Esophageal squamous carcinoma immunotherapy prediction method, device, equipment and medium

Abstract

The invention discloses an esophageal squamous carcinoma immunotherapy prediction method and device, equipment and a medium. The method comprises the following steps: acquiring CT image data of a case; the CT image data is image data before neoadjuvant immunotherapy combined with radiotherapy and chemotherapy; the case is an esophageal squamous cell carcinoma patient to receive neoadjuvant immunotherapy combined radiotherapy and chemotherapy; obtaining a region of interest after the CT image data is sketched; extracting radiomics characteristics of the region of interest; inputting the radiomics characteristics into the target esophageal squamous cell carcinoma immunotherapy prediction model to obtain a prediction result; the prediction result is used for indicating the prognosis state of the case in receiving the neoadjuvant immunotherapy combined radiotherapy and chemotherapy. Therefore, in combination with a non-invasive imaging analysis means of imaging omics and through the pre-constructed target esophageal squamous cell carcinoma immunotherapy prediction model, whether the patient obtains complete pathological relief through combination of new auxiliary immunotherapy and chemoradiotherapy is predicted, so that immunotherapy is guided, and the survival rate of the esophageal squamous cell carcinoma patient is increased.

Description

Technical Field

[0001] The present application relates to the field of machine learning technology, and in particular to a method, device, equipment and medium for predicting immunotherapy for esophageal squamous cell carcinoma. Background Art

[0002] Esophageal squamous cell carcinoma is the most common pathological type of esophageal cancer. It is a highly malignant tumor. Patients are often in the late stage of the disease at the time of initial diagnosis, and the surgical effect is poor, so the survival period is usually short. Neoadjuvant immunotherapy combined with chemoradiotherapy is a new treatment method that is expected to significantly improve the survival prognosis of patients with esophageal squamous cell carcinoma.

[0003] However, current methods for evaluating the expression of local infiltrating lymphocytes and immune checkpoints in tumor tissue all rely on invasive tissue sample acquisition, and then performing relevant analysis based on the tissue samples to obtain the patient's prognostic status after neoadjuvant immunotherapy combined with chemoradiotherapy. In other words, based on current existing technologies, it is impossible to quickly obtain the patient's prognostic status after neoadjuvant immunotherapy combined with chemoradiotherapy. Summary of the invention

[0004] The present application provides a method, device, equipment and medium for predicting immunotherapy of esophageal squamous cell carcinoma, which can quickly predict the prognosis of esophageal squamous cell carcinoma patients after neoadjuvant immunotherapy combined with chemoradiotherapy through non-invasive imaging genomics, thereby improving the survival rate of esophageal squamous cell carcinoma patients to a certain extent.

[0005] In a first aspect, the present application provides a method for predicting immunotherapy for esophageal squamous cell carcinoma, the method comprising:

[0006] Obtaining computed tomography (CT) image data of the case; the CT image data is image data before neoadjuvant immunotherapy combined with chemoradiotherapy; the case is a patient with esophageal squamous cell carcinoma who is to receive neoadjuvant immunotherapy combined with chemoradiotherapy;

[0007] Obtaining a region of interest delineated from the CT image data;

[0008] Extracting radiomic features of the region of interest;

[0009] The imaging genomics features are input into a target esophageal squamous cell carcinoma immunotherapy prediction model to obtain a prediction result; the target esophageal squamous cell carcinoma immunotherapy prediction model is generated by training an extreme gradient boosting algorithm XGBoost model; the prediction result is used to indicate the prognostic status of the case when receiving neoadjuvant immunotherapy combined with chemoradiotherapy.

[0010] Optionally, the method further comprises:

[0011] Acquire sample CT image data corresponding to a plurality of case samples; the case samples are patients with esophageal squamous cell carcinoma; the sample CT image data are image data of the case samples before receiving immunotherapy combined with chemoradiotherapy;

[0012] Obtaining a sample region of interest after outlining the sample CT image data;

[0013] Extracting radiomics features of the sample region of interest to obtain an initial radiomics feature set;

[0014] Performing feature selection and dimensionality reduction on the initial radiomics feature set to obtain a target radiomics feature set;

[0015] Dividing the target radiomics feature set to obtain a model training set and a model test set;

[0016] The extreme gradient boosting algorithm XGBoost model is trained using the model training set to generate an initial esophageal squamous cell carcinoma immunotherapy prediction model, wherein the XGBoost model is a model after hyperparameter optimization and obtaining the optimal hyperparameter combination;

[0017] Performing a model performance evaluation on the initial esophageal squamous cell carcinoma immunotherapy prediction model based on the model test set to obtain an evaluation result;

[0018] When the performance evaluation result of the initial esophageal squamous cell carcinoma immunotherapy prediction model meets the preset requirements, the initial esophageal squamous cell carcinoma immunotherapy prediction model is obtained as the target esophageal squamous cell carcinoma immunotherapy prediction model.

[0019] Optionally, the using the model training set to train the extreme gradient boosting algorithm XGBoost model to generate an initial esophageal squamous cell carcinoma immunotherapy prediction model includes:

[0020] Defining a hyperparameter search space of the XGBoost model, wherein the hyperparameter search space stores hyperparameters to be optimized;

[0021] Using a Bayesian optimization algorithm to search the hyperparameter search space to obtain multiple hyperparameter combinations;

[0022] Use cross-validation to evaluate the performance indicators of multiple hyperparameter combinations, and obtain the hyperparameter combination with the best performance indicator as the optimal hyperparameter combination;

[0023] The XGBoost model is trained based on the model training set to generate the initial esophageal squamous cell carcinoma immunotherapy prediction model, and the XGBoost model sets the optimal hyperparameter combination.

[0024] Optionally, performing feature selection and dimensionality reduction on the initial radiomics feature set to obtain a target radiomics feature set includes:

[0025] Based on at least one of a Spearman correlation analysis method and a Mann-Whitney U test method, feature selection is performed on the initial radiomics feature set to obtain an intermediate radiomics feature set;

[0026] The intermediate radiomics feature set is eliminated by using a support vector machine-recursive feature elimination method to obtain the target radiomics feature set.

[0027] Optionally, the performing a model performance evaluation on the initial esophageal squamous cell carcinoma immunotherapy prediction model based on the model test set to obtain an evaluation result includes:

[0028] Inputting the model test set into the initial esophageal squamous cell carcinoma immunotherapy prediction model to obtain a prediction probability;

[0029] According to the predicted probability and the actual probability in the model test set, the model performance of the initial esophageal squamous cell carcinoma immunotherapy prediction model is evaluated to obtain the evaluation result.

[0030] Optionally, performing a model performance evaluation on the initial esophageal squamous cell carcinoma immunotherapy prediction model according to the predicted probability and the actual probability in the model test set to obtain the evaluation result includes:

[0031] According to the predicted probability and the actual probability in the model test set, a goodness-of-fit HL test is performed on the initial esophageal squamous cell carcinoma immunotherapy prediction model to generate a calibration curve, wherein the calibration curve is generated based on the HL test result;

[0032] Based on the calibration curve, a calibration evaluation result of the initial esophageal squamous cell carcinoma immunotherapy prediction model is obtained.

[0033] Optionally, performing a model performance evaluation on the initial esophageal squamous cell carcinoma immunotherapy prediction model according to the predicted probability and the actual probability in the model test set to obtain the evaluation result includes:

[0034] Generating a receiver operating characteristic (ROC) curve of the initial esophageal squamous cell carcinoma immunotherapy prediction model according to the predicted probability and the actual probability in the model test set;

[0035] Based on the area under the curve (AUC) value of the ROC curve, the accuracy evaluation result of the initial esophageal squamous cell carcinoma immunotherapy prediction model is obtained.

[0036] Optionally, performing a model performance evaluation on the initial esophageal squamous cell carcinoma immunotherapy prediction model according to the predicted probability and the actual probability in the model test set to obtain the evaluation result includes:

[0037] Generating a decision curve of the initial esophageal squamous cell carcinoma immunotherapy prediction model according to the predicted probability and the actual probability in the model test set;

[0038] A decision curve analysis is performed on the decision curve to obtain a net benefit evaluation result, and the net benefit evaluation result represents the application value of the initial esophageal squamous cell carcinoma immunotherapy prediction model in actual clinical decision-making.

[0039] Optionally, dividing the target radiomics feature set to obtain a model training set and a model test set includes:

[0040] The target radiomics feature set is divided using a random seed to obtain the model training set and the model test set.

[0041] Optionally, the imaging omics features include shape features, first-order statistical features, gray-level co-occurrence matrix features, gray-level dependence matrix features, gray-level run length matrix features, gray-level size area matrix features, Laplace features or wavelet features.

[0042] In a second aspect, the present application also provides a device for predicting immunotherapy of esophageal squamous cell carcinoma, the device comprising:

[0043] An acquisition unit is used to acquire computer tomography CT image data of a case; the CT image data is image data before neoadjuvant immunotherapy combined with chemoradiotherapy; the case is a patient with esophageal squamous cell carcinoma who is to receive neoadjuvant immunotherapy combined with chemoradiotherapy;

[0044] An obtaining unit, used for obtaining a region of interest after delineation of the CT image data;

[0045] An extraction unit, used for extracting radiomics features of the region of interest;

[0046] An input unit is used to input the imaging genomics features into a target esophageal squamous cell carcinoma immunotherapy prediction model to obtain a prediction result; the target esophageal squamous cell carcinoma immunotherapy prediction model is generated by training the extreme gradient boosting algorithm XGBoost model; the prediction result is used to indicate the prognostic status of the case when receiving neoadjuvant immunotherapy combined with chemoradiotherapy.

[0047] Optionally, the device further comprises a model building unit, wherein the model building unit comprises:

[0048] The sample acquisition subunit is used to acquire sample CT image data corresponding to a plurality of case samples; the case samples are patients with esophageal squamous cell carcinoma; the sample CT image data are image data of the case samples before receiving immunotherapy combined with radiotherapy and chemotherapy;

[0049] A sample region of interest obtaining subunit, used to obtain the sample region of interest after delineating the sample CT image data;

[0050] A feature extraction subunit, used to extract the radiomics features of the region of interest of the sample to obtain an initial radiomics feature set;

[0051] A preprocessing subunit, used for performing feature selection and dimensionality reduction on the initial radiomics feature set to obtain a target radiomics feature set;

[0052] A division subunit, used to divide the target radiomics feature set to obtain a model training set and a model test set;

[0053] A model training subunit is used to train the extreme gradient boosting algorithm XGBoost model using the model training set to generate an initial esophageal squamous cell carcinoma immunotherapy prediction model, wherein the XGBoost model is a model after hyperparameter optimization and obtaining the optimal hyperparameter combination;

[0054] A model evaluation subunit is used to perform a model performance evaluation on the initial esophageal squamous cell carcinoma immunotherapy prediction model based on the model test set to obtain an evaluation result;

[0055] The model acquisition subunit is used to obtain the initial esophageal squamous cell carcinoma immunotherapy prediction model as the target esophageal squamous cell carcinoma immunotherapy prediction model when the performance evaluation result of the initial esophageal squamous cell carcinoma immunotherapy prediction model meets the preset requirements.

[0056] Optionally, the model training subunit is specifically used to:

[0057] Defining a hyperparameter search space of the XGBoost model, wherein the hyperparameter search space stores hyperparameters to be optimized;

[0058] Using a Bayesian optimization algorithm to search the hyperparameter search space to obtain multiple hyperparameter combinations;

[0059] Use cross-validation to evaluate the performance indicators of multiple hyperparameter combinations, and obtain the hyperparameter combination with the best performance indicator as the optimal hyperparameter combination;

[0060] The XGBoost model is trained based on the model training set to generate the initial esophageal squamous cell carcinoma immunotherapy prediction model, and the XGBoost model sets the optimal hyperparameter combination.

[0061] Optionally, the preprocessing subunit is specifically used to:

[0062] Based on at least one of a Spearman correlation analysis method and a Mann-Whitney U test method, feature selection is performed on the initial radiomics feature set to obtain an intermediate radiomics feature set;

[0063] The intermediate radiomics feature set is eliminated by using a support vector machine-recursive feature elimination method to obtain the target radiomics feature set.

[0064] Optionally, the model evaluation subunit is specifically used to:

[0065] Inputting the model test set into the initial esophageal squamous cell carcinoma immunotherapy prediction model to obtain a prediction probability;

[0066] According to the predicted probability and the actual probability in the model test set, the model performance of the initial esophageal squamous cell carcinoma immunotherapy prediction model is evaluated to obtain the evaluation result.

[0067] Optionally, performing a model performance evaluation on the initial esophageal squamous cell carcinoma immunotherapy prediction model according to the predicted probability and the actual probability in the model test set to obtain the evaluation result includes:

[0068] According to the predicted probability and the actual probability in the model test set, a goodness-of-fit HL test is performed on the initial esophageal squamous cell carcinoma immunotherapy prediction model to generate a calibration curve, wherein the calibration curve is generated based on the HL test result;

[0069] Based on the calibration curve, a calibration evaluation result of the initial esophageal squamous cell carcinoma immunotherapy prediction model is obtained.

[0070] Optionally, performing a model performance evaluation on the initial esophageal squamous cell carcinoma immunotherapy prediction model according to the predicted probability and the actual probability in the model test set to obtain the evaluation result includes:

[0071] Generating a receiver operating characteristic (ROC) curve of the initial esophageal squamous cell carcinoma immunotherapy prediction model according to the predicted probability and the actual probability in the model test set;

[0072] Based on the area under the curve (AUC) value of the ROC curve, the accuracy evaluation result of the initial esophageal squamous cell carcinoma immunotherapy prediction model is obtained.

[0073] Optionally, performing a model performance evaluation on the initial esophageal squamous cell carcinoma immunotherapy prediction model according to the predicted probability and the actual probability in the model test set to obtain the evaluation result includes:

[0074] Generating a decision curve of the initial esophageal squamous cell carcinoma immunotherapy prediction model according to the predicted probability and the actual probability in the model test set;

[0075] A decision curve analysis is performed on the decision curve to obtain a net benefit evaluation result, and the net benefit evaluation result represents the application value of the initial esophageal squamous cell carcinoma immunotherapy prediction model in actual clinical decision-making.

[0076] Optionally, the imaging omics features include shape features, first-order statistical features, gray-level co-occurrence matrix features, gray-level dependence matrix features, gray-level run length matrix features, gray-level size area matrix features, Laplace features or wavelet features.

[0077] Optionally, the division subunit is specifically used to:

[0078] The target radiomics feature set is divided using a random seed to obtain the model training set and the model test set.

[0079] In a third aspect, the present application further provides an electronic device, the electronic device comprising a processor and a memory:

[0080] The memory is used to store computer programs;

[0081] The processor is used to execute the method provided in the first aspect according to the computer program.

[0082] In a fourth aspect, the present application further provides a computer-readable storage medium, wherein the computer-readable storage medium is used to store a computer program, and the computer program is used to execute the method provided in the first aspect above.

[0083] It can be seen that this application has the following beneficial effects:

[0084] The present application provides a method, device, equipment and medium for predicting immunotherapy for esophageal squamous cell carcinoma. In the method, CT image data of the case is obtained; the CT image data is the image data before neoadjuvant immunotherapy combined with chemoradiotherapy; the case is an esophageal squamous cell carcinoma patient to be treated with neoadjuvant immunotherapy combined with chemoradiotherapy; the region of interest after delineation of the CT image data is obtained; the imaging genomics features of the region of interest are extracted; the imaging genomics features are input into the target esophageal squamous cell carcinoma immunotherapy prediction model to obtain the prediction result; the target esophageal squamous cell carcinoma immunotherapy prediction model is generated by training the extreme gradient boosting algorithm XGBoost model; the prediction result is used to indicate the prognosis status of the case in receiving neoadjuvant immunotherapy combined with chemoradiotherapy. In this way, combined with the non-invasive imaging analysis method of imaging genomics, and through the pre-constructed target esophageal squamous cell carcinoma immunotherapy prediction model, it is predicted whether the patient has achieved complete pathological remission through neoadjuvant immunotherapy combined with chemoradiotherapy, thereby guiding immunotherapy. Compared with the invasive operation of obtaining tissue samples in the prior art, it can quickly predict the prognosis status of esophageal squamous cell carcinoma patients after neoadjuvant immunotherapy combined with chemoradiotherapy, and improve the survival rate of esophageal squamous cell carcinoma patients. BRIEF DESCRIPTION OF THE DRAWINGS

[0085] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present application. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0086] Figure 1 This is a flow chart of a method for predicting immunotherapy of esophageal squamous cell carcinoma in an embodiment of the present application;

[0087] Figure 2 This is a schematic diagram of a process for constructing a prediction model for immunotherapy of esophageal squamous cell carcinoma in an embodiment of the present application;

[0088] Figure 3 It is the ROC curve graph and box plot of the model training set and the model test set in the embodiment of the present application;

[0089] Figure 4 It is a calibration curve diagram of the model training set and the model test set in the embodiment of the present application;

[0090] Figure 5 It is a decision curve analysis diagram of the model training set and the model test set in the embodiment of the present application;

[0091] Figure 6 A schematic diagram of the structure of an esophageal squamous cell carcinoma immunotherapy prediction device 600 provided in an embodiment of the present application;

[0092] Figure 7A schematic diagram of the structure of an electronic device 700 provided in an embodiment of the present application. DETAILED DESCRIPTION

[0093] The multiple involved in the embodiments of the present application means greater than or equal to two. It should be noted that in the description of the embodiments of the present application, the words "first", "second", etc. are only used for the purpose of distinguishing the description, and cannot be understood as indicating or implying relative importance, nor can they be understood as indicating or implying an order.

[0094] In order to make the above-mentioned purposes, features and advantages of the present application more obvious and easy to understand, the embodiments of the present application are further described in detail below in conjunction with the accompanying drawings and specific implementation methods. It is understood that the specific embodiments described herein are only used to explain the present application and are not intended to limit the present application. In addition, it should be noted that, for ease of description, only the parts related to the present application are shown in the accompanying drawings, not all structures.

[0095] First, some of the terms used in the embodiments of the present application are explained:

[0096] 1. Spearman correlation analysis is used to measure the nonlinear relationship between two variables. In the feature selection process, the Spearman correlation coefficient between the features is calculated. When the correlation coefficient is greater than 0.95, one of the variables is randomly deleted and the other variable is saved. Spearman correlation analysis helps identify the correlation between variables and helps avoid selecting highly redundant features.

[0097] 2. Mann-Whitney U test, also known as Mann-Whitney rank sum test, is a nonparametric test method used to compare the medians of two independent samples. In the feature selection process, Mann-Whitney U test is used to compare the feature distribution between different groups (such as pathological complete remission pCR group and non-pCR group) and select features with significant differences.

[0098] 3. Support vector machine-recursive feature elimination (SVM-RFE) is a backward selection method. SVM-RFE ranks all features according to their importance, and then recursively removes the least important features according to the performance of the model, thereby selecting key features to build the final model. It should be noted that by selecting key features, the dimension of the data set (i.e., the number of features) can actually be reduced, thereby indirectly achieving the effect of dimensionality reduction. In other words, SVM-RFE can indirectly achieve the effect of dimensionality reduction by selecting key features.

[0099] 4. eXtreme Gradient Boosting (XGBoost) is an ensemble learning algorithm based on the Gradient Boosting Decision Tree (GBDT), which is dedicated to breaking through the computational limits of the boosting tree to achieve the engineering goal of fast operation and excellent performance. When using the training set for model training, XGBoost will gradually optimize the classifier according to the definition of the loss function, generate multiple decision trees, and combine the above decision trees together to form the final strong learner to achieve prediction. Among them, XGBoost continuously fits the residual of the model in an iterative manner to gradually improve the prediction ability. Compared with the traditional gradient boosting algorithm, XGBoost has made many improvements and has been considered to be an advanced evaluator with super high performance in both classification and regression.

[0100] 5. Volume of Interest (VOI) is a commonly used concept in three-dimensional image processing, which is used to extract specific areas in the image for further analysis and processing.

[0101] 6. Bayesian optimization is a hyperparameter tuning method based on Bayesian theorem. Its basic idea is to build a posterior probability distribution model about the objective function (model performance) and optimize the objective function in the process of continuously observing data. Bayesian optimization continuously selects the next optimal hyperparameter in the hyperparameter space for optimization, thereby finding the optimal hyperparameter combination.

[0102] The applicant found through research that the current methods for evaluating the expression of local infiltrating lymphocytes and immune checkpoints in tumor tissues all rely on invasive tissue sample acquisition, and then conduct relevant analysis based on the tissue samples to obtain the patient's prognosis after neoadjuvant immunotherapy combined with chemoradiotherapy. In other words, based on the current existing technology, it is impossible to quickly obtain the patient's prognosis after neoadjuvant immunotherapy combined with chemoradiotherapy. In addition, in the field of locally advanced esophageal squamous cell carcinoma, the prediction of efficacy by imaging omics is mainly focused on neoadjuvant chemotherapy or chemoradiotherapy, but there is a lack of research on predicting neoadjuvant immunotherapy combined with chemoradiotherapy.

[0103] Based on this, the present application provides a method, device, equipment and medium for predicting immunotherapy for esophageal squamous cell carcinoma. In the method, CT image data of the case is obtained; the CT image data is the image data before neoadjuvant immunotherapy combined with chemoradiotherapy; the case is an esophageal squamous cell carcinoma patient to be treated with neoadjuvant immunotherapy combined with chemoradiotherapy; the region of interest after delineation of the CT image data is obtained; the imaging genomics features of the region of interest are extracted; the imaging genomics features are input into the target esophageal squamous cell carcinoma immunotherapy prediction model to obtain the prediction result; the target esophageal squamous cell carcinoma immunotherapy prediction model is generated by training the extreme gradient boosting algorithm XGBoost model; the prediction result is used to indicate the prognostic status of the case when receiving neoadjuvant immunotherapy combined with chemoradiotherapy.

[0104] In this way, by combining radiomics, a non-invasive imaging analysis method, and through the pre-constructed target esophageal squamous cell carcinoma immunotherapy prediction model, it is predicted whether the patient will achieve complete pathological remission (pCR) through neoadjuvant immunotherapy combined with chemoradiotherapy, thereby guiding immunotherapy. Compared with the invasive operation of obtaining tissue samples in the existing technology, it can quickly predict the prognosis of esophageal squamous cell carcinoma patients after neoadjuvant immunotherapy combined with chemoradiotherapy, thereby improving the survival rate of esophageal squamous cell carcinoma patients.

[0105] To facilitate understanding of the specific implementation of the method for predicting immunotherapy of esophageal squamous cell carcinoma provided in the embodiment of the present application, it will be described below with reference to the accompanying drawings.

[0106] It should be noted that the subject implementing the esophageal squamous cell carcinoma immunotherapy prediction method can be the esophageal squamous cell carcinoma immunotherapy prediction device provided in the embodiment of the present application, and the esophageal squamous cell carcinoma immunotherapy prediction device can be carried in an electronic device or a functional module of an electronic device. The electronic device in the embodiment of the present application can be any device capable of implementing the esophageal squamous cell carcinoma immunotherapy prediction method in the embodiment of the present application, for example, it can be an Internet of Things (IoT) device.

[0107] See also Figure 1 , provides a method for predicting immunotherapy for esophageal squamous cell carcinoma, in an embodiment of the present application, for example, the following steps may be included:

[0108] S1, obtain the computed tomography (CT) image data of the case.

[0109] Among them, the CT image data is the image data before neoadjuvant immunotherapy combined with chemoradiotherapy; the case is an esophageal squamous cell carcinoma patient who is about to receive neoadjuvant immunotherapy combined with chemoradiotherapy. Because esophageal squamous cell carcinoma is a highly malignant tumor, patients are often in the locally advanced stage at the time of initial diagnosis, so the case is often a patient with locally advanced esophageal squamous cell carcinoma. In the specific implementation process, the CT image data can be complete esophageal enhanced staging CT image data.

[0110] S2, obtain the CT image data and delineate the region of interest.

[0111] S3, extract radiomic features of the region of interest.

[0112] It should be noted that the imaging genomics features may include shape features, first-order statistical features, Gray Level Co-occurrence Matrix (GLCM) features, Gray Level Dependence Matrix (GLDM) features, Gray Level Run-Length Matrix (GLRLM) features, Gray Level Size Zone Matrix (GLSZM) features, Laplace features or wavelet features. The above imaging genomics features can reflect the internal structure and texture of the tumor, which is of great significance for the diagnosis and treatment of the tumor. Of course, other imaging genomics features may also be included, and the embodiments of the present application are not limited to this.

[0113] S4, input the imaging omics features into the target esophageal squamous cell carcinoma immunotherapy prediction model to obtain the prediction results.

[0114] Among them, the target esophageal squamous cell carcinoma immunotherapy prediction model is generated by training the extreme gradient boosting algorithm XGBoost model; the prediction result is used to indicate the prognosis status of the case receiving neoadjuvant immunotherapy combined with chemoradiotherapy. Prognosis refers to the treatment efficiency and survival time of tumor patients inferred based on clinical data.

[0115] In this way, the esophageal squamous cell carcinoma immunotherapy prediction method provided in the embodiment of the present application can be combined with radiomics, a non-invasive imaging analysis method, and through a pre-constructed target esophageal squamous cell carcinoma immunotherapy prediction model, it is possible to predict whether the patient will achieve complete pathological remission through neoadjuvant immunotherapy combined with chemoradiotherapy, thereby guiding immunotherapy. Compared with the invasive operation of obtaining tissue samples in the prior art, the embodiment of the present application can quickly predict the prognosis of esophageal squamous cell carcinoma patients after neoadjuvant immunotherapy combined with chemoradiotherapy, thereby improving the survival rate of esophageal squamous cell carcinoma patients to a certain extent.

[0116] See also Figure 2 , is a schematic diagram of a process for constructing an esophageal squamous cell carcinoma immunotherapy prediction model provided in an embodiment of the present application, which may include the following steps:

[0117] S01, obtaining sample CT image data corresponding to a plurality of case samples.

[0118] It should be noted that the sample CT image data is the image data of the case sample before receiving immunotherapy combined with chemoradiotherapy, and the case sample is a patient with esophageal squamous cell carcinoma, for example, a patient with locally advanced esophageal squamous cell carcinoma.

[0119] S02, obtaining the sample CT image data to delineate the sample region of interest.

[0120] In a possible implementation, 3DSlicer software may be used to delineate lesions in sample CT image data to obtain a region of interest.

[0121] S03, extracting radiomics features of the sample region of interest to obtain an initial radiomics feature set.

[0122] In the embodiment of the present application, for example, 1316 radiomics features can be extracted, and for the region of interest of the sample CT image data corresponding to each case sample in multiple case samples, the radiomics features of the sample region of interest are extracted to finally obtain an initial radiomics feature set.

[0123] In one possible implementation, in order to make the imaging genomics features better reflect the internal structure and texture of the tumor, the following principles can be followed in the process of extracting imaging genomics features: 1. Ensure that the extracted features are stable and repeatable, so use standardized and normalized methods for feature extraction; 2. Select features related to tumor properties and clinical outcomes to improve the accuracy and reliability of the prediction model; 3. Sufficient verification and optimization should be performed to ensure the generalization ability and practicality of the prediction model. Of course, the embodiments of this application do not limit the process of extracting imaging genomics features, and any method of extracting imaging genomics features should be regarded as part of the embodiments of this application.

[0124] S04, performing feature selection and dimensionality reduction on the initial radiomics feature set to obtain a target radiomics feature set.

[0125] In a possible implementation, S04 may include: S041, based on at least one of the Spearman correlation analysis method and the Mann-Whitney U test method, performing feature selection on the initial imaging omics feature set to obtain an intermediate imaging omics feature set; S042, using a support vector machine-recursive feature elimination method to eliminate the intermediate imaging omics feature set to obtain a target imaging omics feature set.

[0126] It should be noted that feature selection for the initial imaging omics feature set can be based on Spearman correlation analysis and Mann-Whitney U test at the same time, and feature selection can be performed from different angles to improve the accuracy of feature selection. The features to be retained are determined by the screened features using the support vector machine-recursive feature elimination method to obtain the target imaging omics feature set. In the embodiment of the present application, R software can be used to perform Spearman correlation analysis, Mann-Whitney U test and support vector machine-recursive feature elimination method on the initial imaging omics feature set. In the embodiment of the present application, the support vector machine-recursive feature elimination method is used to retain the top five imaging omics features.

[0127] S05, dividing the target radiomics feature set to obtain a model training set and a model test set.

[0128] In one possible implementation, a random seed can be used to divide the target imaging omics feature set to obtain a model training set and a model test set. In the embodiment of the present application, by setting a random seed, it can be ensured that the results of each data set division are consistent, thereby improving the reproducibility and comparability of the experiment.

[0129] S06, using the model training set to train the extreme gradient boosting algorithm XGBoost model to generate an initial esophageal squamous cell carcinoma immunotherapy prediction model, the XGBoost model is a model after hyperparameter optimization and obtaining the optimal hyperparameter combination.

[0130] Among them, hyperparameters refer to the external configuration variables of the model, which are parameters that need to be manually set during the model training process. They usually cannot be learned through the training process itself. Better hyperparameters can further improve the performance of the model.

[0131] In one possible implementation, in order to improve the accuracy of the generated esophageal squamous cell carcinoma immunotherapy prediction model, the hyperparameters of the XGBoost model architecture can be optimized first. Then, S06 provided in the embodiment of the present application may include: S061, defining a hyperparameter search space of the XGBoost model, where the hyperparameter search space stores the hyperparameters to be optimized; S062, searching in the hyperparameter search space using a Bayesian optimization algorithm to obtain multiple hyperparameter combinations; S063, evaluating the performance indicators of multiple hyperparameter combinations using cross-validation, and obtaining the hyperparameter combination with the best performance indicator as the optimal hyperparameter combination; S064, training the XGBoost model based on the model training set to generate an initial esophageal squamous cell carcinoma immunotherapy prediction model, and setting the optimal hyperparameter combination for the XGBoost model.

[0132] It should be noted that the hyperparameter search space of the XGBoost model may include, for example, the learning rate, the number of trees, the depth of the tree, the number of iterations, the batch size, etc. By optimizing the hyperparameters and finding the optimal hyperparameter combination, the generalization ability of the model can be further improved, thereby improving the model performance and accuracy. In the embodiment of the present application, K-fold cross validation (e.g., five-fold cross validation) can be used to evaluate the performance indicators of multiple hyperparameter combinations, and the hyperparameter combination with the best performance indicator is the optimal hyperparameter combination.

[0133] S07, based on the model test set, the model performance of the initial esophageal squamous cell carcinoma immunotherapy prediction model was evaluated to obtain the evaluation results.

[0134] In a possible implementation, S07 may include: S071 inputting the model test set into the initial esophageal squamous cell carcinoma immunotherapy prediction model to obtain a predicted probability; S072 evaluating the model performance of the initial esophageal squamous cell carcinoma immunotherapy prediction model based on the predicted probability and the actual probability in the model test set to obtain an evaluation result.

[0135] In one possible implementation, in order to obtain the calibration of the initial esophageal squamous cell carcinoma immunotherapy prediction model, S072 may include: performing a goodness of fit (Hosmer-Lemeshow) HL test on the initial esophageal squamous cell carcinoma immunotherapy prediction model according to the predicted probability and the actual probability in the model test set, and generating a calibration curve, wherein the calibration curve is generated based on the HL test results; based on the calibration curve, obtaining the calibration evaluation result of the initial esophageal squamous cell carcinoma immunotherapy prediction model.

[0136] In the embodiment of the present application, the HL test is a model fitting index, and its principle is to judge the gap between the predicted value and the true value. If the p value is greater than 0.05, it means that the HL test is passed, that is, there is no very obvious difference between the predicted value and the true value. On the contrary, if the p value is less than 0.05, it means that the HL test is not passed, and there is a significant difference between the predicted value and the true value, that is, the model fit is poor. Calibration curve, the calibration curve is a method of visualizing the HL test results, by comparing the predicted probability and the observed probability at different quantiles, as a graphical tool for evaluating the accuracy of the predicted probability.

[0137] In one possible implementation, in order to obtain the accuracy of the initial esophageal squamous cell carcinoma immunotherapy prediction model, S072 may include: generating a receiver operating characteristic (ROC) curve of the initial esophageal squamous cell carcinoma immunotherapy prediction model based on the predicted probability and the actual probability in the model test set; and obtaining the accuracy evaluation result of the initial esophageal squamous cell carcinoma immunotherapy prediction model based on the area under the curve AUC value of the ROC curve.

[0138] In the embodiment of the present application, the accuracy of the initial esophageal squamous cell carcinoma immunotherapy prediction model is obtained according to the area under the ROC curve (AUC) value. Among them, the larger the AUC value, the higher the accuracy of the initial esophageal squamous cell carcinoma immunotherapy prediction model obtained, wherein the range of the AUC value is [0, 1], and the closer the value is to 1, the more ideal the state is, indicating that the accuracy of the initial esophageal squamous cell carcinoma immunotherapy prediction model is higher, and the evaluation criteria are as follows: AUC values ​​below 0.7 are poor, [0.7, 0.8] are acceptable, [0.8, 0.9] are good, and [0.9, 1.0] are excellent.

[0139] See also Figure 3 and Figure 4 , respectively, the ROC curve graph and box plot of the model training set and the model test set provided in the embodiments of the present application, and the calibration curve graph of the model training set and the model test set, wherein the ROC curve and the calibration curve are used to evaluate the predictive ability of the esophageal squamous cell carcinoma immunotherapy prediction model in the neoadjuvant immunotherapy combined with chemoradiotherapy for locally advanced esophageal squamous cell carcinoma. The results showed that the AUC (area under the curve) of the ROC curve of the model training set was 0.89, the 95% confidence interval was 0.81-0.97, the classification accuracy was 85%, the classification sensitivity was 86%, and the classification specificity was 84%. The AUC (area under the curve) of the ROC curve of the model test set was 0.80, the 95% confidence interval was 0.64-0.97, the classification accuracy was 77%, the classification sensitivity was 84%, and the classification specificity was 69%.

[0140] In addition, the predicted probabilities of patients in the training set and the test set were displayed through box plots to provide a more intuitive understanding of the predictive performance of the model. In addition, in order to verify the stability of the model, the Hosmer-Lemeshow test and calibration curve were performed, which confirmed that the initial esophageal squamous cell carcinoma immunotherapy prediction model had good stability.

[0141] It should be noted that the discrimination is expressed by the ROC curve and the calibration is expressed by the calibration curve, but the above two indicators cannot provide a reference for clinical decision-making, so a decision curve analysis DCA can be performed. Decision curve analysis is a method to evaluate the application value of prediction models in actual clinical decision-making. It is achieved by comparing the net benefits of different decision-making schemes within a specific threshold range. The core principle of DCA is to quantify the utility of the prediction model at a specific clinical decision threshold and compare it with other decision-making strategies (such as treating everyone or not treating anyone).

[0142] In one possible implementation, S072 provided in an embodiment of the present application may include: generating a decision curve of an initial esophageal squamous cell carcinoma immunotherapy prediction model based on the predicted probability and the actual probability in the model test set; performing a decision curve analysis on the decision curve to obtain a net benefit evaluation result, and the net benefit evaluation result represents the application value of the initial esophageal squamous cell carcinoma immunotherapy prediction model in actual clinical decision-making.

[0143] See also Figure 5 , is a decision curve analysis diagram of the model training set and the model test set provided in the embodiments of the present application, Figure 5 In the DCA diagram, the threshold probability is the probability threshold for clinical decision makers to take a certain intervention measure. In other words, it is the predicted probability of the disease. When the probability is exceeded, the expected treatment benefit exceeds the risk. Net benefit is a measure of the utility of the model at a specific threshold probability, which takes into account the benefits of true positives minus the harm caused by false positives. Reference lines, in DCA diagrams, usually include several reference lines as a benchmark for comparison. The most common ones are the net benefit line of "Take no action" (TreatNone). Assuming that no patients are treated, the net benefit at this time is 0. The net benefit line of "TreatAll patients" (TreatAll), assuming that all patients are treated, this line shows the net benefit in this extreme case.

[0144] In the examples of the present application, the stability of the model was further evaluated through decision curve analysis. The results showed that the prediction results of the initial esophageal squamous cell carcinoma immunotherapy prediction model were relatively consistent with the actual observation results. The initial esophageal squamous cell carcinoma immunotherapy prediction model constructed by the method in steps S01 to S06 of the examples of the present application has high stability.

[0145] S08, when the performance evaluation result of the initial esophageal squamous cell carcinoma immunotherapy prediction model meets the preset requirements, the initial esophageal squamous cell carcinoma immunotherapy prediction model is obtained as the target esophageal squamous cell carcinoma immunotherapy prediction model.

[0146] If the performance evaluation results of the initial esophageal squamous cell carcinoma immunotherapy prediction model meet the preset requirements, the initial esophageal squamous cell carcinoma immunotherapy prediction model is obtained as the target esophageal squamous cell carcinoma immunotherapy prediction model; if the performance evaluation results of the initial esophageal squamous cell carcinoma immunotherapy prediction model do not meet the preset requirements, steps S02 to S07 are executed until the initial esophageal squamous cell carcinoma immunotherapy prediction model that meets the preset requirements is obtained.

[0147] It should be noted that all steps of the method for predicting immunotherapy of esophageal squamous cell carcinoma and the scheme for constructing an immunotherapy prediction model for esophageal squamous cell carcinoma provided in the embodiments of the present application are executed by a computer.

[0148] Thus, in the embodiments of the present application, a prediction model for esophageal cancer immunotherapy combined with chemoradiotherapy is constructed through imaging genomics features, thereby improving the efficacy and survival rate of patients with locally advanced esophageal squamous cell carcinoma.

[0149] See also Figure 6 The present application embodiment further provides an esophageal squamous cell carcinoma immunotherapy prediction device 600, the device 600 comprising:

[0150] The acquisition unit 601 is used to acquire computer tomography CT image data of the case; the CT image data is image data before neoadjuvant immunotherapy combined with chemoradiotherapy; the case is a patient with esophageal squamous cell carcinoma who is to receive neoadjuvant immunotherapy combined with chemoradiotherapy;

[0151] An obtaining unit 602 is used to obtain a region of interest after delineation of the CT image data;

[0152] An extraction unit 603 is used to extract radiomics features of the region of interest;

[0153] Input unit 604 is used to input the imaging genomics features into the target esophageal squamous cell carcinoma immunotherapy prediction model to obtain a prediction result; the target esophageal squamous cell carcinoma immunotherapy prediction model is generated by training the extreme gradient boosting algorithm XGBoost model; the prediction result is used to indicate the prognostic status of the case when receiving neoadjuvant immunotherapy combined with chemoradiotherapy.

[0154] Optionally, the device 600 further includes a model building unit 605, and the model building unit 605 includes:

[0155] The sample acquisition subunit is used to acquire sample CT image data corresponding to a plurality of case samples; the case samples are patients with locally advanced esophageal squamous cell carcinoma; the sample CT image data are image data of the case samples before receiving immunotherapy combined with chemoradiotherapy;

[0156] A sample region of interest obtaining subunit, used to obtain the sample region of interest after delineating the sample CT image data;

[0157] A feature extraction subunit, used to extract the radiomics features of the region of interest of the sample to obtain an initial radiomics feature set;

[0158] A preprocessing subunit, used for performing feature selection and dimensionality reduction on the initial radiomics feature set to obtain a target radiomics feature set;

[0159] A division subunit, used to divide the target radiomics feature set to obtain a model training set and a model test set;

[0160] A model training subunit is used to train the extreme gradient boosting algorithm XGBoost model using the model training set to generate an initial esophageal squamous cell carcinoma immunotherapy prediction model, wherein the XGBoost model is a model after hyperparameter optimization and obtaining the optimal hyperparameter combination;

[0161] A model evaluation subunit is used to perform a model performance evaluation on the initial esophageal squamous cell carcinoma immunotherapy prediction model based on the model test set to obtain an evaluation result;

[0162] The model acquisition subunit is used to obtain the initial esophageal squamous cell carcinoma immunotherapy prediction model as the target esophageal squamous cell carcinoma immunotherapy prediction model when the performance evaluation result of the initial esophageal squamous cell carcinoma immunotherapy prediction model meets the preset requirements.

[0163] Optionally, the model training subunit is specifically used to:

[0164] Defining a hyperparameter search space of the XGBoost model, wherein the hyperparameter search space stores hyperparameters to be optimized;

[0165] Using a Bayesian optimization algorithm to search the hyperparameter search space to obtain multiple hyperparameter combinations;

[0166] Use cross-validation to evaluate the performance indicators of multiple hyperparameter combinations, and obtain the hyperparameter combination with the best performance indicator as the optimal hyperparameter combination;

[0167] The XGBoost model is trained based on the model training set to generate the initial esophageal squamous cell carcinoma immunotherapy prediction model, and the XGBoost model sets the optimal hyperparameter combination.

[0168] Optionally, the preprocessing subunit is specifically used to:

[0169] Based on at least one of a Spearman correlation analysis method and a Mann-Whitney U test method, feature selection is performed on the initial radiomics feature set to obtain an intermediate radiomics feature set;

[0170] The intermediate radiomics feature set is eliminated by using a support vector machine-recursive feature elimination method to obtain the target radiomics feature set.

[0171] Optionally, the model evaluation subunit is specifically used to:

[0172] Inputting the model test set into the initial esophageal squamous cell carcinoma immunotherapy prediction model to obtain a prediction probability;

[0173] According to the predicted probability and the actual probability in the model test set, the model performance of the initial esophageal squamous cell carcinoma immunotherapy prediction model is evaluated to obtain the evaluation result.

[0174] Optionally, performing a model performance evaluation on the initial esophageal squamous cell carcinoma immunotherapy prediction model according to the predicted probability and the actual probability in the model test set to obtain the evaluation result includes:

[0175] According to the predicted probability and the actual probability in the model test set, a goodness-of-fit HL test is performed on the initial esophageal squamous cell carcinoma immunotherapy prediction model to generate a calibration curve, wherein the calibration curve is generated based on the HL test result;

[0176] Based on the calibration curve, a calibration evaluation result of the initial esophageal squamous cell carcinoma immunotherapy prediction model is obtained.

[0177] Optionally, performing a model performance evaluation on the initial esophageal squamous cell carcinoma immunotherapy prediction model according to the predicted probability and the actual probability in the model test set to obtain the evaluation result includes:

[0178] Generating a receiver operating characteristic (ROC) curve of the initial esophageal squamous cell carcinoma immunotherapy prediction model according to the predicted probability and the actual probability in the model test set;

[0179] Based on the area under the curve (AUC) value of the ROC curve, the accuracy evaluation result of the initial esophageal squamous cell carcinoma immunotherapy prediction model is obtained.

[0180] Optionally, performing a model performance evaluation on the initial esophageal squamous cell carcinoma immunotherapy prediction model according to the predicted probability and the actual probability in the model test set to obtain the evaluation result includes:

[0181] Generating a decision curve of the initial esophageal squamous cell carcinoma immunotherapy prediction model according to the predicted probability and the actual probability in the model test set;

[0182] A decision curve analysis is performed on the decision curve to obtain a net benefit evaluation result, and the net benefit evaluation result represents the application value of the initial esophageal squamous cell carcinoma immunotherapy prediction model in actual clinical decision-making.

[0183] Optionally, the imaging omics features include shape features, first-order statistical features, gray-level co-occurrence matrix features, gray-level dependence matrix features, gray-level run length matrix features, gray-level size area matrix features, Laplace features or wavelet features.

[0184] Optionally, the division subunit is specifically used to:

[0185] The target radiomics feature set is divided using a random seed to obtain the model training set and the model test set.

[0186] In addition, the present application embodiment also provides an electronic device 700, such as Figure 7 As shown, the electronic device 700 includes a processor 701 and a memory 702:

[0187] The memory 702 is used to store computer programs;

[0188] The processor 701 is used to execute according to the computer program Figure 1 or Figure 2 The method provided.

[0189] In addition, an embodiment of the present application further provides a computer-readable storage medium, wherein the computer-readable storage medium is used to store a computer program, and the computer program is used to execute the method provided by the embodiment of the present application.

[0190] Through the description of the above implementation methods, it can be known that those skilled in the art can clearly understand that all or part of the steps in the above-mentioned embodiment method can be implemented by means of software plus a general hardware platform. Based on such an understanding, the technical solution of the present application can be embodied in the form of a software product, which can be stored in a storage medium, such as a read-only memory (ROM) / RAM, a magnetic disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network communication device such as a router) to execute the methods described in each embodiment of the present application or some parts of the embodiments.

[0191] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the device embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment. The device embodiment described above is only schematic, in which the modules described as separate components may or may not be physically separated, and the components displayed as modules may or may not be physical modules, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules can be selected according to actual needs to achieve the goals of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative work.

[0192] The above is only a preferred embodiment of the present application and is not intended to limit the protection scope of the present application. It should be noted that, for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the present application, and these improvements and modifications should also be regarded as the protection scope of the present application.

Claims

1. A method for predicting immunotherapy of esophageal squamous cell carcinoma, characterized in that: The method comprises: Obtaining computed tomography (CT) image data of the case; the CT image data is image data before neoadjuvant immunotherapy combined with chemoradiotherapy; the case is a patient with esophageal squamous cell carcinoma who is to receive neoadjuvant immunotherapy combined with chemoradiotherapy; Obtaining a region of interest delineated from the CT image data; Extracting radiomic features of the region of interest; The imaging genomics features are input into a target esophageal squamous cell carcinoma immunotherapy prediction model to obtain a prediction result; the target esophageal squamous cell carcinoma immunotherapy prediction model is generated by training an extreme gradient boosting algorithm XGBoost model; the prediction result is used to indicate the prognostic status of the case when receiving neoadjuvant immunotherapy combined with chemoradiotherapy.

2. The method according to claim 1, characterized in that The method further comprises: Acquire sample CT image data corresponding to a plurality of case samples; the case samples are patients with esophageal squamous cell carcinoma; the sample CT image data are image data of the case samples before receiving immunotherapy combined with chemoradiotherapy; Obtaining a sample region of interest after outlining the sample CT image data; Extracting radiomics features of the sample region of interest to obtain an initial radiomics feature set; Performing feature selection and dimensionality reduction on the initial radiomics feature set to obtain a target radiomics feature set; Dividing the target radiomics feature set to obtain a model training set and a model test set; The extreme gradient boosting algorithm XGBoost model is trained using the model training set to generate an initial esophageal squamous cell carcinoma immunotherapy prediction model, wherein the XGBoost model is a model after hyperparameter optimization and obtaining the optimal hyperparameter combination; Performing a model performance evaluation on the initial esophageal squamous cell carcinoma immunotherapy prediction model based on the model test set to obtain an evaluation result; When the performance evaluation result of the initial esophageal squamous cell carcinoma immunotherapy prediction model meets the preset requirements, the initial esophageal squamous cell carcinoma immunotherapy prediction model is obtained as the target esophageal squamous cell carcinoma immunotherapy prediction model.

3. The method according to claim 2, characterized in that The method of using the model training set to train the extreme gradient boosting algorithm XGBoost model to generate an initial esophageal squamous cell carcinoma immunotherapy prediction model includes: Defining a hyperparameter search space of the XGBoost model, wherein the hyperparameter search space stores hyperparameters to be optimized; Using a Bayesian optimization algorithm to search the hyperparameter search space to obtain multiple hyperparameter combinations; Use cross-validation to evaluate the performance indicators of multiple hyperparameter combinations, and obtain the hyperparameter combination with the best performance indicator as the optimal hyperparameter combination; The XGBoost model is trained based on the model training set to generate the initial esophageal squamous cell carcinoma immunotherapy prediction model, and the XGBoost model sets the optimal hyperparameter combination.

4. The method according to claim 2, characterized in that The performing feature selection and dimensionality reduction on the initial radiomics feature set to obtain a target radiomics feature set includes: Based on at least one of a Spearman correlation analysis method and a Mann-Whitney U test method, feature selection is performed on the initial radiomics feature set to obtain an intermediate radiomics feature set; The intermediate radiomics feature set is eliminated by using a support vector machine-recursive feature elimination method to obtain the target radiomics feature set.

5. The method according to claim 2, characterized in that: The model performance evaluation of the initial esophageal squamous cell carcinoma immunotherapy prediction model based on the model test set is performed to obtain an evaluation result, including: Inputting the model test set into the initial esophageal squamous cell carcinoma immunotherapy prediction model to obtain a prediction probability; According to the predicted probability and the actual probability in the model test set, the model performance of the initial esophageal squamous cell carcinoma immunotherapy prediction model is evaluated to obtain the evaluation result.

6. The method according to claim 5, characterized in that The performing of a model performance evaluation on the initial esophageal squamous cell carcinoma immunotherapy prediction model according to the predicted probability and the actual probability in the model test set to obtain the evaluation result comprises: Generating a decision curve of the initial esophageal squamous cell carcinoma immunotherapy prediction model according to the predicted probability and the actual probability in the model test set; A decision curve analysis is performed on the decision curve to obtain a net benefit evaluation result, and the net benefit evaluation result represents the application value of the initial esophageal squamous cell carcinoma immunotherapy prediction model in actual clinical decision-making.

7. The method according to claim 1, characterized in that The imaging omics features include shape features, first-order statistical features, grayscale co-occurrence matrix features, grayscale dependence matrix features, grayscale run length matrix features, grayscale size area matrix features, Laplace features or wavelet features.

8. A device for predicting immunotherapy of esophageal squamous cell carcinoma, characterized in that: The device comprises: An acquisition unit is used to acquire computer tomography CT image data of a case; the CT image data is image data before neoadjuvant immunotherapy combined with chemoradiotherapy; the case is a patient with esophageal squamous cell carcinoma who is to receive neoadjuvant immunotherapy combined with chemoradiotherapy; An obtaining unit, used for obtaining a region of interest after delineation of the CT image data; An extraction unit, used for extracting radiomics features of the region of interest; An input unit is used to input the imaging genomics features into a target esophageal squamous cell carcinoma immunotherapy prediction model to obtain a prediction result; the target esophageal squamous cell carcinoma immunotherapy prediction model is generated by training the extreme gradient boosting algorithm XGBoost model; the prediction result is used to indicate the prognostic status of the case when receiving neoadjuvant immunotherapy combined with chemoradiotherapy.

9. An electronic device, characterized in that: The electronic device comprises a processor and a memory: The memory is used to store computer programs; The processor is configured to execute the method according to any one of claims 1 to 7 according to the computer program.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium is used to store a computer program, and the computer program is used to execute the method according to any one of claims 1 to 7.

Images