Liver cancer microvascular infiltration prediction method and system

Abstract

The invention discloses a liver cancer microvascular infiltration prediction method and system, and relates to the technical field of medical image analysis and artificial intelligence, and the method comprises the steps: obtaining the liver data of a prediction object, carrying out the preprocessing of the liver data of the prediction object, and obtaining the processed liver data of the prediction object, the liver data of the predicted object comprises the pathology of the predicted object, multi-stage enhanced image information and liver related data; screening the processed liver data of the prediction object to obtain a radiomics characteristic index associated with the liver cancer microvascular infiltration degree, and inputting the radiomics characteristic index associated with the liver cancer microvascular infiltration degree into a pre-established liver cancer microvascular infiltration prediction model, outputting to obtain a trained liver cancer microvascular infiltration prediction model; and checking and comparing the trained liver cancer microvascular infiltration prediction model by adopting a mode of calculating an AUC value, determining an optimal liver cancer microvascular infiltration prediction model according to a comparison result, and realizing prediction of liver cancer microvascular infiltration through the optimal liver cancer microvascular infiltration prediction model.

Description

Technical Field

[0001] The present invention relates to the technical fields of medical image analysis and artificial intelligence, and specifically relates to a method and system for predicting microvascular invasion of liver cancer. Background Art

[0002] In recent years, radiomics based on enhanced CT has been proven to be able to effectively and non-invasively predict the preoperative microvascular invasion status of liver cancer. Enhanced CT images can not only clearly present a series of typical imaging features of liver cancer, but also contain rich information that cannot be observed by the naked eye; and radiomics can help to mine this potential information; as an emerging image analysis method, radiomics can extract rich information of tumors and the peritumoral microenvironment from imaging images before surgery, so as to reveal the pathophysiological characteristics of tumors. It shows great potential in characterizing tumor phenotypes and improving cancer diagnosis, prognosis and treatment response. As far as we know, about 85% of MVI occurs around the tumor. Traditional omics research mostly focuses on a single phase of enhanced CT or the tumor itself to predict microvascular invasion, which may ignore some important information in other phases and around the tumor. Relying on a single aspect of clinical, imaging or radiomics has certain limitations and the AUC value of the detection accuracy is usually around 0.7. At the same time, traditional machine learning often lacks interpretability, resulting in the black box problem, which is not conducive to clinical application. Summary of the Invention

[0003] To solve the deficiencies mentioned in the above background art, the purpose of the present invention is to provide a method and system for predicting microvascular invasion of liver cancer.

[0004] In a first aspect, the purpose of the present invention can be achieved by the following technical solutions: A method for predicting microvascular invasion of liver cancer, the method comprising the following steps:

[0005] Obtain the liver data of the prediction object, preprocess the liver data of the prediction object to obtain the processed liver data of the prediction object, wherein the liver data of the prediction object includes: the pathology of the prediction object, multi-phase enhanced imaging information and liver-related data;

[0006] Screen the processed liver data of the prediction object to obtain radiomics feature indicators associated with the degree of microvascular invasion of liver cancer, and input the radiomics feature indicators associated with the degree of microvascular invasion of liver cancer into a pre-established prediction model for microvascular invasion of liver cancer, and output a trained prediction model for microvascular invasion of liver cancer;

[0007] Use the trained prediction model for microvascular invasion of liver cancer to perform a test comparison by calculating the AUC value, determine the optimal prediction model for microvascular invasion of liver cancer according to the comparison result, and realize the prediction of microvascular invasion of liver cancer through the optimal prediction model for microvascular invasion of liver cancer.

[0008] In combination with the first aspect, in certain implementations of the first aspect, the method further includes: The liver data of the prediction object specifically includes: the basic information of the patient, the pathological data of hepatocellular carcinoma microvascular invasion, the multi-phase enhanced CT imaging data, and the clinical and serum laboratory examination data.

[0009] In combination with the first aspect, in certain implementations of the first aspect, the method further includes: The process of preprocessing the liver data of the prediction object:

[0010] The finally extracted radiomics features are normalized, and the normalization formula is as follows:

[0011] Xn normalized = (Xn – Xmin) / (Xmax – Xmin);

[0012] Xn represents any numerical variable, Xn normalized represents the value of the numerical variable Xn after normalization, Xmax represents the maximum value in the numerical variable, and Xmin represents the minimum value in the numerical variable;

[0013] The continuous variables in the clinical and imaging features are binarized, and the expression of the restricted cubic spline function is as follows:

[0014] RCS(X) = β 0 X + β 1 S 1 +…+ β k-2 S k-2

[0015] where Si is the cubic component falling in the i-th knot.

[0016] In combination with the first aspect, in certain implementations of the first aspect, the method further includes: The multi-phase enhanced CT imaging data includes: the maximum tumor diameter, morphology, liver capsule invasion; capsule, low enhancement in the arterial phase, low-density ring sign, intratumoral arterial visualization, intrahepatic metastasis, high perfusion in the peritumoral arterial phase, major vessel invasion, and blood supply situation.

[0017] In combination with the first aspect, in certain implementations of the first aspect, the method further includes: The processed liver data of the prediction object is screened, and the selected indicators are further refined by lasso regression. The cost function of the 5-fold cross-validation of lasso regression is:

[0018]

[0019] In the formula, is the mean square error term, m is the number of samples, y i is the actual value of the i-th sample, h θ (x i) is the predicted value of the model for the i-th sample; is the L1 regularization term, λ is the regularization parameter, and θ j is the j-th parameter of the model, and n is the total number of parameters.

[0020] Combined with the first aspect, in some implementation manners of the first aspect, the method further includes: the process of inputting the radiomics feature indexes associated with the degree of hepatocellular carcinoma microvascular invasion into a pre-established hepatocellular carcinoma microvascular invasion prediction model and outputting the trained hepatocellular carcinoma microvascular invasion prediction model is as follows:

[0021] Dividing the dataset of radiomics feature indexes associated with the degree of hepatocellular carcinoma microvascular invasion into a training sample set and a test sample set;

[0022] Using the Logistic regression method to establish a hepatocellular carcinoma microvascular invasion prediction model and setting the parameters of the hepatocellular carcinoma microvascular invasion prediction model, including the regularization strength and whether to perform feature standardization;

[0023] Inputting the training sample set into the hepatocellular carcinoma microvascular invasion prediction model to complete the training of the model;

[0024] Inputting the test sample set into the hepatocellular carcinoma microvascular invasion prediction model and outputting the relevant data of the hepatocellular carcinoma microvascular invasion status;

[0025] Establishing a receiver operating characteristic curve ROC according to the relevant data of the hepatocellular carcinoma microvascular invasion status.

[0026] Combined with the first aspect, in some implementation manners of the first aspect, the method further includes: when predicting the hepatocellular carcinoma microvascular invasion through the optimal hepatocellular carcinoma microvascular invasion prediction model, the optimal hepatocellular carcinoma microvascular invasion prediction model is preferably selected as the intratumoral omics feature data in the arterial phase, and the Radscore calculation method is as follows:

[0027] Radscore = log.sigma.1.mm.3D_firstorder_Variance × 0.1 + log.sigma.1.mm.3D_glszm_LowGrayLevelZoneEmphasis × -0.207 + wavelet.HH - glcm_InverseVariance × -0.228 + exponential_ngtdm_Contrast × -0.025 + wavelet.LL_glrlm_RunVariance × 0.284 + wavelet.LH_gldm_LargeDependenceLowGrayLevelEmphasis × -0.053 + squareroot_glszm_GrayLevelNonUniformity × 0.152 + firstorder_RootMeanSquared × -0.206 - 0.569。

[0028] In the formula, firstorder_RootMeanSquared and log.sigma.1.mm.3D_firstorder_Variance are first - order statistical features, log.sigma.1.mm.3D_glszm_LowGrayLevelZoneEmphasis and squareroot_glszm_GrayLevelNonUniformity are gray - level run - length matrix features, wavelet.HH - glcm_InverseVariance is a gray - level co - occurrence matrix feature, wavelet.LH_gldm_LargeDependenceLowGrayLevelEmphasis is a gray - level dependence matrix feature, wavelet.LL_glrlm_RunVariance and wavelet.HL_glrlm_RunEntropy are gray - level run - length matrix features, exponential_ngtdm_Contrast is a gray - level neighborhood difference matrix feature, and wavelet.LL_glrlm_RunVariance, wavelet.HH - glcm_InverseVariance, wavelet.LH_gldm_LargeDependenceLowGrayLevelEmphasis, wavelet.HL_glrlm_RunEntropy are wavelet transform features.

[0029] Second aspect, to achieve the above object, the present invention discloses a prediction system for hepatocellular carcinoma microvascular invasion, comprising:

[0030] A data processing module, configured to obtain the liver data of a prediction object, preprocess the liver data of the prediction object to obtain the processed liver data of the prediction object, wherein the liver data of the prediction object includes: the pathology of the prediction object, multi-phase enhanced imaging information, and liver-related data;

[0031] An index screening module, configured to screen the processed liver data of the prediction object to obtain imaging feature indexes associated with the degree of hepatocellular carcinoma microvascular invasion, input the imaging feature indexes associated with the degree of hepatocellular carcinoma microvascular invasion into a pre-established prediction model for hepatocellular carcinoma microvascular invasion, and output a trained prediction model for hepatocellular carcinoma microvascular invasion;

[0032] A prediction module, configured to test and compare the trained prediction model for hepatocellular carcinoma microvascular invasion by calculating the AUC value, determine the optimal prediction model for hepatocellular carcinoma microvascular invasion according to the comparison result, and realize the prediction of hepatocellular carcinoma microvascular invasion through the optimal prediction model for hepatocellular carcinoma microvascular invasion.

[0033] In another aspect of the present invention, to achieve the above object, a terminal device is disclosed, comprising a memory, a processor, and a computer program stored in the memory and capable of running on the processor. The memory stores a computer program capable of running on the processor. When the processor loads and executes the computer program, it adopts a prediction method for hepatocellular carcinoma microvascular invasion as described above.

[0034] In yet another aspect of the present invention, to achieve the above object, a computer-readable storage medium is disclosed. The computer-readable storage medium stores a computer program, characterized in that when the computer program is loaded and executed by a processor, it adopts a prediction method for hepatocellular carcinoma microvascular invasion as described above.

[0035] Advantages of the present invention:

[0036] The advantages of the present invention are mainly reflected in the following technical aspects:

[0037] Improve prediction accuracy: By integrating multi-phase enhanced imaging information, pathological data, and liver-related clinical data, the present invention can extract features related to hepatocellular carcinoma microvascular invasion from multiple dimensions. This multi-modal data fusion method significantly enhances the model's ability to identify microvascular invasion. Compared with the prediction method using a single data source, it can more accurately evaluate the microvascular invasion risk of patients.

[0038] Optimizing Feature Selection and Model Performance: The present invention adopts advanced radiomics analysis techniques to screen out feature indicators significantly related to microvascular invasion in liver cancer, effectively reducing the impact of data noise on prediction results. By calculating the AUC value to test and compare the model, the performance of the model is further optimized to ensure that it can maintain high prediction accuracy and stability on different datasets.

[0039] Improving the Efficiency of Clinical Decision Support: The present invention provides an efficient and accurate risk assessment tool for predicting microvascular invasion in liver cancer. By predicting the risk of microvascular invasion in advance, doctors can more reasonably arrange the follow-up plans and intervention measures for patients, thereby improving the utilization efficiency of medical resources and the postoperative management effect of patients.

[0040] Enhancing the Interpretability of the Model: Through a clear feature screening process and the determination of significant correlation indicators, the present invention not only improves the prediction accuracy but also enhances the interpretability of the model. Clinicians can more intuitively understand which factors have important impacts on the risk of microvascular invasion, thus better applying this prediction model in clinical practice.

[0041] Reducing the Misdiagnosis Rate and Missed Diagnosis Rate: The prediction model of the present invention is developed based on a large amount of clinical data and advanced machine learning techniques, and can effectively reduce the misdiagnosis and missed diagnosis situations that may occur in traditional diagnostic methods. By accurately predicting the risk of microvascular invasion, it helps to detect potential risks at an early stage, take intervention measures in a timely manner, and improve the survival rate and quality of life of patients. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, for those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings;

[0043] Figure 1 It is a schematic diagram of the method flow of the present invention;

[0044] Figure 2 It is a schematic diagram of using lasso regression 5-fold cross-validation to screen non-zero coefficient variables as the input variables for final modeling for the intratumoral omics data of the arterial phase with positive microvascular invasion in liver cancer involved in the present invention;

[0045] Figure 3 It is a schematic diagram of using lasso regression 5-fold cross-validation to screen non-zero coefficient variables as the input variables for final modeling for the intratumoral omics data of the portal vein phase with positive microvascular invasion in liver cancer involved in the present invention;

[0046] Figure 4Schematic diagram of using lasso regression 5-fold cross-validation to screen non-zero coefficient variables as input variables for final modeling for intratumoral omics data in the delayed phase of hepatocellular carcinoma with positive microvascular invasion;

[0047] Figure 5 Schematic diagram of using lasso regression 5-fold cross-validation to screen non-zero coefficient variables as input variables for final modeling for peritumoral omics data in the arterial phase of hepatocellular carcinoma with positive microvascular invasion;

[0048] Figure 6 Schematic diagram of using lasso regression 5-fold cross-validation to screen non-zero coefficient variables as input variables for final modeling for peritumoral omics data in the portal phase of hepatocellular carcinoma with positive microvascular invasion;

[0049] Figure 7 Schematic diagram of using lasso regression 5-fold cross-validation to screen non-zero coefficient variables as input variables for final modeling for peritumoral omics data in the delayed phase of hepatocellular carcinoma with positive microvascular invasion;

[0050] Figure 8 Schematic diagram of using lasso regression 5-fold cross-validation to screen non-zero coefficient variables as input variables for final modeling for intratumoral + peritumoral omics data in the arterial phase of hepatocellular carcinoma with positive microvascular invasion;

[0051] Figure 9 Schematic diagram of using lasso regression 5-fold cross-validation to screen non-zero coefficient variables as input variables for final modeling for intratumoral + peritumoral omics data in the portal phase of hepatocellular carcinoma with positive microvascular invasion;

[0052] Figure 10 Schematic diagram of using lasso regression 5-fold cross-validation to screen non-zero coefficient variables as input variables for final modeling for intratumoral + peritumoral omics data in the delayed phase of hepatocellular carcinoma with positive microvascular invasion;

[0053] Figure 11 Schematic diagram of AUC value of logistic regression model constructed based on lasso regression 5-fold cross-validation for screening multi-phase CT enhanced imaging omics data;

[0054] Figure 12 Schematic diagram of the preprocessing process of multi-phase CT enhanced imaging omics data involved in the present invention;

[0055] Figure 13 Schematic diagram of the preprocessing process of clinical and imaging feature information based on restricted cubic spline (RCS) analysis involved in the present invention;

[0056] Figure 14The receiver operating characteristic curve (ROC) graph for predicting microvascular invasion of liver cancer by the training group based on the Catboost machine learning algorithm involved in the present invention;

[0057] Figure 15 The receiver operating characteristic curve (ROC) graph for predicting microvascular invasion of liver cancer by the validation group based on the Catboost machine learning algorithm involved in the present invention;

[0058] Figure 16 The decision curve analysis (DCA) graph for predicting microvascular invasion of liver cancer by the validation group based on the Catboost machine learning algorithm involved in the present invention;

[0059] Figure 17 The confusion matrix graph for predicting microvascular invasion of liver cancer by the optimal combined model training group based on the Catboost machine learning algorithm involved in the present invention;

[0060] Figure 18 The confusion matrix graph for predicting microvascular invasion of liver cancer by the optimal combined model validation group based on the Catboost machine learning algorithm involved in the present invention;

[0061] Figure 19 The summary plot of the global interpretation of the machine learning model by the Shapley additive explanation of the optimal combined model based on the Catboost machine learning algorithm involved in the present invention;

[0062] Figure 20 The waterfall plot of the local interpretation of the machine learning model by the Shapley additive explanation of the optimal combined model based on the Catboost machine learning algorithm involved in the present invention;

[0063] Figure 21 The force plot of the local interpretation of the machine learning model by the Shapley additive explanation of the optimal combined model based on the Catboost machine learning algorithm involved in the present invention;

[0064] Figure 22 The schematic diagram of the system structure of the present invention. Detailed implementation manners

[0065] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0066] Embodiment 1:

[0067] As Figure 1 shown, a method for predicting hepatocellular carcinoma microvascular invasion, the method includes the following steps:

[0068] S101: Obtain the liver data of the prediction object, preprocess the liver data of the prediction object to obtain the processed liver data of the prediction object, wherein the liver data of the prediction object includes: the pathology of the prediction object, multi-phase enhanced imaging information, and liver-related data;

[0069] The liver data of the prediction object specifically includes: the basic information of the patient, the pathological data of hepatocellular carcinoma microvascular invasion, the multi-phase enhanced CT imaging data, and the clinical and serum laboratory examination data.

[0070] The process of preprocessing the liver data of the prediction object:

[0071] (1) The finally extracted radiomics features need to be normalized, and the normalization formula is as follows:

[0072] Xn normalized = (Xn – Xmin) / (Xmax – Xmin);

[0073] Xn represents any numerical variable, Xn normalized represents the value of the numerical variable Xn after normalization, Xmax represents the maximum value in the numerical variable, and Xmin represents the minimum value in the numerical variable;

[0074] The continuous variables in clinical and imaging features need to be binarized ( Figure 13 ), and the expression of the restricted cubic spline function is as follows:

[0075] RCS(X) = β0X + β1S1 + … + βk-2Sk-2

[0076] where Si is the cubic component falling in the i-th knot.

[0077] Among them,

[0078] The described clinical data includes age, gender, hepatitis history, and cirrhosis history;

[0079] The described laboratory serum indicators include alpha-fetoprotein (ng / ml), alanine aminotransferase (U / L), aspartate aminotransferase (AST) (U / L), alkaline phosphatase (ALP) (U / L), gamma-glutamyltransferase (GGT) (U / L), lactate dehydrogenase (LDH) (U / L), total bilirubin (TB) (μmol / L), total bile acid (TBA) (μmol / L), white blood cell count (WBC) (10^9 / L), and platelet count (PLT) (10^9 / L). Calculate the evaluation indicators for liver function and cirrhosis: A / G, ALP / ALT, AST / ALP, modified albumin-bilirubin mALBI (albumin-bilirubin score), APRI index, FIB-4, DeRitis ratio (AST / ALT);

[0080] The formula for the APRI index is:

[0081] APRI = (AST (IU / L) / ULN) × 100 / (Platelet count (10^9 / L))

[0082] The formula for FIB-4 is:

[0083] FIB-4 = age (years) × AST (IU / L) / (Platelet count (10^9 / L) × ALT (IU / L)^1 / 2)

[0084] The formula for mALBI is:

[0085] ALBI = (log10 bilirubin × 0.66) + (albumin × -0.085)

[0086] Among them, the unit of bilirubin is μmol / L, the unit of albumin is g / L, mALBI grade 1: ALBI ≤ -2.60; mALBI grade 2a: -2.60 < ALBI ≤ -2.00; mALBI grade 2b: -2.00 < ALBI ≤ -1.39; mALBI grade 3: ALBI > -1.39;

[0087] The described multi-phase enhanced CT imaging data include the maximum tumor diameter, morphology, liver capsule invasion; capsule, low enhancement in the arterial phase, low-density ring sign, intratumoral arterial visualization, intrahepatic metastasis, high perfusion in the arterial phase around the tumor, major vessel invasion, and blood supply situation (arterial, portal, dual blood supply or atypical).

[0088] The described multi-phase enhanced CT imaging data include: the maximum tumor diameter, morphology, liver capsule invasion; capsule, low enhancement in the arterial phase, low-density ring sign, intratumoral arterial visualization, intrahepatic metastasis, high perfusion in the arterial phase around the tumor, major vessel invasion, and blood supply situation.

[0089] S102: Screen the processed liver data of the prediction object to obtain the radiomics feature indexes associated with the degree of microvascular invasion in liver cancer, and input the radiomics feature indexes associated with the degree of microvascular invasion in liver cancer into the pre-established prediction model of microvascular invasion in liver cancer, and output the trained prediction model of microvascular invasion in liver cancer;

[0090] Select a part of the collected data as the training group. Here, the Second Affiliated Hospital of Anhui Medical University is used as the training center, and the First Affiliated Hospital of Anhui Medical University is used as the model verification center;

[0091] After that, relevant analysis is carried out in the training group to screen out the radiomics feature indexes with strong robustness;

[0092] After that, further refined screening is carried out on the selected indexes through lasso regression. The cost function of 5-fold cross-validation of the lasso regression is:

[0093]

[0094] is the mean squared error term, m is the number of samples, y i is the actual value of the i-th sample, h θ (x i ) is the predicted value of the model for the i-th sample;

[0095] is the L1 regularization term, λ is the regularization parameter, θ j is the j-th parameter of the model, and n is the total number of parameters;

[0096] During the 5-fold cross-validation process, the data set is divided into 5 subsets. Each time, 4 of the subsets are used for training, and the remaining 1 subset is used for verification. In this way, the model can be trained and verified multiple times to evaluate the performance of the model on different data subsets. Finally, by comparing the model performance under different regularization parameters λ, the λ that minimizes the cost function is selected as the optimal parameter;

[0097] After that, the process of screening multi-phase CT enhanced radiomics data based on 5-fold cross-validation of lasso regression is as Figures 2 - 10 In it, 5-fold cross-validation of lasso regression is used to screen non-zero coefficient variables as the input variables for the final modeling. The abscissa is all log lambda (logλ), and the ordinate is the binomial deviance and the coefficient value.

[0098] S103: The trained prediction model for hepatocellular carcinoma microvascular invasion is tested and compared by calculating the AUC value, and the optimal prediction model for hepatocellular carcinoma microvascular invasion is determined according to the comparison results, and the prediction of hepatocellular carcinoma microvascular invasion is realized through the optimal prediction model for hepatocellular carcinoma microvascular invasion.

[0099] Use the Logistic regression algorithm to train the selected indicators and fit the microvascular invasion status of hepatocellular carcinoma; Logistic regression is a statistical model widely used in binary classification problems. It makes predictions by estimating the probability of an event occurring. Compared with other models, Logistic regression has the advantages of simple model, easy to interpret, and high computational efficiency, and is suitable for processing medium-sized data sets. Logistic regression determines the model parameters through maximum likelihood estimation to make the predicted probability closest to the actual occurrence probability. To improve the prediction performance and generalization ability of the model, the model can be regularized, such as L1 regularization for feature selection and L2 regularization for preventing overfitting. Set the regularization strength parameter to 0.1; to further optimize the model, the features can be standardized to eliminate the influence of different feature dimensions and make the model more stable; among them, the steps of constructing a prediction model for hepatocellular carcinoma microvascular invasion based on multiple clinical data using the Logistic regression algorithm are as follows:

[0100] Divide the data set of radiomics feature indicators related to the degree of microvascular invasion of hepatocellular carcinoma into a training sample set and a test sample set;

[0101] Use the Logistic regression method to establish a prediction model for hepatocellular carcinoma microvascular invasion, and set the model parameters, including the regularization strength and whether to perform feature standardization;

[0102] Input the training sample set into the prediction model for hepatocellular carcinoma microvascular invasion to complete the training of the model;

[0103] Input the test sample set into the prediction model for hepatocellular carcinoma microvascular invasion and output the relevant data of the microvascular invasion status of hepatocellular carcinoma;

[0104] Establish a receiver operating characteristic curve ROC based on the relevant data of the microvascular invasion status of hepatocellular carcinoma.

[0105] Substitute the established model into the validation center for validation, and at the same time test and compare multiple models;

[0106] Use the AUC value for effect evaluation and comparison between models;

[0107] As Figure 11 shown, although the AUC value of the intratumoral + peritumoral arterial phase radiomics model is the highest, the AUC value of the intratumoral arterial phase radiomics model has no difference from it, and it has the advantages of simpler implementation and fewer feature numbers;

[0108] After that, the best radiomics prediction model was optimally selected as the intratumoral omics feature data in the arterial phase, and the Radscore calculation method is as follows:

[0109] Radscore = log.sigma.1.mm.3D_firstorder_Variance × 0.1 + log.sigma.1.mm.3D_glszm_LowGrayLevelZoneEmphasis × -0.207 + wavelet.HH - glcm_InverseVariance × -0.228 + exponential_ngtdm_Contrast × -0.025 + wavelet.LL_glrlm_RunVariance × 0.284 + wavelet.LH_gldm_LargeDependenceLowGrayLevelEmphasis × -0.053 + squareroot_glszm_GrayLevelNonUniformity × 0.152 + firstorder_RootMeanSquared × -0.206 + wavelet.HL_glrlm_RunEntropy × 0.161 - 0.569.

[0110] Among them, firstorder_RootMeanSquared and log.sigma.1.mm.3D_firstorder_Variance are first-order statistical features, log.sigma.1.mm.3D_glszm_LowGrayLevelZoneEmphasis and squareroot_glszm_GrayLevelNonUniformity are gray-level run length matrix features, wavelet.HH-glcm_InverseVariance is a gray-level co-occurrence matrix feature, wavelet.LH_gldm_LargeDependenceLowGrayLevelEmphasis is a gray-level dependence matrix feature, wavelet.LL_glrlm_RunVariance and wavelet.HL_glrlm_RunEntropy are gray-level run length matrix features, exponential_ngtdm_Contrast is a gray-level neighborhood difference matrix feature, and wavelet.LL_glrlm_RunVariance, wavelet.HH-glcm_InverseVariance, wavelet.LH_gldm_LargeDependenceLowGrayLevelEmphasis, and wavelet.HL_glrlm_RunEntropy are wavelet transform features.

[0111] As Figures 14 - 18 shown.

[0112] Figures 14 - 15 is the receiver operating characteristic curve (ROC) graph, which is used to evaluate the classification performance of different combined models of the best radiomics data, clinical, and imaging feature data. It can be considered that the higher the curve, the better the ability. Figure 14 is the training group, Figure 15 is the validation group;

[0113] Figure 16 The decision curve analysis (DCA) graph is used to compare the benefits provided by different combined models of the best radiomics data, clinical, and imaging feature data. It can be considered that the higher the curve, the higher the benefit;

[0114] Confirmation of the best combined model:

[0115] Figure 17It is a Confusion Matrix diagram, which respectively shows the relationship between the model prediction results and the actual results in the form of a matrix for the best combined model training group and the validation group. Through the confusion matrix diagram, it can be clearly seen in which categories the model makes mistakes, such as false positives (predicting negative classes as positive classes) and false negatives (predicting positive classes as negative classes). The diagonal elements of the confusion matrix diagram represent the number of samples correctly classified by the model. The larger the value on the diagonal, the higher the classification accuracy of the model in that category;

[0116] Feature screening process:

[0117] Data grouping: A part of the data collected from a certain hospital is selected as the training group, and a part of the data collected from another hospital is used as the validation group;

[0118] Screening radiomics features with strong robustness: Using between-group and within-group consistency analysis, radiomics features with an ICC value of the correlation coefficient greater than 0.95 are screened out for further construction of a logistic regression radiomics model;

[0119] Screening clinical and imaging feature indicators: In the training group, those with a P value less than 0.1 are included in the multivariate logistic regression analysis through univariate logistic regression analysis first, and finally, the feature indicators with a P value less than 0.05 are included as meaningful feature indicators in the construction of the Catboost machine learning prediction model;

[0120] Using collinearity analysis to check whether there is collinearity in the finally included feature indicators (Variance Inflation Factor < 2, Tollerence > 0.5 is regarded as no collinearity);

[0121] Based on Shapley additive explanations, global and local explanations are made for this machine learning model, specifically:

[0122] Figure 19 The summary plot of the global explanation of the machine learning model by the Shapley additive explanation of the optimal combined model based on the Catboost machine learning algorithm involved in the present invention is shown. The vertical axis arranges each feature in the model from high to low according to the importance of its influence on the prediction result. A point in the figure represents a sample; the redder the point, the larger the original value, the bluer the point, the smaller the value, and purple represents the intermediate value; the horizontal axis value represents the SHAP value. A positive value represents a positive influence on the prediction result, and a negative value represents a negative influence; the larger the SHAP value, the greater the influence on the model and the more important it is;

[0123] Figures 20 - 21The waterfall plot and force plot of the Shapley additive explanation of the optimal joint model based on the Catboost machine learning algorithm for local interpretation of the machine learning model involved in the present invention illustrate the prediction process of a sample in the model. Red represents a positive push to the prediction result, and blue represents a negative push. The width of each feature represents the degree of influence.

[0124] Specifically:

[0125] Figure 20 The figure shows the results of the feature importance analysis of a prediction model, specifically the contribution of each feature to the model prediction value f(x). Each bar in the chart represents the influence of a feature on the prediction value, and the length and direction of the bar represent the magnitude and direction of the influence. f(x) = -1.101: This is the prediction value of the model, representing the prediction result under the current feature combination. The contribution of Radscore to the prediction value is -1.36. In this example, the value of Radscore is low, indicating that this feature has a greater negative impact on the prediction value, resulting in a significant decrease in the prediction value. The contribution value of Liver_capsule_invasion (hepatic capsule invasion) is -0.39. In this example, it means that this feature has a moderate negative impact on the prediction value, and the presence of hepatic capsule invasion causes the prediction value to decrease. The contribution value of Intrahepatic_metastasis (intrahepatic metastasis) is -0.23. In this example, it means that this feature has a relatively small negative impact on the prediction value, and the presence of intrahepatic metastasis causes a slight decrease in the prediction value. The expected value E[f(X)] = 0.876, which is the average prediction value of the model on all training data, representing the baseline prediction value when there is no influence of specific features. Through the contribution values of these features, it can be clearly seen that Radscore has the greatest impact on the prediction value, followed by hepatic capsule invasion and intrahepatic metastasis. The combined influence of these features results in a final prediction value of -1.101, which is lower than the expected value of 0.876. This indicates that under the current feature combination, the model predicts a relatively low recurrence risk.

[0126] Figure 21Shows the results of the feature importance analysis of another prediction model, specifically showing the contribution of each feature to the model's predicted value f(x). Each bar represents the impact of a feature on the predicted value, and its length and direction represent the magnitude and direction of the impact respectively. The predicted value f(x) of the model is -1.101, which is the prediction result under the current feature combination. Among them, the contribution value of Radscore is -1.36, which is relatively low and has a greater negative impact on the predicted value, resulting in a significant decrease in the predicted value. The contribution value of Liver_capsule_invasion (hepatic capsule invasion) is -0.39, with a moderate negative impact on the predicted value, resulting in a decrease in the predicted value. The contribution value of Intrahepatic_metastasis (intrahepatic metastasis) is -0.23, with a relatively small negative impact on the predicted value, resulting in a slight decrease in the predicted value. The expected value E[f(X)] is 0.876, which is the average predicted value of the model on all training data, representing the baseline predicted value when there is no influence from specific features. From the contribution values of these features, it can be seen that Radscore has the greatest impact on the predicted value, followed by hepatic capsule invasion and intrahepatic metastasis. The combined impact of these features makes the final predicted value -1.101, which is lower than the expected value 0.876, indicating that under the current feature combination, the recurrence risk predicted by the model is relatively low.

[0127] Example 2: Second aspect, to achieve the above object, the present invention discloses a hepatocellular carcinoma microvascular invasion prediction system, including:

[0128] A data processing module 11, configured to obtain the liver data of the prediction object, preprocess the liver data of the prediction object, and obtain the processed liver data of the prediction object, wherein the liver data of the prediction object includes: the pathology of the prediction object, multi-phase enhanced imaging information, and liver-related data;

[0129] An index screening module 12, configured to screen the processed liver data of the prediction object to obtain imaging feature indexes associated with the degree of hepatocellular carcinoma microvascular invasion, and input the imaging feature indexes associated with the degree of hepatocellular carcinoma microvascular invasion into a pre-established hepatocellular carcinoma microvascular invasion prediction model, and output a trained hepatocellular carcinoma microvascular invasion prediction model;

[0130] A prediction module 13, configured to test and compare the trained hepatocellular carcinoma microvascular invasion prediction model by calculating the AUC value, determine the optimal hepatocellular carcinoma microvascular invasion prediction model according to the comparison result, and realize the prediction of hepatocellular carcinoma microvascular invasion through the optimal hepatocellular carcinoma microvascular invasion prediction model.

[0131] Based on the same inventive concept, the present invention also provides a computer device, which includes: one or more processors, and a memory for storing one or more computer programs; the program includes program instructions, and the processor is configured to execute the program instructions stored in the memory. The processor may be a Central Processing Unit (CPU), or may also be other general-purpose processors, Digital Signal Processors (DSPs), Application Specific Integrated Circuits (ASICs), Field-Programmable Gate Arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. It is the computing core and control core of the terminal, and is used to implement one or more instructions. Specifically, it is used to load and execute one or more instructions in the computer storage medium to implement the above method.

[0132] It should be further noted that, based on the same inventive concept, the present invention also provides a computer storage medium, on which a computer program is stored, and the computer program, when run by a processor, executes the above method. The storage medium may adopt any combination of one or more computer-readable media. The computer-readable medium may be a computer-readable signal medium or a computer-readable storage medium. The computer-readable storage medium may, for example, but not be limited to, be an electrical, magnetic, optical, electrical, magnetic, infrared, or semiconductor system, apparatus, or device, or any combination of the above. More specific examples (non-exhaustive list) of the computer-readable storage medium include: an electrical connection having one or more wires, a portable computer disk, a hard disk, a Random Access Memory (RAM), a Read Only Memory (ROM), an Erasable Programmable Read Only Memory (EPROM or flash memory), an optical fiber, a portable compact disk read only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present invention, the computer-readable storage medium may be any tangible medium that contains or stores a program, and the program may be used by or combined with an instruction execution system, apparatus, or device.

[0133] In the description of this specification, the descriptions referring to terms such as "one embodiment", "example", "specific example", etc. mean that the specific features, structures, materials, or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present disclosure. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described may be combined in any one or more embodiments or examples in a suitable manner.

[0134] The above shows and describes the basic principles, main features and advantages of the present disclosure. Those skilled in the art should understand that the present disclosure is not limited by the above embodiments. What is described in the above embodiments and the specification only illustrates the principles of the present disclosure. Without departing from the spirit and scope of the present disclosure, the present disclosure will have various changes and improvements, and these changes and improvements all fall within the scope of the present disclosure claimed.

Claims

1. A method for predicting microvascular infiltration of liver cancer, characterized in that: The method comprises the following steps: Acquiring liver data of the predicted object, preprocessing the liver data of the predicted object to obtain processed liver data of the predicted object, wherein the liver data of the predicted object includes: pathology of the predicted object, multi-phase enhanced imaging information and liver-related data; The processed liver data of the prediction object are screened to obtain imaging omics feature indicators associated with the degree of microvascular infiltration of liver cancer, and the imaging omics feature indicators associated with the degree of microvascular infiltration of liver cancer are input into a pre-established liver cancer microvascular infiltration prediction model, and the trained liver cancer microvascular infiltration prediction model is output; The trained liver cancer microvascular invasion prediction model was tested and compared by calculating the AUC value, and the optimal liver cancer microvascular invasion prediction model was determined according to the comparison results. The prediction of liver cancer microvascular invasion was achieved through the optimal liver cancer microvascular invasion prediction model.

2. A method for predicting liver cancer microvascular infiltration according to claim 1, characterized in that: The predicted object liver data specifically include: basic information of the patient, liver cancer microvascular infiltration pathological data, multi-phase enhanced CT imaging data, clinical and serum laboratory test data.

3. The method for predicting microvascular invasion of liver cancer according to claim 1, characterized in that: The process of preprocessing the liver data of the predicted object: The final extracted imaging features are normalized, and the normalization formula is as follows: Xn normalized=(Xn–Xmin) / (Xmax–Xmin); Xn represents any numeric variable, Xn normalized represents the normalized value of the numeric variable Xn, Xmax represents the maximum value of the numeric variable, and Xmin represents the minimum value of the numeric variable; The continuous variables in clinical and imaging characteristics were binarized, and the expression of the restricted cubic spline function was as follows: RCS(X)=β0X+β1S1+…+β k-2 S k-2 Where Si is the cubic component falling in the i-th node.

4. A method for predicting microvascular invasion of liver cancer according to claim 2, characterized in that: The multi-phase enhanced CT imaging data include: maximum diameter, morphology, liver capsule invasion of the tumor; capsule, low enhancement in the arterial phase, low-density ring sign, intratumoral arterial development, intrahepatic metastasis, peritumoral arterial phase high perfusion, large vessel invasion and blood supply.

5. The method for predicting liver cancer microvascular infiltration according to claim 1, characterized in that: The processed prediction object liver data is screened by using lasso regression to further fine-screen the screened indicators. The cost function of lasso regression 5-fold cross validation is: In the formula, is the mean square error term, m is the number of samples, and y i is the actual value of the i-th sample, h θ (x i ) is the model's predicted value for the i-th sample; is the L1 regularization term, λ is the regularization parameter, θ j is the jth parameter of the model and n is the total number of parameters.

6. The method for predicting microvascular invasion of liver cancer according to claim 1, characterized in that: The process of inputting the imaging genomics characteristic index associated with the degree of liver cancer microvascular infiltration into a pre-established liver cancer microvascular infiltration prediction model and outputting a trained liver cancer microvascular infiltration prediction model is as follows: The data set of imaging omics feature indicators associated with the degree of microvascular invasion of liver cancer is divided into a training sample set and a test sample set; Logistic regression method was used to establish a prediction model for liver cancer microvascular invasion, and the parameters of the prediction model for liver cancer microvascular invasion were set, including regularization strength and whether to perform feature standardization. Input the training sample set into the prediction model of liver cancer microvascular infiltration to complete the training of the model; Input the test sample set into the liver cancer microvascular invasion prediction model, and output the relevant data of the liver cancer microvascular invasion status; The receiver operating characteristic (ROC) curve was established based on the data related to the microvascular invasion status of liver cancer; The pre-established prediction model for liver cancer microvascular invasion is as follows: in: represents the predicted output of the model, which is the probability of microvascular invasion of liver cancer; σ is the Logistic function, defined as: β0 is the intercept term, β n is the regression coefficient of the model, which indicates the influence weight of each feature on the prediction result. n is the input feature, i.e., the imaging genomics feature index associated with the degree of microvascular infiltration of liver cancer, and the regression coefficient β n Represents each feature X n The degree of influence on the microvascular invasion status of liver cancer: a positive regression coefficient indicates that when the characteristic value increases, the probability of invasion increases; a negative regression coefficient indicates that when the characteristic value increases, the probability of invasion decreases.

7. The method for predicting liver cancer microvascular infiltration according to claim 1, characterized in that: When the prediction of liver cancer microvascular invasion is achieved by using the optimal liver cancer microvascular invasion prediction model, the optimal liver cancer microvascular invasion prediction model is selected as the intratumor omics feature data in the arterial phase, and the Radscore calculation method is as follows: Radscore=log.sigma.1.mm.3D_firstorder_Variance×0.1+log.sigma.1.mm.3D_glszm_LowGrayLevelZoneEmphasis×-0.207+wavelet.HH-glcm_InverseVariance×-0.228+exponential_ngtdm_Contrast×-0.025+wavelet.LL_glrlm _RunVariance×0.284+wavelet.LH_gldm_LargeDependenceLowGrayLevelEmphasis×-0.053+squareroot_glszm_GrayLevelNonUniformity×0.152+firstorder_RootMeanSquared×-0.206+wavelet.HL_glrlm_RunEntropy×0.161-0.569 Where firstorder_RootMeanSquared and log.sigma.1.mm.3D_firstorder_Variance are first-order statistical features, log.sigma.1.mm.3D_glszm_LowGrayLevelZoneEmphasis and squareroot_glszm_GrayLevelNonUniformity are gray-level run-length matrix features, wavelet.HH-glcm_InverseVariance is the gray-level co-occurrence matrix feature, wavelet.LH_gldm_LargeDependenceLowGrayLevelEmphasi s is the grayscale dependency matrix feature, wavelet.LL_glrlm_RunVariance and wavelet.HL_glrlm_RunEntropy are the grayscale run length matrix features, exponential_ngtdm_Contrast is the grayscale neighborhood difference matrix feature, wavelet.LL_glrlm_RunVariance, wavelet.HH-glcm_InverseVariance, wavelet.LH_gldm_LargeDependenceLowGrayLevelEmphasis, and wavelet.HL_glrlm_RunEntropy are wavelet transform features.

8. A liver cancer microvascular infiltration prediction system, characterized in that: include: A data processing module, used to obtain the predicted object liver data, pre-process the predicted object liver data, and obtain the processed predicted object liver data, wherein the predicted object liver data includes: predicted object pathology, multi-phase enhanced imaging information and liver-related data; An index screening module is used to screen the processed liver data of the prediction object to obtain the imaging genomics feature index associated with the degree of microvascular infiltration of liver cancer, input the imaging genomics feature index associated with the degree of microvascular infiltration of liver cancer into a pre-established liver cancer microvascular infiltration prediction model, and output a trained liver cancer microvascular infiltration prediction model; The prediction module is used to test and compare the trained liver cancer microvascular infiltration prediction model by calculating the AUC value, determine the optimal liver cancer microvascular infiltration prediction model according to the comparison result, and realize the prediction of liver cancer microvascular infiltration through the optimal liver cancer microvascular infiltration prediction model.

9. A terminal device, comprising a memory, a processor, and a computer program stored in the memory and capable of running on the processor, characterized in that: The memory stores a computer program that can be run on the processor. When the processor loads and executes the computer program, a method for predicting liver cancer microvascular infiltration according to any one of claims 1 to 7 is adopted.

10. A computer-readable storage medium having a computer program stored therein, characterized in that: When the computer program is loaded and executed by the processor, a method for predicting liver cancer microvascular infiltration according to any one of claims 1 to 7 is adopted.

Images