A method for constructing a prediction model for chronic hepatitis B cirrhosis and its prediction method

Abstract

A method for constructing a prediction model for chronic hepatitis B cirrhosis and a prediction method thereof. Existing prediction methods for liver fibrosis and cirrhosis cannot be widely used due to cost issues, and the prediction effect is not accurate enough. A method for constructing a prediction model for chronic hepatitis B cirrhosis and a prediction method thereof, delineating an ROI in a CT plain scan image at the right portal vein level of the liver, wherein ROI means region of interest in Chinese, extracting imaging genomics features within the delineated ROI, and extracting imaging genomics features including first-order statistics, texture features, and wavelet transforms; selecting imaging genomics features using repeatability between and within observers and lasso regression; constructing a preliminary imaging genomics model using a support vector machine, and constructing a prediction model by combining a simple imaging genomics model with an independent predictor of serum indicators based on an imaging genomics nomogram. The method of the present invention can perform non-invasive prediction of liver cirrhosis by combining CT images with independent predictors of serum indicators.

Description

Technical Field

[0001] The present invention relates to the construction and prediction of a liver cirrhosis prediction model, and in particular to a method for constructing a chronic hepatitis B liver cirrhosis prediction model and a prediction method thereof. Background Art

[0002] Current noninvasive methods for predicting the extent of liver fibrosis include serum indices and elastography. TE and MRE are known to have excellent diagnostic performance for liver fibrosis staging. However, these high-performance equipment methods have not been widely used due to their high prices. In China, enhanced CT is usually recommended for HBV carriers to determine whether they have tumors, but due to limited cost-effectiveness, many patients only receive plain CT. Although enhanced CT or MRI can provide more information than plain CT, we hope to study noninvasive models to predict cirrhosis at a relatively low cost based on easily available data. Summary of the invention

[0003] The purpose of the present invention is to solve the problem that the existing liver fibrosis and cirrhosis prediction methods cannot be widely used due to cost issues and the prediction effect is not accurate enough, and to propose a method for constructing a chronic hepatitis B cirrhosis prediction model and its prediction method.

[0004] A method for constructing a prediction model for chronic hepatitis B cirrhosis, characterized in that the method is implemented by the following steps:

[0005] Step 1: Delineate ROI in the CT plain scan image at the level of the right portal vein of the liver, where ROI means region of interest in Chinese;

[0006] Step 2: Extract radiomics features within the defined ROI, including first-order statistics, texture features, and wavelet transform;

[0007] Step 3, select radiomics features using intra- and intra-observer reproducibility and lasso regression;

[0008] Step 4: Use support vector machine to build a preliminary radiomics model. Based on the radiomics nomogram, the simple radiomics model and the independent predictive factors of serum indicators are combined to build a prediction model Y:

[0009] Y=0.734×ALT * +0.468×AST * +1.385×GLOB * +2.372×INR * –0.454;

[0010] ALT *The value of ALT is 0 when ALT≤25.9U / L; 1 when ALT>25.9U / L. * The value of AST is 0 when AST≤32.6U / L and 1 when AST>32.5U / L.

[0011] GLOB * The value selection rule is: when GLOB≤33.9g / L, the value is 0; when GLOB>33.9g / L, the value is 1.

[0012] INR * The value of is: when INR≤1.10, the value is 0; when INR>1.10, the value is 1;

[0013] Among them, ALT means alanine aminotransferase; AST means aspartate aminotransferase; GLOB means globulin.

[0014] Preferably, in the step of defining the ROI in the CT plain scan image at the right portal vein level of the liver in step 1,

[0015] The ROI area was selected from the right portal vein level CT plain scan images of multiple livers using 3Dslicer software. The ROI area was defined at the level of the right portal vein along the edge of the right liver lobe without the large vessels on the plain scan CT images. The area range of the ROI was 19-106 cm 2 .

[0016] Preferably, the average area of ​​the ROI is 47±15 cm 2 .

[0017] Preferably, the step of extracting radiomics features in step 2 comprises:

[0018] First, the open source Pyradiomics software package was used for image preprocessing and feature extraction;

[0019] Second, the voxel spacing was standardized to a size of 1 × 1 × 1 mm, and the voxel intensity values ​​were discretized with a bin width of 25 HU;

[0020] Third, 828 radiomics features were extracted from each ROI, including 18 first-order statistics, 74 texture features, and 736 wavelet-based transform features;

[0021] Fourth, feature values ​​were standardized using z-scores in the training cohort; standard scores were applied in the validation cohort using the means and standard deviations determined in the training cohort.

[0022] Preferably, the step of selecting radiomics features using intra-observer and intra-observer repeatability and lasso regression in step 3 is specifically:

[0023] First, we randomly selected CT plain scan images of the liver at the level of the right portal vein, and quantified the reproducibility of each radiological feature using intra- and inter-class correlation coefficients; one quantitative method was to repeat the ROI segmentation twice a week, and the other quantitative method was to perform the ROI segmentation independently and then calculate the intra- and inter-observer reproducibility, respectively, where the minimum acceptable threshold of ICC was 0.8;

[0024] Among them, the intraclass and interclass correlation coefficients are referred to as ICC, and the full English name is intraclass correlation coefficient;

[0025] Second, radiomics features were selected;

[0026] First, features with high reproducibility were retained for subsequent analysis; the features with high reproducibility were those with intra-observer and inter-observer ICC values ​​greater than 0.8;

[0027] Secondly, the least absolute shrinkage and selection operator logistic regression algorithm was used, and the penalty parameter was adjusted by 10-fold cross validation, and the features with non-zero coefficients were considered as the features associated with cirrhosis.

[0028] Preferably, in the step of constructing a preliminary radiomics model using a support vector machine in step 4, and combining the simple radiomics model and the serum index independent predictor based on the radiomics nomogram, the preliminary radiomics model constructed using a support vector machine specifically refers to:

[0029] A binary classifier of pure radiomics was established using the support vector machine algorithm to distinguish liver fibrosis lesions; the support vector machine was implemented using the e1071 package on the R software based on the selected radiomics features.

[0030] The independent predictor of serum index is obtained by the following method, specifically:

[0031] First, Spearman correlation analysis was used to preliminarily screen out relevant factors for subsequent analysis, and correlation analysis results less than 0.05 were considered to meet the screening criteria;

[0032] Secondly, independent factors predicting cirrhosis were selected using forward conditional logistic multivariate analysis (input and output P values ​​were 0.05 and 0.1, respectively); the optimal cutoff value of each independent factor was determined by ROC analysis using the maximum Youden index;

[0033] The steps of constructing the prediction model are: a model for predicting cirrhosis in a training cohort was established by performing multivariate logistic regression analysis; a nomogram was constructed to provide a more understandable outcome measure; and then the critical values ​​derived from the training cohort were internally tested for the performance of the model in an independent validation cohort.

[0034] A method for predicting the risk of chronic hepatitis B cirrhosis, characterized in that the method comprises:

[0035] Obtain the subjects' predictive indicators related to the risk of developing liver cirrhosis;

[0036] Inputting the prediction index related to the risk of developing liver cirrhosis into a pre-established liver cirrhosis prediction model to calculate the risk of chronic hepatitis B liver cirrhosis;

[0037] When the calculated result is greater than or equal to 4, it is judged that there is a risk of developing liver cirrhosis.

[0038] The beneficial effects of the present invention are:

[0039] The present invention provides the first evidence of the feasibility of plain scan CT radiomics analysis, and uses machine learning methods to establish a reliable simple radiomics model. In addition, independent serum predictors are included in the nomogram model to provide more clinical benefits. The method of the present invention can achieve the establishment of a non-invasive and user-friendly radiomics nomogram model that integrates radiomics based on plain scan CT and independent predictors of serum indicators to evaluate the cirrhosis status of HBV patients. The radiomics nomogram model can help clinical decision-making and may provide more benefits for clinicians and specific patients.

[0040] The present invention established and validated a radiomics-based nomogram model to predict the cirrhosis status of HBV patients. The simple radiomics model included 25 stable radiomics features and had a strong ability to identify patients as cirrhotic or non-cirrhotic cases. Compared with CT subjective assessment of cirrhosis, APRI and FIB-4, the user-friendly nomogram model integrating radiomics, ALT, AST, GLOB and INR significantly improved the diagnostic performance and provided more clinical benefits.

[0041] The method also establishes reproducibility tests, including intra-observer and inter-observer ICC calculations, with a minimum threshold of 0.80. By selecting a smaller subset of radiomic features with higher reproducibility, the reproducibility of functional clustering of radiomic features can be significantly improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 A flowchart of a method for constructing a chronic hepatitis B cirrhosis prediction model according to the present invention;

[0043] Figure 2a The original plain scan CT image of the present invention;

[0044] Figure 2b A schematic diagram of selecting a region of interest on a plain scan CT original image according to the present invention;

[0045] Figure 3a A schematic diagram of determining the optimal λ value of the LASSO model involved in the present invention by using 10-fold cross validation through a minimum criterion;

[0046] Figure 3b LASSO coefficient curves of 85 selected features listed for the present invention;

[0047] Figure 4 It is a diagram of the principle of image omics feature extraction and screening involved in the present invention;

[0048] Figure 5 A schematic diagram of the construction process of the liver cirrhosis prediction model involved in the present invention;

[0049] Figure 6a A nomogram model integrating a simple radiomics model, ALT, AST, GLOB and INR was established from a training cohort for the present invention;

[0050] Figure 6b The ROC curve diagram of the nomogram model involved in the present invention compared with CT subjective evaluation, APRI and FIB-4 in the training group;

[0051] Figure 6c The ROC curve diagram of the nomogram model involved in the present invention compared with CT subjective assessment, APRI and FIB-4 in the validation group;

[0052] Figure 6d A probability correction curve diagram of the nomogram model involved in the present invention in a training group;

[0053] Figure 6e A probability correction curve diagram of the nomogram model involved in the present invention in the validation group;

[0054] Figure 7a Decision curve analysis for each model in the training data set involved in the present invention;

[0055] Figure 7b This is a decision curve analysis of each model in the validation data set involved in the present invention. DETAILED DESCRIPTION Specific implementation method one:

[0057] A method for constructing a chronic hepatitis B cirrhosis prediction model in this embodiment is as follows: Figure 1 and Figure 5 As shown, the method is implemented by the following steps:

[0058] Step 1: Delineate ROI in the CT plain scan image at the level of the right portal vein of the liver, where ROI means region on interest in Chinese.

[0059] Step 2: Extract radiomics features within the defined ROI, including first-order statistics, texture features, and wavelet transform;

[0060] Step 3, select radiomics features using intra- and intra-observer reproducibility and lasso regression;

[0061] Step 4: Use support vector machine to build a preliminary radiomics model. Based on the radiomics nomogram, the simple radiomics model and the independent predictive factors of serum indicators are combined to build a prediction model Y:

[0062] Y=0.734×ALT * +0.468×AST * +1.385×GLOB * +2.372×INR * –0.454;

[0063] ALT * The value of ALT is 0 when ALT≤25.9U / L; 1 when ALT>25.9U / L. * The value of AST is 0 when AST≤32.6U / L and 1 when AST>32.5U / L.

[0064] GLOB * The value selection rule is: when GLOB≤33.9g / L, the value is 0; when GLOB>33.9g / L, the value is 1.

[0065] INR * The value of is: when INR≤1.10, the value is 0; when INR>1.10, the value is 1;

[0066] Among them, ALT means alanine aminotransferase; AST means aspartate aminotransferase; GLOB means globulin;

[0067] Step 5. Evaluate the performance of the established model through receiver operating characteristic (ROC), calibration curve and decision curve analysis (DCA). Specific implementation method 2:

[0069] Different from the first embodiment, in the method for constructing a prediction model of chronic hepatitis B cirrhosis in this embodiment, in the step of defining the ROI in the CT plain scan image at the right portal vein level of the liver as described in step 1,

[0070] The ROI area was selected from the right portal vein level CT plain scan images of multiple livers using 3Dslicer software. The ROI area was defined at the level of the right portal vein along the edge of the right liver lobe, excluding the large vessels of the liver on the plain scan CT images. Figure 2a Shown is a plain scan CT image of the liver at the level of the right portal vein. Figure 2b The figure shows the delineation of the ROI region in the CT plain scan image of the liver at the level of the right portal vein; the area range of the ROI is 19-106cm 2 . Specific implementation method three:

[0072] Different from the second embodiment, in the method for constructing a prediction model of chronic hepatitis B cirrhosis in this embodiment, the average area of ​​the ROI is 47±15cm 2 . Specific implementation method four:

[0074] Different from the first, second or third specific implementation modes, in the method for constructing a prediction model for chronic hepatitis B cirrhosis in this implementation mode, the step of extracting radiomics features in step 2 includes:

[0075] First, the open source Pyradiomics software package was used for image preprocessing and feature extraction;

[0076] Second, the voxel spacing was standardized to a size of 1 × 1 × 1 mm, and the voxel intensity value was discretized with a bin width of 25 HU to reduce the interference of image noise and normalize the intensity;

[0077] Third, 828 radiomics features were extracted from each ROI, including 18 first-order statistics, 74 texture features, and 736 wavelet-based transform features;

[0078] Fourth, feature values ​​were standardized using z-scores in the training cohort; the standard scores (z-scores) were applied in the validation cohort using the mean and standard deviation determined in the training cohort.

[0079] like Figure 3a and 3b As shown, among the 828 extracted features, 85 features (8 first-order statistics, 21 texture features, and 56 wavelet-based transforms) with high reproducibility were selected for subsequent analysis. Twenty-five cirrhosis-related features with non-zero coefficients in the lasso regression model were selected based on the training cohort. Specific implementation method five:

[0081] Different from the fourth embodiment, in the method for constructing a prediction model of chronic hepatitis B cirrhosis in this embodiment, the step of selecting radiomics features by using inter-observer and intra-observer repeatability and lasso regression in step three is shown in the schematic diagram of the principle as follows: Figure 4 As shown, specifically:

[0082] First, we randomly selected CT plain scan images of the liver at the level of the right portal vein, and based on 50 randomly selected patients, we quantified the reproducibility of each radiological feature using intra- and inter-class correlation coefficients; one quantitative method was to repeat the ROI segmentation twice a week, and the other quantitative method was to perform the ROI segmentation independently and then calculate the intra- and inter-observer reproducibility, respectively, where the minimum acceptable threshold of ICC was 0.8;

[0083] Among them, the intraclass and interclass correlation coefficients are referred to as ICC, and the full English name is intraclass correlation coefficient;

[0084] Second, select important radiomics features;

[0085] First, features with high reproducibility were retained for subsequent analysis; the features with high reproducibility were those with intra-observer and inter-observer ICC values ​​greater than 0.8;

[0086] Secondly, the least absolute shrinkage and selection operator (LASSO) logistic regression algorithm was used, and the penalty parameter was adjusted through 10-fold cross validation to consider features with non-zero coefficients as features related to liver cirrhosis, such as Figure 3a and 3b shown.

[0087] Figure 3a and 3b is a diagram showing the selection of radiomics features using LASSO regression; Figure 3aShown is the LASSO model using 10-fold cross validation to determine the optimal λ value through the minimum criterion. The binomial deviation curve is plotted against log(λ). The dashed vertical line is drawn at the optimal value by using the minimum criterion and 1 standard error of the minimum criterion (1–standard error condition). Here, the optimal λ value is 0.0383. Figure 3b Shown is a LASSO coefficient curve listing 85 selected features. Specific implementation method six:

[0089] Different from the fifth specific embodiment, in the method for constructing a prediction model of chronic hepatitis B cirrhosis in this embodiment, in the step of constructing a preliminary radiomics model using a support vector machine in step 4, the simple radiomics model and the serum index independent predictor based on the radiomics nomogram, the preliminary radiomics model constructed using a support vector machine specifically refers to:

[0090] The support vector machine (SVM) algorithm was used to establish a binary classifier of pure radiomics to distinguish liver fibrosis lesions of F0-3 and F4; the support vector machine SVM was implemented by the e1071 package on R software based on the selected radiomics features.

[0091] The independent predictor of serum index is obtained by the following method, specifically:

[0092] First, Spearman correlation analysis was used to preliminarily screen out factors with significant correlations for subsequent analysis, and correlation analysis results less than 0.05 were considered to meet the screening criteria;

[0093] Secondly, independent factors predicting cirrhosis were selected using forward conditional logistic multivariate analysis (input and output P values ​​were 0.05 and 0.1, respectively); the optimal cutoff value of each independent factor was determined by ROC analysis using the maximum Youden index;

[0094] The steps of constructing the prediction model are: establishing a model for predicting cirrhosis in a training cohort by performing multivariate logistic regression analysis; constructing a nomogram to provide a more understandable outcome measure; and then internally testing the performance of the model in an independent validation cohort using the formula and the critical value derived from the training cohort.

[0095] in,

[0096] (1) The relevant description of the training group and validation group data involved in the present invention is shown in Table 1.

[0097] There were no statistical differences in clinical data, radiology report data, and pathological features between the training and validation groups. There was no difference in the incidence of cirrhosis between the two groups (training group: 43.8% [63 / 144]; validation group: 38.0%, [57 / 150]; P = 0.34). The overall identification accuracy of subjective assessment report of CT for cirrhosis was 77.6% (228 / 294), sensitivity was 62.5% (75 / 120), specificity was 87.9% (153 / 174), positive predictive value was 78.1% (75 / 96), and negative predictive value was 77.3% (153 / 198).

[0098] Table 1. Baseline information

[0099]

[0100] Note: Unless otherwise specified, data are numbers of patients, and percentages are in brackets. ALT, alanine aminotransferase; APRI, aspartate aminotransferase to platelet ratio; AST, aspartate aminotransferase; FIB-4, fibrosis-4; GGT, glutamyl transpeptidase; INR, international normalized ratio.

[0101] *Data are medians, and the data in brackets are quartiles

[0102] (2) Related instructions for radiomics analysis are as follows:

[0103] A radiomics-only model was constructed using the SVM algorithm, with the type of SVM being "c-classification" and the kernel function being radial basis, and the total number of support vectors being 133: 62 for noncirrhotics and 71 for cirrhotics. Significant differences in radiomics scores were found between patients with cirrhosis and those without cirrhosis in the training cohort (mean, 0.279 vs -0.649, P < .001), which were then confirmed in the validation cohort (mean, 0.141 vs -0.585, P < .001). The radiomics-only model showed good discriminatory power, with an AUC of 0.879 (95% confidence interval [CI]: 0.827, 0.932) in the training cohort and 0.858 (95% CI: 0.795, 0.921) in the validation cohort.

[0104] (3) The relevant instructions for screening of clinical laboratory test factors are as follows:

[0105] As shown in Table 2, in the training cohort, WBC, PLT, ALT, AST, CB, ALB, GLOB, TBA, TC, LDL-C, ApoB, PT, and INR were simultaneously associated with cirrhosis (P < .05, Spearman correlation analysis). Multivariate conditional logistic regression analysis identified ALT, AST, GLOB, and INR as independent predictors of cirrhosis (Table 2).

[0106] Table 2. Clinical characteristics associated with liver cirrhosis in the training group

[0107]

[0108] Note: b coefficients are from multivariate logistic regression. Clinical variables found to be significantly associated with cirrhosis by Spearman correlation analysis were included in the forward conditional logistic regression multivariate analysis. ALB, albumin; ALP, alkaline phosphatase; ALT, alanine aminotransferase; Apo A1, apolipoprotein A1; Apo B, apolipoprotein B; AST, aspartate aminotransferase; CB, conjugated bilirubin; CRP, C-reactive protein; GGT, glutamyl transpeptidase; GLOB, globulin; Hb, hemoglobin; HDL-C, high-density lipoprotein cholesterol; INR, international normalized ratio; LAP, leucine alanylamidase; LDH, lactate dehydrogenase; LDL-C, low-density lipoprotein cholesterol; PLT, platelets; PT, prothrombin time; RBC, red blood cells; TB, total serum bilirubin; TBA, total bile acid; TC, total cholesterol; WBC, white blood cells.

[0109] As shown in Table 3, according to ROC analysis, the optimal cutoff values ​​of ALT, AST, GLOB and INR were 25.9 U / L, 32.6 U / L, 33.9 g / L and 1.10, respectively.

[0110] Table 3. Optimal cutoff values ​​for independent predictors

[0111]

[0112] Note: AUC, area under the curve; ALT, alanine aminotransferase; AST, aspartate aminotransferase; GLOB, globulin; INR, international normalized ratio; ROC, receiver operating characteristic curve

[0113] (4) Statistical analysis related instructions are as follows:

[0114] Categorical variables and continuous variables were compared using the chi-square test and the Mann-Whitney U test, respectively. All statistical analyses were performed using R software. The diagnostic performance of the established models was evaluated by ROC curves and area under the curve (AUC) values. AUC values ​​were compared using the Delong test. Decision curve analysis (DCA) was used to calculate the net benefit of the models using different threshold probabilities. Calibration curves were plotted to evaluate the probability calibration of the radiomics models, accompanied by the Hosmer-Lemeshow test. Two-sided P values ​​less than 0.05 were considered statistically significant. Specific implementation method seven:

[0116] A method for predicting the risk of chronic hepatitis B cirrhosis in this embodiment comprises:

[0117] Obtain the subjects' predictive indicators related to the risk of developing liver cirrhosis;

[0118] Inputting the prediction index related to the risk of developing liver cirrhosis into a pre-established liver cirrhosis prediction model to calculate the risk of chronic hepatitis B liver cirrhosis;

[0119] When the calculated result is greater than or equal to 4, it is judged that there is a risk of developing liver cirrhosis.

[0120] Example:

[0121] Modeling using the method of the present invention and verification of the built model:

[0122] Figure 6a-6e Shown are the nomogram model, ROC curve, and probability correction curve. Figure 6a As shown, a nomogram model integrating the simple radiomics model, ALT, AST, GLOB, and INR was established from the training cohort; Figure 6b , Figure 6c The ROC curves of the nomogram model compared with CT subjective assessment, APRI and FIB-4 in the training group and the validation group are shown respectively; Figure 6d , Figure 6e The probability correction curves of the nomogram model in the training group and the validation group are shown respectively. ALT, alanine aminotransferase; APRI, ratio of aspartate aminotransferase to platelets; AST, aspartate aminotransferase; FIB-4, fibrosis-4; GLOB, globulin; INR, international normalized ratio; ROC, receiver operating characteristic curve.

[0123] As shown in Table 4 , in the training cohort, the radiomics-based nomogram model had the best discriminatory ability, with an AUC value of 0.915 (95% CI: 0.869, 0.961), which was significantly higher than the subjective CT assessment of cirrhosis (AUC: 0.752; 95% CI: 0.683, 0.821; P < .001), APRI (AUC: 0.752; 95% CI: 0.683, 0.821; P < .001), and FIB-4 (AUC: 0.664; 95% CI: 0.575, 0.753; P < .001). In the validation cohort, the radiomics nomogram model also produced the highest AUC of 0.872 (95% CI: 0.814, 0.930) compared with subjective assessment of cirrhosis by CT alone (AUC: 0.755; 95% CI: 0.683, 0.827; P = .006), APRI (AUC: 0.731; 95% CI: 0.649, 0.814; P = .003), and FIB-4 (AUC: 0.688; 95% CI: 0.601, 0.775; P < .001). The optimal cutoff value of the nomogram model was 0.014, determined by the point of the maximum Youden index in the entire cohort study. The overall correct classification rate of the nomogram was 82.0%, the sensitivity was 77.5%, the specificity was 85.1%, the positive predictive value was 78.2%, and the negative predictive value was 84.6%.

[0124] Table 4. Discrimination ability of all noninvasive prediction models in the training and validation groups

[0125]

[0126]

[0127] Note: The data in brackets are 95% confidence intervals. APRI, aspartate aminotransferase to platelet ratio; AUROC, area under the receiver operating characteristic curve; FIB-4, fibrosis-4.

[0128] The probability correction curves of the nomogram model showed a strong agreement between the predicted cirrhosis and the actual cirrhosis in the training cohort (see Figure 6d The Hosmer-Lemeshow test had a p value of 0.317, indicating that the model did not deviate from a good fit. Figure 6e ) and Hosmer-Lemeshow test results (P = 0.534), the nomogram model also achieved a good calibration effect in the validation cohort.

[0129] Figure 7a Shown is the decision curve analysis of each model in the training data set. Figure 7bDecision curve analysis for each model in the validation dataset. Net benefit is measured on the y-axis. Within the threshold probability range, the application of the radiomics nomogram model to predict cirrhosis status was more beneficial than treating all or abandoning all patients, subjective assessment of cirrhosis status by CT, APRI, and FIB-4. APRI, aspartate aminotransferase to platelet ratio; FIB-4, fibrosis-4.

[0130] The DCA curves of the radiomics nomogram model, subjective CT assessment of cirrhosis, APRI, and FIB-4 are shown in Figure 7. Within the reasonable threshold probability range of the two cohorts (training group and validation group), DCA vividly shows in a graphical way that the radiomics nomogram model can provide a greater net benefit for predicting cirrhosis compared with the other three prediction methods.

Claims

1. A method for constructing a prediction model for chronic hepatitis B cirrhosis, characterized in that: The method is implemented by the following steps: Step 1: Delineate ROI in the CT plain scan image at the level of the right portal vein of the liver, where ROI means region on interest in Chinese. Step 2: Extract radiomics features within the defined ROI, including first-order statistics, texture features, and wavelet transform; Step 3, select radiomics features using intra- and intra-observer reproducibility and lasso regression; Step 4: Use support vector machine to build a preliminary radiomics model. Based on the radiomics nomogram, the simple radiomics model and the independent predictive factors of serum indicators are combined to build a prediction model Y: Y=0.734×ALT * +0.468×AST * +1.385×GLOB * +2.372×INR * –0.454; ALT * The value of ALT is 0 when ALT≤25.9U / L; 1 when ALT>25.9U / L. AST * The value of AST is 0 when AST≤32.6U / L and 1 when AST>32.5U / L. GLOB * The value selection rule is: when GLOB≤33.9g / L, the value is 0; when GLOB>33.9g / L, the value is 1. INR * The value of is: when INR≤1.10, the value is 0; when INR>1.10, the value is 1; Among them, ALT means alanine aminotransferase; AST means aspartate aminotransferase; GLOB means globulin; The steps of extracting radiomics features in step 2 include: First, the open source Pyradiomics software package was used for image preprocessing and feature extraction; Second, the voxel spacing was standardized to a size of 1 × 1 × 1 mm, and the voxel intensity values ​​were discretized with a bin width of 25 HU; Third, 828 radiomics features were extracted from each ROI, including 18 first-order statistics, 74 texture features, and 736 wavelet-based transform features; Fourth, feature values ​​were standardized using z scores in the training cohort; standard scores were applied in the validation cohort using the mean and standard deviation determined in the training cohort; The steps for selecting radiomics features using intra- and intra-observer repeatability and lasso regression described in step 3 are as follows: First, we randomly selected CT plain scan images of the liver at the level of the right portal vein, and quantified the reproducibility of each radiological feature using intra- and inter-class correlation coefficients; one quantitative method was to repeat the ROI segmentation twice a week, and the other quantitative method was to perform the ROI segmentation independently and then calculate the intra- and inter-observer reproducibility, respectively, where the minimum acceptable threshold of ICC was 0.8; Among them, the intraclass and interclass correlation coefficients are referred to as ICC, and the full English name is intraclass correlation coefficient; Second, radiomics features were selected; First, features with high reproducibility were retained for subsequent analysis; the features with high reproducibility were those with intra-observer and inter-observer ICC values ​​greater than 0.8; Secondly, the least absolute shrinkage and selection operator logistic regression algorithm was used, and the penalty parameter was adjusted by 10-fold cross validation, and the features with non-zero coefficients were considered as the features associated with liver cirrhosis; In step 4, the use of support vector machine to construct a preliminary radiomics model, and the step of combining the simple radiomics model and the serum index independent predictor based on the radiomics nomogram, the use of support vector machine to construct a preliminary radiomics model specifically refers to: The independent predictor of serum index is obtained by the following method, specifically: First, Spearman correlation analysis was used to preliminarily screen out relevant factors for subsequent analysis, and correlation analysis results less than 0.05 were considered to meet the screening criteria; Secondly, forward conditional logistic multivariate analysis was used to select independent factors for predicting liver cirrhosis, with input and output values ​​of 0.05 and 0.1, respectively; the optimal cutoff value of each independent factor was determined by ROC analysis using the maximum Youden index; The steps of constructing the prediction model are: a model for predicting cirrhosis in a training cohort was established by performing multivariate logistic regression analysis; a nomogram was constructed to provide a more understandable outcome measure; and then the critical values ​​derived from the training cohort were internally tested for the performance of the model in an independent validation cohort.

2. The method for constructing a prediction model for chronic hepatitis B cirrhosis according to claim 1, characterized in that: In the step of defining the ROI in the CT plain scan image at the level of the right portal vein of the liver as described in step 1, The ROI area was selected from the right portal vein level CT plain scan images of multiple livers using 3Dslicer software. The ROI area was defined at the level of the right portal vein along the edge of the right liver lobe without the large vessels on the plain scan CT images. The area range of the ROI was 19-106 cm 2 .

3. The method for constructing a chronic hepatitis B cirrhosis prediction model according to claim 2, characterized in that: The average area of ​​the ROI is 47±15cm 2 .

Images