A method, system, device and medium for predicting the efficacy of anti-HER-2 treatment for liver metastasis of breast cancer

Abstract

The present invention provides a method, system, device and medium for predicting the efficacy of anti-HER‑2 treatment of liver metastasis of breast cancer, which belongs to the field of medical technology. The method comprises: obtaining an abdominal enhanced CT image to be identified of a subject, i.e., a patient with liver metastasis of breast cancer treated with anti-HER‑2; inputting the abdominal enhanced CT image to be identified into an efficacy prediction model, wherein the efficacy prediction model performs a quantitative analysis operation on the abdominal enhanced CT image to be identified, and obtains the efficacy prediction result of anti-HER‑2 treatment of liver metastasis of breast cancer of the subject. The present invention studies for the first time the value of imaging genomics features based on abdominal enhanced CT in predicting the efficacy of anti-HER‑2 treatment of BCLM, thereby providing reference information and relevant clinical guidance for the prediction of the efficacy of anti-HER‑2 treatment of BCLM patients with HER‑2 positive BCLM and the overall treatment plan.

Description

Technical Field

[0001] The present invention belongs to the field of medical technology, and in particular relates to a method, system, device and medium for predicting the efficacy of anti-HER-2 treatment of liver metastasis of breast cancer. Background Art

[0002] The information disclosed in this background technology section is only intended to enhance the understanding of the overall background of the invention, and should not necessarily be regarded as an admission or any form of suggestion that the information constitutes the prior art already known to a person skilled in the art.

[0003] Breast cancer is the most common malignant tumor in women worldwide. Its various treatment methods have always attracted people's attention. Studies have shown that about one-third of breast cancer patients will have distant non-lymph node metastasis. Once distant metastasis occurs, the 5-year survival rate drops to 23%. Common metastatic sites of breast cancer include bones, lungs, liver, and brain. Among them, about 50% of metastatic breast cancer (MBC) patients have liver metastasis, and their natural survival period is only 4 to 8 months. It is one of the main causes of death in advanced breast cancer (ABC). Therefore, breast cancer liver metastasis (BCLM) has always been one of the hot spots and difficulties in MBC research. Improving the treatment effect of BCLM, optimizing treatment plans, and improving patient prognosis are currently a major clinical problem. BCLM has made certain progress in chemotherapy, targeted and endocrine therapy, but the current treatment plan still has limited benefits, and the patient's prognosis is relatively poor. The average overall survival of BCLM is only 3 years. In addition, breast cancer is a malignant tumor with high heterogeneity at the molecular level. Breast cancer patients with different molecular subtypes have significant differences in response to clinical treatment methods and prognosis. HER-2 positive breast cancer is a subtype of breast cancer, which is associated with high tumor invasiveness, high risk of recurrence, rapid progression, and poor prognosis. It is an independent adverse prognostic factor for breast cancer. With the development of breast cancer treatment, the application of anti-HER-2 targeted drugs has greatly prolonged the survival of HER-2 positive breast cancer patients. As the first humanized monoclonal antibody targeting HER-2, the advent of trastuzumab has changed the prognosis of HER-2 positive breast cancer patients, affected the diagnosis and treatment mode of breast cancer, and is an important breakthrough in breast cancer drug treatment. Other anti-HER-2 drugs such as pyrotinib and lapatinib have also been proven in clinical trials to significantly prolong the survival of patients with advanced HER-2 positive breast cancer. However, due to the heterogeneity between and within tumors, the efficacy of effective treatments may vary greatly among different cancer patients. Some patients can achieve therapeutic effects greater than median progression-free survival (mPFS) and median overall survival (mOS) using anti-HER-2 drugs such as trastuzumab, pyrotinib, and lapatinib. However, there are some patients with the same molecular classification who also have poor therapeutic effects using these drugs and cannot achieve mPFS or mOS. What affects the difference in therapeutic effects between these two groups of patients has aroused our curiosity.

[0004] In recent years, artificial intelligence, especially radiomics, has developed rapidly, providing new ideas and new tools for medical research. Radiomics is an emerging technology that extracts a large amount of imaging information from images (CT, MRI, PET, etc.) with high throughput to achieve tumor segmentation, feature extraction and model building. It can indirectly reflect tumor heterogeneity that cannot be identified by the naked eye, discover the correlation between quantitative data and pathological phenotypes, and perform non-invasive evaluation of the entire tumor. In addition, some studies have confirmed that radiomics features can be used as indicators of disease prognosis. However, whether radiomics can be used to discover some potential features that cannot be obtained on a visual basis from breast cancer liver metastases to predict the efficacy of anti-HER-2 treatment in patients has not been reported in the literature. Summary of the invention

[0005] The present invention provides a method, system, device and medium for predicting the efficacy of anti-HER-2 treatment of breast cancer liver metastasis. The present invention is the first to study the value of imaging genomics features based on abdominal enhanced CT in predicting the efficacy of anti-HER-2 treatment of BCLM, and provides reference information and relevant clinical guidance for the prediction of anti-HER-2 treatment efficacy and overall treatment plan for HER-2 positive BCLM patients.

[0006] In order to achieve the above object, the present invention adopts the following technical solutions:

[0007] The first aspect of the present invention provides a method for predicting the efficacy of anti-HER-2 treatment for liver metastasis of breast cancer, the method comprising:

[0008] Obtaining an abdominal enhanced CT image to be identified from a subject, i.e., a patient with liver metastasis of breast cancer treated with anti-HER-2 therapy;

[0009] Inputting the to-be-identified abdominal enhanced CT image into the efficacy prediction model, wherein the efficacy prediction model performs a quantitative analysis operation on the to-be-identified abdominal enhanced CT image to obtain a prediction result of the efficacy of anti-HER-2 treatment for breast cancer liver metastasis of the subject;

[0010] The efficacy prediction model is obtained by training the model using a machine learning algorithm using the pre-collected imaging genomics-related feature data of patients with liver metastasis of breast cancer treated with anti-HER-2 therapy.

[0011] Furthermore, the specific method for constructing the efficacy prediction model includes:

[0012] S1. Evaluate and outline the lesion area on the patient's abdominal enhanced CT image to obtain the region of interest (ROI);

[0013] S2. Extract the radiomics features in the ROI and construct a radiomics model; perform data training and testing on the radiomics model, and use the trained model as a therapeutic efficacy prediction model.

[0014] A second aspect of the present invention provides a therapeutic efficacy prediction system for anti-HER-2 treatment of breast cancer liver metastasis, and the therapeutic efficacy prediction system may include:

[0015] A data acquisition module, used for acquiring an abdominal enhanced CT image to be identified of a subject, i.e., a patient with liver metastasis of breast cancer treated with anti-HER-2 therapy;

[0016] Data processing and output module: used to input the abdominal enhanced CT image to be identified into the efficacy prediction model, the efficacy prediction model performs quantitative analysis operations on the abdominal enhanced CT image to be identified, and outputs the efficacy prediction results of anti-HER-2 treatment of breast cancer liver metastasis of the subject.

[0017] Among them, in the data processing and output module, the efficacy prediction model is obtained by using a machine learning algorithm to train the model using the pre-collected imaging omics-related feature data of patients with liver metastasis of breast cancer treated with anti-HER-2.

[0018] The third aspect of the present invention provides an electronic device, comprising a memory and a processor, and computer instructions stored in the memory and executed on the processor, wherein when the computer instructions are executed by the processor, the steps of the above-mentioned method for predicting the efficacy of anti-HER-2 treatment of liver metastasis of breast cancer are completed.

[0019] A fourth aspect of the present invention provides a computer-readable storage medium for storing computer instructions, which, when executed by a processor, complete the steps of the above-mentioned method for predicting the efficacy of anti-HER-2 treatment of breast cancer liver metastasis.

[0020] Beneficial technical effects of one or more of the above technical solutions:

[0021] The above technical scheme is the first to use the radiomics features of abdominal enhanced CT to predict the efficacy of anti-HER-2 treatment of BCLM. Experiments have shown that the radiomics model established based on texture features is more accurate and objective. Radiomics can become an objective, precise, simple and easy non-invasive tool to judge the efficacy of anti-HER-2 treatment in patients with breast cancer liver metastasis. Among them, the RF algorithm model of the portal venous phase of enhanced CT has the best classification efficiency, thereby providing reference information and relevant clinical guidance for the prediction of anti-HER-2 treatment efficacy and overall treatment plan for HER-2-positive BCLM patients. Therefore, it has good practical application value. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] The accompanying drawings in the specification, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0023] Figure 1 Schematic diagram of ROI delineation in Example 1 of the present invention, (a) is the ROI in the arterial phase, (b) is the ROI in the portal venous phase, and (c) is the ROI in the delayed phase.

[0024] Figure 2 This is a horizontal bar chart of variance thresholds for feature selection in Example 1 of the present invention. The dark grey bars represent all the extracted radiomics features, the light grey bars represent the radiomics features after screening by the variance threshold method, and the ordinate represents 15 types of filtering methods (variance threshold = 0.8).

[0025] Figure 3 This is a horizontal bar chart of the best K feature selection in Example 1 of the present invention. The horizontal axis is the p-value of the feature, and the vertical axis is the features with a p-value < 0.05 screened out by the SelectKBest method.

[0026] Figure 4 Schematic diagram of feature screening using the Lasso method in Example 1 of the present invention; wherein, (a) Lasso path, the horizontal axis is the log value of α, and the vertical axis represents the coefficient of the feature; (b) MSE path, the horizontal axis represents the log value of α, and the vertical axis represents the mean square error; (c) Lasso model regression coefficient, the horizontal axis represents the regression coefficient, and the vertical axis represents the screened feature.

[0027] Figure 5 The test set ROC curve of the portal venous phase RF model in Example 1 of the present invention is shown in Figure 1. The light gray curve is the poor prognosis group, and the dark gray curve is the good prognosis group. The AUC value of the poor prognosis group is 0.889 (sensitivity and specificity are 0.70 and 0.78, respectively), and the AUC value of the good prognosis group is 0.889 (sensitivity and specificity are 0.78 and 0.70, respectively). DETAILED DESCRIPTION

[0028] It should be noted that the following detailed descriptions are all illustrative and intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present invention belongs.

[0029] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components and / or combinations thereof. It should be understood that the scope of protection of the present invention is not limited to the specific embodiments described below; it should also be understood that the terms used in the embodiments of the present invention are intended to describe specific embodiments, not to limit the scope of protection of the present invention. If the experimental methods of the specific conditions are not indicated in the following specific embodiments, they are generally in accordance with conventional methods and conditions within the art.

[0030] Specifically, in a typical embodiment of the present invention, a method for predicting the efficacy of anti-HER-2 treatment for liver metastasis of breast cancer is provided, the method comprising:

[0031] Obtaining an abdominal enhanced CT image to be identified from a subject, i.e., a patient with liver metastasis of breast cancer treated with anti-HER-2 therapy;

[0032] Inputting the to-be-identified abdominal enhanced CT image into the efficacy prediction model, wherein the efficacy prediction model performs a quantitative analysis operation on the to-be-identified abdominal enhanced CT image to obtain a prediction result of the efficacy of anti-HER-2 treatment for breast cancer liver metastasis of the subject;

[0033] The efficacy prediction model is obtained by training the model using a machine learning algorithm using the pre-collected imaging genomics-related feature data of patients with liver metastasis of breast cancer treated with anti-HER-2 therapy.

[0034] In another specific embodiment of the present invention, the specific method for constructing the efficacy prediction model includes:

[0035] S1. Evaluate and outline the lesion area on the patient's abdominal enhanced CT image to obtain the region of interest (ROI);

[0036] S2. Extract the radiomics features in the ROI and build a radiomics model; perform data training and testing on the radiomics model, and use the trained model as a therapeutic effect prediction model;

[0037] In another specific implementation of the present invention, in step S1,

[0038] The evaluation and delineation can be performed manually, for example, by at least two senior radiologists using a double-blind method, and the second senior physician will review the delineation after the first physician has finished the delineation, thereby ensuring the accuracy of the delineation result; of course, the evaluation and delineation can also be performed using a related delineation device, thereby identifying and delineating the lesion area of ​​the image, and outputting the delineated image;

[0039] In another specific embodiment of the present invention, the step S1 further includes further processing the obtained ROI, such as grayscale normalization, to reduce the impact of contrast and brightness changes. In a specific embodiment of the present invention, the obtained ROI includes an arterial phase ROI, a portal venous phase ROI, and a delayed phase ROI;

[0040] In another specific embodiment of the present invention, in step S2, extracting radiomics features in ROI includes quantifying ROI to obtain ROI features and screening to obtain effective radiomics features;

[0041] In another specific embodiment of the present invention, the ROI is quantified, including extracting quantitative imaging features from CT images using the Radcloud platform; grouping the quantitative imaging features, and the grouping method can be descriptors, shape and size features, and texture features; then the ROI is quantified using the intra-class correlation coefficient (ICC), and the quantified ROI can be screened using a variance threshold, SelectKBest or LASSO model to extract effective imaging omics features;

[0042] Among them, the descriptor quantitatively describes the distribution of voxel intensity in CT images through commonly used basic metrics; the shape and size-based features can be three-dimensional features that reflect the shape and size of the region; the texture features are obtained by quantifying the regional heterogeneity differences based on the gray level run and gray level co-occurrence texture matrix, and the texture features describe the repeated local patterns in the image and their arrangement rules, including gray level co-occurrence matrix (GLCM) features, gray level run length matrix (GLRLM) features, and gray level size zone matrix (GLSZM) features, etc. 75 features. Preferably, 14 filters such as exponential, logarithmic, gradient, square value, square root, lbp-2D and wavelet (wavelet-LHL, wavelet-LHH, wavelet-HLL, wavelet-LLH, wavelet-HLH, wavelet-HHH, wavelet-HHL, wavelet-LLL) can be used to filter the image and analyze the texture at a finer scale.

[0043] The radiomics model can be obtained by using a classifier model, and the classifier model includes but is not limited to KNN, SVM, XGBoost, RF, LR and DT6;

[0044] More specifically, the abdominal enhanced CT images of patients with ROI outlined are randomly divided into a training group and a test group, so as to train and test the imaging omics model, and the trained model is used as a therapeutic effect prediction model. At the same time, the receiver operating characteristic (ROC) curve, that is, the area under the curve (AUC), can be used to evaluate the prediction performance. Specifically, the prediction efficiency of the model with 0.5<AUC≤0.7 is low, the prediction efficiency of the model with 0.7<AUC≤0.9 is medium, and the prediction efficiency of the model with AUC>0.9 is high. Further preferably, four indicators can be used to evaluate the performance of the model, including precision (P) (referring to the proportion of all samples predicted to be correct), recall (R) (referring to the proportion of samples predicted to be correct), F1 score (F1-score) (based on the harmonic mean of precision and recall), and support (the total number of samples involved).

[0045] In a specific embodiment of the present invention, random grouping is used to evaluate the accuracy of the test set results of the machine learning model classification. The average of the verification score numbers for each time is taken to establish an accuracy (score) matrix, so as to screen out the appropriate ROI and select the optimal machine learning model.

[0046] The study found that the AUC value of the SVM classifier in the arterial phase was the highest, which was 0.756, and the AUC value of the RF classifier in the delayed phase was 0.600, and the classification efficiency of the model was average; when analyzing the ROI in the portal venous phase, the AUC value was significantly improved, and the RF classifier had the highest AUC value of 0.889, and the misdiagnosis rate and missed diagnosis rate were reduced, which shows that CT images of liver metastases in the portal venous phase are meaningful in predicting the efficacy of anti-HER-2 drugs.

[0047] In another specific embodiment of the present invention, a system for predicting the efficacy of anti-HER-2 treatment for liver metastasis of breast cancer is provided, and the system for predicting the efficacy may include:

[0048] A data acquisition module, used for acquiring an abdominal enhanced CT image to be identified of a subject, i.e., a patient with liver metastasis of breast cancer treated with anti-HER-2 therapy;

[0049] Data processing and output module: used to input the abdominal enhanced CT image to be identified into the efficacy prediction model, the efficacy prediction model performs quantitative analysis operations on the abdominal enhanced CT image to be identified, and outputs the efficacy prediction results of anti-HER-2 treatment of breast cancer liver metastasis of the subject.

[0050] Among them, in the data processing and output module, the efficacy prediction model is obtained by using a machine learning algorithm to train the model using the pre-collected imaging omics-related feature data of patients with liver metastasis of breast cancer treated with anti-HER-2.

[0051] In another specific embodiment of the present invention, the specific method for constructing the efficacy prediction model includes:

[0052] S1. Evaluate and outline the lesion area on the patient's abdominal enhanced CT image to obtain the region of interest (ROI);

[0053] S2. Extract the radiomics features in the ROI and build a radiomics model; perform data training and testing on the radiomics model, and use the trained model as a therapeutic effect prediction model;

[0054] Wherein, in the step S1,

[0055] The evaluation and delineation can be performed manually, for example, by at least two senior radiologists using a double-blind method, and the second senior physician will review the delineation after the first physician has finished the delineation, thereby ensuring the accuracy of the delineation result; of course, the evaluation and delineation can also be performed using a related delineation device, thereby identifying and delineating the lesion area of ​​the image, and outputting the delineated image;

[0056] In another specific embodiment of the present invention, the step S1 further includes further processing the obtained ROI, such as grayscale normalization, to reduce the impact of contrast and brightness changes. In a specific embodiment of the present invention, the obtained ROI includes an arterial phase ROI, a portal venous phase ROI, and a delayed phase ROI;

[0057] In step S2, extracting radiomics features in ROI includes quantifying ROI to obtain ROI features and screening to obtain effective radiomics features;

[0058] More specifically, the ROI is quantified, including extracting quantitative imaging features from CT images using a Radcloud platform; grouping the quantitative imaging features, and the grouping method may be descriptors, shape and size features, and texture features; then the ROI is quantified using an intra-class correlation coefficient (ICC), and the quantified ROI may be screened using a variance threshold, SelectKBest, or LASSO model to extract effective imaging omics features;

[0059] Among them, the descriptor quantitatively describes the distribution of voxel intensity in CT images through commonly used basic metrics; the shape and size-based features can be three-dimensional features that reflect the shape and size of the region; the texture features are obtained by quantifying the regional heterogeneity differences based on the gray level run and gray level co-occurrence texture matrix, and the texture features describe the repeated local patterns in the image and their arrangement rules, including gray level co-occurrence matrix (GLCM) features, gray level run length matrix (GLRLM) features, and gray level size zone matrix (GLSZM) features, etc. 75 features. Preferably, 14 filters such as exponential, logarithmic, gradient, square value, square root, lbp-2D and wavelet (wavelet-LHL, wavelet-LHH, wavelet-HLL, wavelet-LLH, wavelet-HLH, wavelet-HHH, wavelet-HHL, wavelet-LLL) can be used to filter the image and analyze the texture at a finer scale.

[0060] The radiomics model can be obtained by using a classifier model, and the classifier model includes but is not limited to KNN, SVM, XGBoost, RF, LR and DT6;

[0061] More specifically, the abdominal enhanced CT images of patients with ROI outlined are randomly divided into a training group and a test group, so as to train and test the imaging omics model, and the trained model is used as a therapeutic effect prediction model. At the same time, the receiver operating characteristic (ROC) curve, that is, the area under the curve (AUC), can be used to evaluate the prediction performance. Specifically, the prediction efficiency of the model with 0.5<AUC≤0.7 is low, the prediction efficiency of the model with 0.7<AUC≤0.9 is medium, and the prediction efficiency of the model with AUC>0.9 is high. Further preferably, four indicators can be used to evaluate the performance of the model, including precision (P) (referring to the proportion of all samples predicted to be correct), recall (R) (referring to the proportion of samples predicted to be correct), F1 score (F1-score) (based on the harmonic mean of precision and recall), and support (the total number of samples involved).

[0062] In a specific embodiment of the present invention, random grouping is used to evaluate the accuracy of the test set results of the machine learning model classification. The average of the verification score numbers for each time is taken to establish an accuracy (score) matrix, so as to screen out the appropriate ROI and select the optimal machine learning model.

[0063] The study found that the AUC value of the SVM classifier in the arterial phase was the highest, which was 0.756, and the AUC value of the RF classifier in the delayed phase was 0.600, and the classification efficiency of the model was average; when analyzing the ROI in the portal venous phase, the AUC value was significantly improved, and the RF classifier had the highest AUC value of 0.889, and the misdiagnosis rate and missed diagnosis rate were reduced. Among them, the classification efficiency of the RF algorithm model in the portal venous phase of enhanced CT was the best.

[0064] In another specific embodiment of the present invention, an electronic device is provided, comprising a memory and a processor, and computer instructions stored in the memory and executed on the processor. When the computer instructions are executed by the processor, the steps of the above-mentioned method for predicting the efficacy of anti-HER-2 treatment of liver metastasis of breast cancer are completed.

[0065] In another specific embodiment of the present invention, a computer-readable storage medium is provided for storing computer instructions. When the computer instructions are executed by a processor, the steps of the above-mentioned method for predicting the efficacy of anti-HER-2 treatment of liver metastasis of breast cancer are completed.

[0066] In order to enable those skilled in the art to more clearly understand the technical solution of the present invention, the technical solution of the present invention will be described in detail below in conjunction with specific embodiments.

[0067] Example 1

[0068] Materials and Methods

[0069] 1. Research subjects

[0070] This study was approved by the Clinical Research Center of Qilu Hospital of Shandong University. Patients with breast cancer liver metastasis treated with anti-HER-2 therapy who had hospitalization records in the Central Campus of Qilu Hospital of Shandong University and the East Campus of Qilu Hospital of Shandong University from January 1, 2011 to November 20, 2021 were collected. All cases were confirmed by pathological diagnosis of primary or metastatic lesions, and the abdominal enhanced CT images of the enrolled patients were retrospectively analyzed. A total of 61 patients (0 males and 61 females; aged 31-71 years) were enrolled in this study. All patients regularly used anti-HER-2 drugs after liver metastasis (trastuzumab in 49 cases, pyrotinib in 6 cases, and lapatinib in 6 cases). Then, the 61 enrolled patients were divided into two groups with different prognoses based on the conclusions of existing important clinical trials (for MBC with HER-2 overexpression, the median PFS of chemotherapy combined with the anti-HER-2 monoclonal antibody drug trastuzumab was 7.4 months), namely, the poor prognosis group and the good prognosis group. They were stratified and randomly divided into a training group of 42 cases and a test group of 19 cases in a ratio of 7:3.

[0071] Inclusion criteria: 1. All cases were diagnosed with HER-2 positive breast cancer by pathology obtained through surgery or puncture, and confirmed by immunohistochemistry or FISH analysis; 2. Liver metastasis lesions were shown on CT images, and the puncture pathology or imaging report confirmed metastasis; 3. Anti-HER-2 drugs such as trastuzumab and pyrotinib were used regularly after liver metastasis was confirmed. Exclusion criteria: 1. Breast cancer liver metastasis was HER-2 negative; 2. CT images had motion artifacts, poor image quality, large differences in scanning conditions, and inconsistent layer thickness; 3. Cases where the progression-free survival period of anti-HER-2 drugs could not be clearly determined.

[0072] 2. CT Scanning Method

[0073] SOMATOM Definition AS 64-row spiral CT was used for enhanced abdominal CT scan, with a scanning range of 2 cm above the diaphragm to the lower edge of the liver, and scanning conditions: tube voltage 120 kV, automatic tube current, matrix 512×512, scanning layer thickness 5 mm, and layer spacing 5 mm. Iopromide injection was injected from the elbow vein with a flow rate of 3.0-3.5 mL / s and a dose of 1.0 mL / kg. Arterial phase, portal venous phase, and delayed phase scans were performed at 25-30s, 60-70s, and 120-180s after contrast agent injection, respectively. All patients collected were able to cooperate with the examination normally.

[0074] 3. Image sketching

[0075] All regions of interest (ROIs) of lesions on enhanced abdominal CT images were evaluated and outlined by two radiologists with 5 and 10 years of work experience, respectively, in a double-blind manner. After the first doctor finished outlining, the second senior physician reviewed it. If the difference was ≥5%, the latter decided the boundary and re-outlined it. Then grayscale normalization was performed to reduce the impact of contrast and brightness changes. Finally, 181 ROIs (61 ROIs in the arterial phase, 60 ROIs in the portal venous phase, and 60 ROIs in the delayed phase) were segmented from the CT images of 61 patients for subject analysis.

[0076] 4. Feature extraction and screening

[0077] A total of 1409 quantitative imaging features were extracted from CT images using the Radcloud platform (http: / / radcloud.cn / ). These features can be divided into three groups. The first group (first-order statistics) consists of 126 descriptors that quantitatively describe the distribution of voxel intensity in CT images through commonly used basic metrics. The second group (shape- and size-based features) contains 14 three-dimensional features that reflect regional shape and size. According to the gray level run length and gray level co-occurrence texture matrix, 525 texture features that can quantify regional heterogeneity differences are classified into the third category (texture features). Texture features characterize the recurring local patterns in the image and their arrangement rules, including 75 features such as gray level co-occurrence matrix (GLCM) features, gray level run length matrix (GLRLM) features, and gray level size zone matrix (GLSZM) features. In addition, 14 filters including exponential, logarithmic, gradient, square value, square root, lbp-2D and wavelet (wavelet-LHL, wavelet-LHH, wavelet-HLL, wavelet-LLH, wavelet-HLH, wavelet-HHH, wavelet-HHL, wavelet-LLL) are used to filter the images to analyze the texture at a finer scale.

[0078] For intra-observer and inter-observer delineation variation, the intra-observer and inter-observer agreement of each feature was quantified by calculating the intra-class correlation coefficient (ICC) between feature pairs. Features with low reproducibility were excluded from subsequent analysis, and any feature with an ICC less than 0.85 was discarded. Dimensionality reduction and feature selection were used to reduce redundant features and obtain the best results. The feature selection methods used variance thresholding (variance threshold = 0.8), SelectKBest, and LASSO models. For the variance thresholding method, the threshold was 0.8, and feature values ​​with variance less than 0.8 were removed. The SelectKBest method is a univariate feature selection method, and the p-value is used to analyze the relationship between the feature and the classification result. All features with a p-value less than 0.05 will be used. For the LASSO model, the L1 regularizer was used as the cost function, the error value of cross-validation was 5, and the maximum number of iterations was 1000.

[0079] 5. Construction of radiomics model

[0080] For the clinical data of the patients and the subsequent radiomics statistical analysis, the training data set and the test data set were randomly sampled in a ratio of 7:3 to establish the training group (n=42) and the test group (n=19), and the number of random seeds was 170. Six classifiers, including KNN, SVM, XGBoost, RF, LR and DT, were used to build a radiomics-based machine learning model to model the poor prognosis group and the good prognosis group.

[0081] 6. Evaluation Indicators

[0082] In order to evaluate the prediction performance, the receiver operating characteristic (ROC) curve, i.e., the area under the curve (AUC), was used in the training data set and the test data set respectively. The prediction efficiency of the model with 0.5<AUC≤0.7 is low, the prediction efficiency of the model with 0.7<AUC≤0.9 is medium, and the prediction efficiency of the model with AUC>0.9 is high. In addition, four indicators are used to evaluate the performance of the model, including precision (P) (referring to the proportion of all samples predicted to be correct), recall (R) (referring to the proportion of samples predicted to be correct), F1 score (F1-score) (based on the harmonic mean of precision and recall), and support (the total number of samples involved). The test set results are used as the evaluation method of machine learning by random grouping. The accuracy of the overall evaluation model classification is taken, and the average of the verification score numbers is taken each time to establish a precision (score) matrix, so as to screen out the appropriate ROI and select the optimal machine learning model.

[0083] 7. Statistical Analysis

[0084] SPSS software was used for statistical analysis of the data. Age was expressed as ± s, and the independent sample t test was used for comparison between groups; hormone receptor status, HER-2 status, physical status, previous chemotherapy drug use, and recurrence and metastasis status were compared using x 2 Test. The radiomics labels of each patient were formed in turn by the linear combination of the selected features and the corresponding weighted coefficients, and the risk score of each patient based on each radiomics label was calculated accordingly. The radiomics features of the poor prognosis group and the good prognosis group were statistically analyzed in the training group and the validation group, and the accuracy (score) matrix was established to compare and evaluate the training results of different radiomics models. The receiver operating characteristic (ROC) curve was used to evaluate the model performance. P < 0.05 indicated statistical significance.

[0085] result

[0086] 1. General Information

[0087] There was no statistical difference in age, physical condition, hormone receptor (HR) status, HER-2 status, and previous use of chemotherapy drugs between the poor prognosis group and the good prognosis group (P<0.05), as shown in Table 1.

[0088] Table 1 General status of subjects

[0089]

[0090] 2. Feature extraction and screening results

[0091] In this study, 351 features were selected from 1409 CT features tested by ICC using the variance threshold method ( Figure 2 ), and then use the SelectKBest method to filter out 12 features ( Figure 3 ), and finally the LASSO algorithm was used to select the 6 best features ( Figure 4 ). Based on these six features and their regression coefficients, the radiomics scoring (Rad-score) formula was constructed: Rad-score = feature * coefficient (Table 2).

[0092] Table 2 Radiomic features of ROIs screened in the venous phase and their categories, filters, and regression coefficients

[0093]

[0094] 3. Discrimination performance of radiomics models

[0095] The accuracy matrix scores of the six classifiers in the arterial phase, portal venous phase, and delayed phase are shown in Table 3. The ROC curve analysis results of all classifiers in the arterial phase ROI test set are shown in Table 4, and the ROC curve analysis results of all classifiers in the portal venous phase ROI test set are shown in Table 5. Among them, the RF classifier scored the highest. When trained with this classifier, the AUC value of the test set in the poor prognosis group was 0.889 (95% credible interval: 0.66-1.00, sensitivity 0.70, specificity 0.78), and the AUC value of the test set in the good prognosis group was 0.889 (95% credible interval: 0.66-1.00, sensitivity 0.78, specificity 0.70). The ROC curve analysis results of all classifiers in the delayed pulse phase ROI test set are shown in Table 6, and the four indicators (accuracy, recall, F1 score, support) of the portal venous phase ROI test set are shown in Table 7.

[0096] Table 3 Accuracy matrix scores of the six classifiers in the arterial phase, portal venous phase and delayed phase training and test sets

[0097]

[0098]

[0099] Table 4 ROC curve analysis results of arterial phase ROI test set classifier

[0100]

[0101] Table 5 ROC curve analysis results of portal venous phase ROI test set classifier

[0102]

[0103]

[0104] Table 6 ROC curve analysis results of the delayed period ROI test set classifier

[0105]

[0106] Table 7 Evaluation results of four indicators of portal venous phase ROI test set

[0107]

[0108]

[0109] Breast cancer is the most common malignant tumor in women. In recent years, it has become younger, seriously endangering women's health. Advanced breast cancer with cancer cell spread and metastasis greatly increases the risk of death. Among them, breast cancer metastases to the liver mainly through the hepatic artery. Relevant studies have confirmed that whether the liver is involved is an independent predictor of the overall survival (OS) of breast cancer patients. Once liver metastasis (LM) occurs, the patient usually has a poor prognosis. According to the NCCN guidelines and China's diagnosis and treatment routine, when the disease progresses, patients with advanced Her-2 positive breast cancer should continue anti-Her-2 treatment as one of the main treatment methods. Anti-HER-2 drugs such as trastuzumab, pyrotinib, and lapatinib have shown good therapeutic effects in clinical practice.

[0110] However, due to the temporal and spatial differences in tumor growth, pathological biopsy sometimes cannot fully reflect the heterogeneity of tumor tissue. The ER, PR and HER-2 status between primary breast cancer and liver metastasis, and even within metastasis, can be very different, making it difficult for some patients to benefit from anti-HER-2 therapy. How to screen out patients whose pathological results cannot truly reflect the body's HER-2 expression level is an urgent problem we need to solve.

[0111] In this study, the value of imaging genomics features based on abdominal enhanced CT in predicting the efficacy of anti-HER-2 treatment of BCLM was explored for the first time. The images retrospectively analyzed in this study were all abdominal enhanced CT images used to evaluate the efficacy of BCLM patients during treatment in our hospital, and there was no additional increase in the radiation dose and economic burden of the patients. Based on the conclusions of existing clinical trials, for HER-2-positive MBC, the mPFS of chemotherapy combined with the anti-HER-2 monoclonal antibody drug trastuzumab was 7.4 months. In addition, the results of a phase 3 clinical trial for HER-2-positive MBC patients who had received trastuzumab and taxane treatment showed that the mPFS of pyrotinib combined with capecitabine was 12.5 months, and the mPFS of lapatinib combined with capecitabine was 6.8 months. According to the 2021CSCO breast cancer diagnosis and treatment guidelines, trastuzumab is the first choice as an advanced anti-HER-2 drug for patients who have not used trastuzumab and have used trastuzumab but are eligible for reuse. A Phase II clinical study of pyrotuzumab included some patients who had not used trastuzumab before. Therefore, the expert group agreed that pyrotinib can also be used for patients who have failed trastuzumab treatment. For patients who have failed trastuzumab treatment, pyrotinib and lapatinib can both be used as optional chemotherapy options for advanced HER-2-positive breast cancer. Therefore, in order to make the research results more accurate and more valuable for reference, we selected the mPFS of trastuzumab combined with chemotherapy as the grouping criterion for this study, and divided the enrolled patients into two groups with different prognoses to compare the differences in their imaging characteristics.

[0112] In radiomics, there are a variety of machine learning methods that can be used to build classification models, and they have their own advantages for different tasks. In this study, six commonly used classifier models (KNN, SVM, XGBoost, RF, LR, and DT) were selected to compare the poor prognosis group and the good prognosis group. The optimal radiomics algorithm model was selected, and 181 ROIs of 61 HER-2 positive BCLM patients were extracted. The training group and the test group were established by stratified random sampling in a ratio of 7:3. The manually drawn ROIs were extracted and mined. The variance threshold, SelectKBest, and the least absolute shrinkage and selection operator (LASSO) logistic regression model were used for feature screening, and the differences between the two groups on CT images were analyzed. The AUC value of the SVM classifier in the arterial phase was the highest, which was 0.756, and the AUC value of the RF classifier in the delayed phase was 0.600. The classification efficiency of the model was average. When analyzing the ROI in the portal venous phase, the AUC value was significantly improved, and the RF classifier had the highest AUC value of 0.889. The misdiagnosis rate and missed diagnosis rate were reduced, which shows that the portal venous phase liver metastasis CT images are meaningful in predicting the efficacy of anti-HER-2 drugs.

[0113] The three phases of enhanced CT of liver metastases each have their own characteristics, mainly manifested as rapid and obvious ring enhancement in the arterial phase, continuous ring enhancement at the edge of the portal venous phase, and low density of lesions in the delayed phase. Studies have shown that the portal venous phase can detect more and more specific liver metastasis lesions and show the characteristics of metastases more clearly and accurately. Compared with primary liver cancer, the onset time and peak time of enhancement of liver metastases are prolonged, which is consistent with the conclusion that the portal venous phase is clearer. HER-2 receptor-mediated signals can promote the secretion of vascular endothelial growth factor (VEGF), cause an increase in tumor-related angiogenesis, and promote tumor growth and development. Therefore, we believe that liver metastases in patients with stronger HER-2 expression have richer blood supply, clearer and more persistent enhancement in the portal venous phase, and can better show their imaging characteristics. Other studies have shown that HER-2 expression in gastric cancer is related to the portal venous phase of enhanced CT rather than other phases. Therefore, we believe that the imaging characteristics of the portal venous phase of enhanced CT can well reflect the HER-2 expression level of liver metastases and predict the effectiveness of anti-HER-2 treatment in patients. Radiomics models can become a new clinical tool, providing clinicians with additional information and promoting the development of personalized precision treatment.

[0114] In summary, the radiomics model based on texture features is more accurate and objective. Radiomics can be an objective, precise, simple and easy non-invasive tool to judge the efficacy of anti-HER-2 treatment in patients with breast cancer liver metastasis. Among them, the RF algorithm model of enhanced CT portal venous phase has the best classification efficiency.

[0115] Example 2

[0116] A therapeutic effect prediction system for anti-HER-2 treatment of breast cancer liver metastasis, the therapeutic effect prediction system may include:

[0117] A data acquisition module, used for acquiring an abdominal enhanced CT image to be identified of a subject, i.e., a patient with liver metastasis of breast cancer treated with anti-HER-2 therapy;

[0118] Data processing and output module: used to input the abdominal enhanced CT image to be identified into the efficacy prediction model, the efficacy prediction model performs quantitative analysis operations on the abdominal enhanced CT image to be identified, and outputs the efficacy prediction results of anti-HER-2 treatment of breast cancer liver metastasis of the subject.

[0119] Among them, in the data processing and output module, the efficacy prediction model is obtained by using a machine learning algorithm to train the model using the pre-collected imaging omics-related feature data of patients with liver metastasis of breast cancer treated with anti-HER-2.

[0120] The specific methods for constructing the efficacy prediction model include:

[0121] S1. Evaluate and outline the lesion area on the patient's abdominal enhanced CT image to obtain the region of interest (ROI);

[0122] S2. Extract the radiomics features in the ROI and build a radiomics model; perform data training and testing on the radiomics model, and use the trained model as a therapeutic effect prediction model;

[0123] Wherein, in the step S1,

[0124] The evaluation and delineation can be performed manually, for example, by at least two senior radiologists using a double-blind method, and the second senior physician will review the delineation after the first physician has finished the delineation, thereby ensuring the accuracy of the delineation result; of course, the evaluation and delineation can also be performed using a related delineation device, thereby identifying and delineating the lesion area of ​​the image, and outputting the delineated image;

[0125] The step S1 also includes further processing the obtained ROI, such as grayscale normalization, to reduce the impact of contrast and brightness changes. In a specific embodiment of the present invention, the obtained ROI includes an arterial phase ROI, a portal venous phase ROI, and a delayed phase ROI;

[0126] In step S2, extracting radiomics features in ROI includes quantifying ROI to obtain ROI features and screening to obtain effective radiomics features;

[0127] The ROI is quantified, including extracting quantitative imaging features from the CT image using the Radcloud platform; grouping the quantitative imaging features, and the grouping method can be descriptors, shape and size features, and texture features; then the ROI is quantified using an intra-class correlation coefficient (ICC), and the quantified ROI can be screened using a variance threshold, SelectKBest or LASSO model to extract effective imaging omics features;

[0128] Among them, the descriptor quantitatively describes the distribution of voxel intensity in CT images through commonly used basic metrics; the shape and size-based features can be three-dimensional features that reflect the shape and size of the region; the texture features are obtained by quantifying the regional heterogeneity differences based on the gray level run and gray level co-occurrence texture matrix, and the texture features describe the repeated local patterns in the image and their arrangement rules, including gray level co-occurrence matrix (GLCM) features, gray level run length matrix (GLRLM) features, and gray level size zone matrix (GLSZM) features, etc. 75 features. Preferably, 14 filters such as exponential, logarithmic, gradient, square value, square root, lbp-2D and wavelet (wavelet-LHL, wavelet-LHH, wavelet-HLL, wavelet-LLH, wavelet-HLH, wavelet-HHH, wavelet-HHL, wavelet-LLL) can be used to filter the image and analyze the texture at a finer scale.

[0129] The radiomics model can be obtained by using a classifier model, and the classifier model includes but is not limited to KNN, SVM, XGBoost, RF, LR and DT6;

[0130] Specifically, the abdominal enhanced CT images of patients with ROI outlined are randomly divided into a training group and a test group, so as to train and test the imaging omics model, and the trained model is used as a therapeutic effect prediction model. At the same time, the receiver operating characteristic (ROC) curve, that is, the area under the curve (AUC), can be used to evaluate the prediction performance. Specifically, the prediction efficiency of the model with 0.5<AUC≤0.7 is low, the prediction efficiency of the model with 0.7<AUC≤0.9 is medium, and the prediction efficiency of the model with AUC>0.9 is high. Further preferably, four indicators can be used to evaluate the performance of the model, including precision (P) (referring to the proportion of all samples predicted to be correct), recall (R) (referring to the proportion of samples predicted to be correct), F1 score (F1-score) (based on the harmonic mean of precision and recall), and support (the total number of samples involved).

[0131] The test set results are used as the evaluation method for machine learning by random grouping. The accuracy of the overall evaluation model classification is taken, and the average of each verification score is taken to establish an accuracy (score) matrix to screen out the appropriate ROI and select the optimal machine learning model.

[0132] Example 3

[0133] An electronic device includes a memory and a processor, and computer instructions stored in the memory and executed on the processor. When the computer instructions are executed by the processor, each operation in the method of Example 1 is completed. For the sake of brevity, they are not described here in detail.

[0134] The electronic device can be a mobile terminal or a non-mobile terminal. The non-mobile terminal includes a desktop computer, and the mobile terminal includes a smart phone (such as an Android phone, an IOS phone, etc.), smart glasses, a smart watch, a smart bracelet, a tablet computer, a laptop computer, a personal digital assistant, and other mobile Internet devices that can perform wireless communication.

[0135] It should be understood that in the present invention, the processor may be a central processing unit CPU, and the processor may also be other general-purpose processors, digital signal processors DSP, application-specific integrated circuits ASIC, off-the-shelf programmable gate arrays FPGA or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0136] The memory may include a read-only memory and a random access memory, and provide instructions and data to the processor, and a portion of the memory may also include a non-volatile random access memory. For example, the memory may also store information about the device type.

[0137] In the implementation process, each step of the above method can be completed by an integrated logic circuit of hardware in a processor or an instruction in the form of software. The steps of the method disclosed in the present invention can be directly embodied as being executed by a hardware processor, or being executed by a combination of hardware and software modules in a processor. The software module can be located in a mature storage medium in the art such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory, or an electrically erasable programmable memory, a register, etc. The storage medium is located in a memory, and the processor reads the information in the memory and completes the steps of the above method in combination with its hardware. To avoid repetition, it is not described in detail here. A person of ordinary skill in the art can appreciate that the units of each example described in combination with the embodiments disclosed herein, i.e., the algorithm steps, can be implemented in electronic hardware or a combination of computer software and electronic hardware. Whether these functions are executed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of this application.

[0138] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0139] In the several embodiments provided in the present application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units is only a division of logical functions. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, and the indirect coupling or communication connection of the device or unit can be electrical, mechanical or other forms.

[0140] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application can be essentially or partly embodied in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server or a network device, etc.) to perform all or part of the steps of the methods described in each embodiment of the present application. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0141] It should be noted that the above examples are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention is described in detail with reference to the given examples, those skilled in the art may modify or replace the technical solution of the present invention as needed without departing from the spirit and scope of the technical solution of the present invention.

Claims

1. A method for predicting the efficacy of anti-HER-2 treatment for breast cancer liver metastasis, characterized in that: The method comprises: The median progression-free survival of trastuzumab combined with chemotherapy was selected as the grouping criterion, and the enrolled patients were divided into two groups with different prognoses to compare the differences in their imaging characteristics; Obtaining an abdominal enhanced CT image to be identified from a subject, i.e., a patient with liver metastasis of breast cancer treated with anti-HER-2 therapy; Inputting the to-be-identified abdominal enhanced CT image into the efficacy prediction model, wherein the efficacy prediction model performs a quantitative analysis operation on the to-be-identified abdominal enhanced CT image to obtain a prediction result of the efficacy of anti-HER-2 treatment for breast cancer liver metastasis of the subject; The efficacy prediction model is obtained by training the model using a machine learning algorithm using the pre-collected imaging omics-related feature data of patients with liver metastasis of breast cancer treated with anti-HER-2 therapy; The specific method for constructing the efficacy prediction model includes: S1. Evaluate and outline the lesion area on the patient's abdominal enhanced CT image to obtain the region of interest (ROI), which includes the portal venous phase ROI; S2. Extract the radiomics features in the ROI and build a radiomics model; perform data training and testing on the radiomics model, and use the trained model as a therapeutic effect prediction model; In step S2, extracting radiomics features in ROI includes quantifying ROI to obtain ROI features and screening to obtain effective radiomics features, specifically: 351 features were selected from 1409 CT features tested by ICC by using the variance threshold method, and then 12 features were selected by using the SelectKBest method. Finally, 6 optimal features were selected by using the LASSO algorithm. The radiomics scoring formula was constructed based on these 6 features and their regression coefficients: Rad-score = feature * coefficient. The test set results are used as the evaluation method of machine learning by random grouping. The accuracy of the overall evaluation model classification is taken, and the average of each verification score is taken to establish an accuracy matrix, so as to screen out the appropriate ROI and select the optimal machine learning model. The radiomics model is obtained by using a classifier model, and the classifier model includes RF.

2. The therapeutic effect prediction method according to claim 1, characterized in that: The step S1 also includes further processing the acquired ROI, including grayscale normalization.

3. A system for predicting the efficacy of anti-HER-2 treatment for liver metastasis of breast cancer, characterized in that: The therapeutic effect prediction system comprises: The data acquisition module selects the mPFS of trastuzumab combined with chemotherapy as the grouping standard, divides the enrolled patients into two groups with different prognoses to compare the differences in their imaging characteristics; it is used to obtain the abdominal enhanced CT images to be identified in the subjects, i.e., patients with liver metastases of breast cancer treated with anti-HER-2 therapy; Data processing and output module: used for inputting the to-be-identified abdominal enhanced CT image into the efficacy prediction model, wherein the efficacy prediction model performs quantitative analysis on the to-be-identified abdominal enhanced CT image and outputs the efficacy prediction result of the anti-HER-2 treatment of breast cancer liver metastasis of the subject; Among them, in the data processing and output module, the efficacy prediction model is obtained by training the model using a machine learning algorithm using the pre-collected imaging omics-related feature data of patients with liver metastasis of breast cancer treated with anti-HER-2 therapy; The specific method for constructing the efficacy prediction model includes: S1. Evaluate and outline the lesion area on the patient's abdominal enhanced CT image to obtain the region of interest (ROI), which includes the portal venous phase ROI; S2. Extract the radiomics features in the ROI and build a radiomics model; perform data training and testing on the radiomics model, and use the trained model as a therapeutic effect prediction model; In step S2, extracting radiomics features in ROI includes quantifying ROI to obtain ROI features and screening to obtain effective radiomics features, specifically: 351 features were selected from 1409 CT features tested by ICC by using the variance threshold method, and then 12 features were selected by using the SelectKBest method. Finally, 6 optimal features were selected by using the LASSO algorithm. The radiomics scoring formula was constructed based on these 6 features and their regression coefficients: Rad-score = feature * coefficient. The test set results are used as the evaluation method of machine learning by random grouping. The accuracy of the overall evaluation model classification is taken, and the average of each verification score is taken to establish an accuracy matrix, so as to screen out the appropriate ROI and select the optimal machine learning model. The radiomics model is obtained by using a classifier model, and the classifier model includes RF.

4. The system according to claim 3, characterized in that In step S1, the evaluation and delineation is performed manually or by a delineation device.

5. The system according to claim 4, characterized in that The step S1 also includes further processing the acquired ROI, including grayscale normalization.

6. The system according to claim 3, characterized in that quantifying the ROI, including extracting quantitative imaging features from the CT image using a Radcloud platform; grouping the quantitative imaging features, wherein the grouping method includes descriptors, shape and size-based features, and texture-based features; Then, the intraclass correlation coefficient was used to quantify the ROIs, and the variance threshold, SelectKBest or LASSO model were used to screen the quantified ROIs to extract effective radiomics features.

7. An electronic device, characterized in that: The method comprises a memory and a processor and computer instructions stored in the memory and executed on the processor. When the computer instructions are executed by the processor, the steps of the method for predicting the efficacy of anti-HER-2 treatment of liver metastasis of breast cancer according to any one of claims 1 to 2 are completed.

8. A computer-readable storage medium, characterized in that: Used to store computer instructions, which, when executed by a processor, complete the steps of the method for predicting the efficacy of anti-HER-2 treatment of liver metastasis of breast cancer according to any one of claims 1 to 2.

Images