcoholic fatty liver disease are implemented.

Claims

1. A method for constructing a non-invasive screening model for non-alcoholic fatty liver disease. Its characteristics are: The construction method is achieved by the following steps: Step 1: Collect serum, imaging and pathological data of patients with steatohepatitis; Collect data, including basic information of patients, NASH and non-NASH pathological biopsy data, liver B-ultrasound images and serum laboratory test data; and ultrasound images of the liver in BMP format of the corresponding patients; in, The Chinese meaning of NASH is non-alcoholic fatty hepatitis; Basic information includes age, gender, height, weight, BMI calculation, drinking history, whether there is diabetes, and hepatitis history; BMI means body mass index, also known as body mass index, also known as weight, in English, Body Mass Index, referred to as BMI; Serum laboratory test data included white blood cell count, platelet count, alanine aminotransferase, aspartate aminotransferase, glutamyl transpeptidase, total bilirubin, direct bilirubin, clotting time, alkaline phosphatase, albumin, blood cholesterol, international normalized ratio of coagulation, blood ferritin, fasting blood glucose, uric acid, blood lipids, fasting insulin, and HOMA index; Step 2: Preprocessing of serum characteristic data; Excluding NASH and non-NASH pathological biopsy data, the remaining serum variables were standardized as follows: (1) Arrange each numerical variable from small to large, and consider those less than the 2.5th percentile or greater than the 97.5th percentile as outliers and change them to 2.5th and 97.5th percentiles, respectively; (2) Normalize all numerical variables. The formula is: n normalized=(X n –X min ) / (X max –X min ), where X n represents any numeric variable, X n normalized represents a numeric variable X n The normalized value of X max Indicates the maximum value of a numeric variable, X min Represents the minimum value of a numeric variable; Step 3: Statistical screening of serum characteristics; A portion of the collected data is selected as a training group; then, in the training group, the indicators with significant correlation are screened out through Spearman correlation analysis; then, the serum indicators associated with NASH are statistically screened out through forward conditional multivariate logistic regression; wherein, the serum indicators are aspartate aminotransferase, platelet count, blood lipids, BMI calculation, and uric acid; Step 4: B-ultrasound feature extraction and screening; The B-ultrasound feature extraction refers to calculating and extracting the grayscale texture features of liver tissue; wherein the grayscale texture features include grayscale co-occurrence matrix, grayscale walking distance matrix, intensity histogram and invariant moment; The screening operation includes the following three steps: (1) The intra-group and inter-group observer consistency test exceeded 0.8; (2) The screening results obtained in step (1) are screened as follows: a preliminary screening operation is performed using the variance threshold method, and the threshold involved is set to 1.0; (3) The screening results obtained in step (2) are screened as follows: the lasso regression method is used for final screening; specifically: First, the Lasso regression model is used to screen out effective B-ultrasound features. The cost function of the Lasso regression model is: Among them, x ij represents the jth B-ultrasound characteristic variable of the i-th patient data, y i is the response variable of the i-th patient data, q represents the total number of B-ultrasound feature variables, m is the total number of patient data, λ is the penalty function, For the best result; Then, the non-zero coefficient variables screened by the lasso regression method are used as effective B-ultrasound features for the final modeling; Step 5: Use support vector machine (SVM) to integrate effective B-ultrasound features; The support vector machine (SVM) model is used to predict the features of the effective B-ultrasound image group to obtain the predicted value R-NASH; wherein the steps of predicting by the SVM model include the following steps: (1) Select Gaussian kernel function: In space, x and z are any two points in space, and σ is the width parameter of the function; (2) Constructing the support vector machine model: where α i and b are the coefficients to be determined; α i It is obtained by solving the following dual problem: 0≤α i ,i=1,2,...,N y i For input x i Corresponding R-NASH value y i ; i represents the i-th, j represents the j-th, a j 、a i For the jth, i-th sample, the required parameter x i 、x j represents the B-ultrasound input features of the i-th and j-th samples; y j , yi represent the integrated B-ultrasound features R-NASH corresponding to the jth and ith samples, and N represents the number of samples; the optimal solution is obtained: Select α * A component of Satisfy the conditions Get the optimal: (3) Using the obtained SVM model, the characteristics of the B-ultrasound image group are predicted to obtain the predicted value R-NASH: Among them, x is the vector composed of the features of the effective B-ultrasound image group; Step 6: The acquired serum index characteristics and B-ultrasound R-NASH characteristics are combined, and a multivariate logistic regression method is used to obtain a non-alcoholic fatty liver disease diagnostic model, including the following steps: (1) Assume x 1 ,x 2 ,...,x n is the predictive value of R-NASH and other serum index characteristics, g is the fibrosis index, and the regression model is as follows: z=h θ (x)=θ 0 +θ 1 x 1 +θ 2 x 2 +…+θ n x n Among them, θ 0 ,θ 1 ,θ 2 …θ n is the model parameter of multivariate logistic regression, n is the number of variables, and the features used include: R-NASH, PLT, AST, BMI, Lipid, and Uric acid; (2) Obtain the model parameters and obtain the diagnostic model as follows: Y NASH =2.145×R-NASH–0.306×PLT+0.267×AST+1.152×BMI+1.185×Lipid+ 0.114 × Uric acid–3.375; Step 7: Model confirmation; Through model comparison and performance analysis, the model threshold range for evaluating NASH was confirmed according to the maximum Youden index of the ROC curve, so as to obtain the final model with fixed parameters for prediction.

2. A method for constructing a non-invasive screening model for non-alcoholic fatty liver disease according to claim 1, Its characteristics are: In step 4, the features used in the step of B-ultrasound feature extraction are: B-ultrasound feature extraction: The B-ultrasound image was read by screening software, and then the image was standardized through the grayscale stretching step, and then the grayscale texture features of the liver were calculated and extracted; the grayscale texture features included grayscale co-occurrence matrix, grayscale walk-length matrix, intensity histogram and invariant moment; Among them, the algorithm for extracting features by the screening software is as follows: The calculation method for extracting liver grayscale texture features, wherein the grayscale texture features include grayscale co-occurrence matrix, grayscale walk run matrix, intensity histogram and invariant moment, is to describe the texture according to the spatial correlation characteristics of grayscale, including: Autocorrelation feature: refers to the correlation between the expected values ​​of random error terms, which is used to evaluate the clarity of the image; Correlation: It measures the similarity of the elements of the spatial grayscale co-occurrence matrix in the row or column direction. Therefore, the size of the correlation value reflects the local grayscale correlation in the image. When the matrix element values ​​are uniformly equal, the correlation value is large. On the contrary, if the matrix pixel values ​​differ greatly, the correlation value is small. If there is horizontal texture in the image, the COR of the horizontal matrix is ​​greater than the COR value of the rest of the matrix. COR stands for Correlation, which means correlation in Chinese. Cluster prominence: The prominence of an object in the image. The greater the contrast between the lines and patterns, the greater the prom value. Prom stands for Cluster prominence, which means cluster prominence in Chinese. Cluster shadow, the full name in English is Clustershade: the size of the shadow is related to the quality of the picture, and more intuitively it is related to the degree of wrinkles in the clothing. The smoother it is, the smaller the shade value is; it measures the skewness of the matrix and measures the concept of uniformity; when the value is high, the image is asymmetrical; shade means shadow in Chinese; Dissimilarity: When calculating contrast, the weight grows exponentially with the distance of the matrix element from the diagonal. If it grows linearly instead, dissimilarity is obtained; Energy: It is the sum of the squares of the grayscale co-occurrence matrix element values, so it is also called energy, which reflects the uniformity of the image grayscale distribution and the coarseness of the texture; if all the values ​​of the co-occurrence matrix are equal, the ASM value is small; on the contrary, if some values ​​are large and other values ​​are small, the ASM value is large; when the elements in the co-occurrence matrix are concentrated, the ASM value is large; a large ASM value indicates a more uniform and regularly changing texture pattern; Homogeneity: Measures the local uniformity of an image, with lower values ​​for non-uniform images and higher values ​​for uniform images; in contrast to contrast or dissimilarity, the weight of homogeneity decreases exponentially as the element value moves away from the diagonal. Maximum probability, the full English name is Maximumprobability: it represents the texture feature that appears most frequently in the image.

3. A method for constructing a non-invasive screening model for non-alcoholic fatty liver disease according to claim 2, Its characteristics are: The method for calculating the grayscale run length matrix is ​​as follows: the grayscale run length matrix is ​​a matrix composed of the lengths of grayscale value runs, which is only a measure and statistics of image pixel information. In actual use, it is also necessary to calculate the generated grayscale run length matrix to obtain image feature information based on the grayscale co-occurrence matrix; Includes short-term emphasis, long-term emphasis, run length non-uniformity, gray level non-uniformity, run percentage, low gray level run emphasis, high gray level run emphasis, short-term low gray level emphasis, short-term high gray level emphasis, long-term low gray level emphasis, and long-term high gray level emphasis.

4. A method for constructing a non-invasive screening model for non-alcoholic fatty liver disease according to claim 2 or 3, Its characteristics are: The calculation extracts the liver grayscale texture features; The grayscale texture features include the grayscale co-occurrence matrix, the grayscale walking run matrix, the intensity histogram and the invariant moment. The intensity histogram is calculated in the following way: the relevant features of the intensity histogram are first-order features calculated based on the intensity histogram, including the statistical mean of the intensity histogram, the statistical variance of the intensity histogram, the statistical entropy value of the intensity histogram, and the statistical energy sum of the intensity histogram; the characteristic number that characterizes the peak height of the probability density distribution curve at the average value, describes whether the peak of the distribution composed of the sample is abrupt or flat, and describes the symmetry of the distribution composed of the sample; Among them, the situation describing the symmetry of the distribution composed of Sample is to calculate the skewness of the time series x, and the skewness is used to measure the symmetry of x; if the skewness is negative, the dispersion on the left side of the mean of x is stronger than the right side; if the skewness is positive, the dispersion on the left side of the mean of x is weaker than the right side; for a normal distribution, the skewness is equal to 0.

5. A method for constructing a non-invasive screening model for non-alcoholic fatty liver disease according to claim 2 or 3, Its characteristics are: The calculation extracts the liver grayscale texture features; Grayscale texture features include grayscale co-occurrence matrix, grayscale walk run matrix, intensity histogram and invariant moment. The calculation method of invariant moment is: Using the second-order and third-order central moments, seven invariant moment groups (Φ1Φ7) are derived; Among them, moment is a digital characteristic of a random variable, the first-order origin moment is the expectation, the first-order central moment μ1 is 0, the second-order central moment μ2 is the variance Var(X) of X, the third-order central moment μ3 is used to measure whether the distribution is biased, and the fourth-order central moment μ4 is used to measure the steepness of the distribution around the mean.

6. A prediction system applied to a method for constructing a non-invasive screening model for non-alcoholic steatohepatitis according to any one of claims 1 to 5, It is characterized in that The system comprises: A data acquisition module is used to pre-process the serum, imaging and pathological data of patients with fatty liver disease; The serum characteristic data preprocessing module removes NASH and non-NASH pathological biopsy data and performs the following standardized processing on the remaining serum variables; the specific processing method is as follows: (1) Arrange each numerical variable from small to large, and consider those less than the 2.5th percentile or greater than the 97.5th percentile as outliers and change them to 2.5th and 97.5th percentiles, respectively; (2) Normalize all numerical variables. The formula is: n normalized=(X n –X min ) / (X max –X min ), where X n represents any numeric variable, X n normalized represents a numeric variable X n The normalized value of X max Indicates the maximum value of a numeric variable, X min Represents the minimum value of a numeric variable; Serum characteristic statistical screening, used for serum characteristic statistical screening; specifically, a part of the collected data is selected as a training group; then, in the training group, the indicators with significant correlation are screened out through Spearman correlation analysis; then, the serum indicators associated with NASH are statistically screened out through forward conditional multivariate logistic regression; wherein, the serum indicators are aspartate aminotransferase, platelet count, blood lipids, BMI calculation, and uric acid; The B-ultrasound feature extraction and screening module is used to extract and screen B-ultrasound features. The non-zero coefficient variables screened out by the lasso regression method are used as the effective B-ultrasound features for the final modeling, so as to calculate and extract the grayscale texture features of liver tissue; wherein, the Lasso regression model is used to screen out effective B-ultrasound features, and the cost function of the Lasso regression model is: Among them, x ij represents the jth B-ultrasound characteristic variable of the i-th patient data, y i is the response variable of the i-th patient data, q represents the total number of B-ultrasound feature variables, m is the total number of patient data, λ is the penalty function, For the best result; A feature integration module is used to predict the features of the effective B-ultrasound image group using a support vector machine (SVM) model, obtain a predicted value R-NASH, and integrate effective B-ultrasound features; The prediction model building module is used to combine the acquired serum index characteristics and the R-NASH characteristics of B-ultrasound, and use the multivariate logistic regression method to obtain a non-alcoholic fatty liver disease screening model; The prediction model confirmation module is equivalent to confirming the model threshold range for evaluating NASH, so as to obtain a model with fixed parameters; wherein, the model threshold range for evaluating NASH is confirmed by model comparison and performance analysis based on the maximum Youden index of the ROC curve, and is ultimately used for prediction.

7. A detection device for non-invasive screening of non-alcoholic fatty liver disease model construction method, It is characterized in that include: processor; a memory storing executable instructions of the processor; Wherein, the processor is configured to execute the steps of a method for constructing a non-invasive screening model for non-alcoholic fatty liver disease according to any one of claims 1 to 5 by executing the executable instructions.

8. A computer-readable storage medium for storing a program, It is characterized in that When the program is executed, the steps of a method for constructing a non-invasive screening model for non-alcoholic fatty liver disease described in any one of claims 1 to 5 are implemented.

Citation Information

Patent Citations

  • Diagnosis method of alcholic or non-alcoholic steato-hepatitis using biochemical markers

    US20060172286A1