Esophageal cancer neoadjuvant chemotherapy and post-immunization pathological complete remission prediction system and method

Abstract

The invention discloses an esophageal cancer neoadjuvant chemotherapy and post-immunization pathology complete remission prediction system and method, and relates to the field of medical image processing, and the system comprises an image processing module which is used for carrying out the standardized processing of an input image, and segmenting a tumor region in the image; the feature extraction module is used for extracting voxel-level radiomics features from the tumor area; the feature screening module is used for screening out a plurality of voxel-level radiomics features with the most predictive capability; and the pathology prediction module is used for generating a comprehensive feature vector according to the segmented input image and the screened voxel-level radiomics features, and predicting the curative effect of the combination of neoadjuvant chemotherapy and immunotherapy according to the comprehensive feature vector. According to the technical scheme provided by the invention, the accuracy of predicting complete pathologic remission after neoadjuvant chemotherapy and immunotherapy of esophageal cancer can be improved, the defect that the heterogeneity in tumors cannot be considered in a traditional imaging omics technology is overcome, and the method has a relatively high application value.

Description

Technical Field

[0001] The present invention relates to the field of medical image processing, and in particular to a system and method for predicting complete pathological remission of esophageal cancer after neoadjuvant chemotherapy and immunization. Background Art

[0002] Esophageal squamous cell carcinoma is the most common type of esophageal cancer in Asia. Compared with other cancers, its morbidity and mortality are at a higher level. Neoadjuvant chemotherapy combined with immunotherapy has become one of the main treatments for locally advanced esophageal squamous cell carcinoma. However, there is currently a lack of biomarkers that can accurately predict the efficacy of neoadjuvant chemotherapy combined with immunotherapy in clinical practice.

[0003] Radiomics is a medical technology that extracts quantitative features from medical images to predict the treatment response and clinical outcomes of cancer patients. Traditional radiomics extracts statistical features of the tumor as a whole, such as texture and shape, and represents them with a single value. This technology can effectively reflect the macroscopic characteristics of the tumor, but it is difficult to take into account the complex heterogeneity and local details inside the tumor, which affects the accuracy of the judgment of the tumor condition. Especially in the treatment of esophageal cancer, the heterogeneity of the tumor is significant, and simple macroscopic features cannot fully reflect the microscopic changes of the tumor before and after treatment. In order to solve this problem, voxel-level radiomics technology came into being. This technology extracts features from each voxel of the tumor image to generate a feature map that reflects the subtle differences inside the tumor. It can capture the complex spatial heterogeneity inside the tumor, such as the changes in the edge and center of the tumor before and after treatment, or the comparison between the necrotic area and the active cell area inside the tumor. Through this meticulous feature extraction, the biological behavior of the tumor can be more accurately reflected, thereby providing more predictive information.

[0004] The development of deep learning technology has provided a new solution for predicting the efficacy of neoadjuvant chemotherapy combined with immunotherapy for esophageal cancer. It can automatically learn and extract deep features in images, but this technology is prone to overfitting when facing relatively small data sets, which in turn affects its versatility on different data sets. Since medical imaging data usually have the problem of limited sample size, deep learning technology is difficult to be well applied to the field of predicting the efficacy of neoadjuvant chemotherapy combined with immunotherapy for cancer. However, the emerging deep learning model Mamba has broken this dilemma. The model was originally designed in the field of RGB image processing, but its characteristics in processing long sequence data and capturing global contextual information enable it to effectively process high-dimensional data and capture complex spatial and temporal associations, and has high application potential in the field of medical image analysis.

[0005] Therefore, on this basis, this application proposes a system and method for predicting pathological complete remission of esophageal cancer after neoadjuvant chemotherapy and immunization that combines voxel-level imaging genomics technology and the Mamba deep learning model. Summary of the invention

[0006] Technical Purpose In order to solve the above problems, the purpose of the present invention is to provide a system and method for predicting pathological complete remission after neoadjuvant chemotherapy and immunotherapy for esophageal cancer, which can not only effectively improve the accuracy and reliability of the prediction of the probability of pathological complete remission after neoadjuvant chemotherapy combined with immunotherapy, but also can process multidimensional medical imaging data and capture complex spatiotemporal associations, so as to extract key information that is helpful for predicting the efficacy of neoadjuvant chemotherapy combined with immunotherapy under a limited sample size. The system and method can not only capture the complex heterogeneity and local details inside the tumor more carefully, but also reduce the risk of overfitting of the model by enriching the input information of the model. It has a wide range of application scenarios and provides an important decision-making basis for the clinical diagnosis of neoadjuvant chemotherapy combined with immunotherapy for esophageal cancer, greatly improving the treatment effect and survival probability of esophageal cancer patients.

[0007] Technical Solution In order to achieve the above-mentioned objectives, the present invention provides a system and method for predicting pathological complete remission of esophageal cancer after neoadjuvant chemotherapy and immunization. The system and method combine CT images and voxel-level imaging genomics features, and use deep learning technology to extract several key features that have the greatest impact on the prediction results from medical imaging data, thereby constructing a prediction model. In addition, the invention also involves an optimization strategy aimed at reducing the risk of overfitting and ensuring the versatility of the model on different data sets.

[0008] In a first aspect, the present invention provides a system for predicting pathological complete remission of esophageal cancer after neoadjuvant chemotherapy and immunization, comprising: An image processing module, used for standardizing the input image and segmenting the tumor area in the image; A feature extraction module to extract voxel-level radiomic features from tumor regions; Feature screening module, used to screen out several most predictive voxel-level radiomic features; The pathology prediction module is used to generate a comprehensive feature vector based on the segmented input image and the screened voxel-level radiomics features, and predict the efficacy of neoadjuvant chemotherapy combined with immunotherapy based on the comprehensive feature vector.

[0009] Furthermore, the input image includes a CT image.

[0010] Furthermore, the image processing module standardizes the input image in a manner that includes resampling the input image to an isotropic voxel size having a specific value.

[0011] Furthermore, the image processing module segments the tumor region according to the boundary of the tumor in the image.

[0012] Furthermore, the boundaries of the tumor are determined by multiple diagnostic methods.

[0013] Furthermore, the diagnostic method includes at least one of PET-CT, esophageal angiography and esophagoscopy.

[0014] Furthermore, PET-CT diagnosis was used to determine the metabolically active areas and anatomical boundaries of the tumor.

[0015] Furthermore, the invasion of the tumor into the esophageal cavity is obtained through esophageal contrast diagnosis.

[0016] Furthermore, the morphological characteristics of the tumor were obtained through esophagoscopy.

[0017] Furthermore, before the tumor boundary is finally determined, it is reviewed and modified by a senior doctor with more than 25 years of industry experience.

[0018] Furthermore, the output end of the image processing module is connected to the input end of the feature extraction module, and the tumor area image segmented by the image processing module is transmitted to the feature extraction module to extract voxel-level radiomics features of the tumor area.

[0019] Furthermore, the voxel-level radiomics features are extracted in a voxel-level manner.

[0020] Furthermore, the voxel-level radiomics features in the feature extraction module include intensity features and texture features.

[0021] Furthermore, the pathology prediction module determines the spatial heterogeneity and local changes within the tumor based on the voxel-level radiomics features.

[0022] Furthermore, the feature screening module screens voxel-level radiomics features through statistical analysis and machine learning methods.

[0023] Furthermore, the feature screening module determines the predictive ability of the feature by calculating the correlation between the voxel-level radiomics feature and the prediction target.

[0024] Furthermore, the feature screening module verifies the contribution of the voxel-level radiomics features to the system prediction results through a cross-validation method, and eliminates features with low contribution or redundant features based on the verification results.

[0025] Furthermore, the most predictive voxel-level radiomics features should show consistently high expression predictive ability in multiple cross-validations.

[0026] Furthermore, the pathology prediction module includes a Vision-Mamba model, which takes multidimensional data including voxel-level radiomics features and segmented tumor area images as input, and outputs a classification prediction of pathological complete remission after neoadjuvant chemotherapy combined with immunotherapy.

[0027] Furthermore, the Vision-Mamba model performs preliminary processing on multidimensional data through a series of 3D convolution operations, and captures long-range dependencies and global context information in medical images through a state-space model. The final comprehensive feature vector is classified and predicted through a fully connected layer, and the prediction result is optimized through the BCEWithLogitsLoss loss function to output the prediction result of whether the patient can achieve pathological complete remission after receiving neoadjuvant chemotherapy combined with immunotherapy.

[0028] Further optimization methods to make the Vision-Mamba model suitable for 3D medical imaging data include: ①Change the convolution layer from a two-dimensional convolution layer to a three-dimensional convolution layer; ②Change the batch normalization layer from a two-dimensional batch normalization layer to a three-dimensional batch normalization layer; ③ Flatten the 3D voxel data into a sequence form; ④ Adjust the state update equation of the selective scanning mechanism according to the characteristics of the three-dimensional data.

[0029] Furthermore, each input channel of the Vision-Mamba model is designed with an independent convolution kernel, and after multiple convolutions, pooling and batch normalization, the convolution results are spliced. Through multi-input structural design and optimized model architecture, especially the introduction of the state space model, the risk of overfitting of the model on small sample data sets is reduced, and the versatility and robustness of the model are improved.

[0030] Furthermore, the pathology prediction module pre-processes the received multi-dimensional data through the 3D convolution method, flattens and tokenizes the feature map, and divides it into multiple parts for the processing of the state space model. When processing high-dimensional data, the optimized 3D convolution operation and state space model effectively reduce the complexity of the calculation, have higher calculation efficiency under the same accuracy, and are suitable for operation in an environment with limited computing resources.

[0031] Furthermore, the pathology prediction module performs bidirectional scanning and long sequence dependency modeling on the spliced ​​features through a selective scanning mechanism and state-space sequence processing features to obtain the spatial-temporal dependency and global context information contained in the three-dimensional medical imaging data.

[0032] Furthermore, the pathology prediction module performs classification prediction and outputs the prediction result by passing the generated comprehensive feature vector through a fully connected layer. By combining CT images and voxel-level radiomics features, the model can more comprehensively capture the spatial heterogeneity and local changes within the tumor, improving the accuracy of the system's prediction of the probability of complete pathological remission of esophageal cancer after neoadjuvant chemotherapy combined with immunotherapy.

[0033] Furthermore, the Vision-Mamba model adjusts the weights parameter in the BCEWithLogitsLoss loss function so that the model can better balance the learning weights of different categories, and gives higher weights to label parameters with fewer samples to alleviate the problem of data imbalance, thereby improving the model's recognition ability for minority category samples and achieving more balanced classification performance. In the case of uneven distribution of positive and negative samples, the BCEWithLogitsLoss loss function handles the problem of category imbalance, ensuring that the performance of the model on minority categories will not be ignored, thereby improving the reliability of prediction.

[0034] Furthermore, the training of the Vision-Mamba model combines the Sigmoid activation function and the binary cross entropy loss function on the basis of the BCEWithLogitsLoss loss function, and handles the problem of class imbalance by adjusting the weights of positive samples.

[0035] Furthermore, interpretability analysis of the prediction results specifically includes calculating the contribution of each feature to the system prediction results.

[0036] In a second aspect, the present invention also provides a method for predicting pathological complete remission of esophageal cancer after neoadjuvant chemotherapy and immunization, the method is based on the system described in the first aspect, and comprises: The method comprises: Normalize the input image and segment the tumor area in the image; Extract voxel-level radiomic features from tumor regions; Several voxel-level radiomic features with the most predictive power were screened through statistical analysis and machine learning methods; Generate a comprehensive feature vector based on the segmented input image and the screened voxel-level radiomic features, and predict the efficacy of neoadjuvant chemotherapy combined with immunotherapy based on the comprehensive feature vector; Perform interpretability analysis on the prediction results and generate a SHAP value graph based on the analysis results.

[0037] In a third aspect, the present invention also provides a computer device, comprising a processor and a memory, wherein the processor is connected to the memory, the memory is used to store a computer program, and the processor is used to execute the computer program stored in the memory, so that the computer device executes at least one step of the aforementioned method for predicting pathological complete remission of esophageal cancer after neoadjuvant chemotherapy and immunization.

[0038] In a fourth aspect, the present invention further provides a computer-readable storage medium, wherein a computer program is stored in the computer-readable storage medium, and when the computer program is executed by a processor, at least one step of the aforementioned method for predicting pathological complete remission of esophageal cancer after neoadjuvant chemotherapy and immunization is implemented.

[0039] The present invention combines CT images and voxel-level radiomic features as the input data of the model, so that the model can capture the complex heterogeneity and local details inside the tumor. Compared with traditional radiomics methods, it can provide more detailed data support for neoadjuvant chemotherapy combined with immunotherapy; based on the original Mamba model, it has been modified to enable the model to adapt to multi-input three-dimensional medical imaging data, so that the model can process multiple types of medical imaging data input in parallel, thereby improving the processing efficiency and adaptability of the model. The system and method have a wide range of application scenarios, providing an important decision-making basis for the clinical diagnosis of neoadjuvant chemotherapy combined with immunotherapy for esophageal cancer, and greatly improving the treatment effect and survival probability of esophageal cancer patients.

[0040] Beneficial Effects By implementing the system and method for predicting complete pathological remission of esophageal cancer after neoadjuvant chemotherapy and immunization provided by the present invention, the following technical effects are achieved: (1) The present invention combines CT images and voxel-level radiomic features as the input data of the model, so that the model can capture the complex heterogeneity and local details inside the tumor. Compared with traditional radiomics methods, it can provide more detailed data support for neoadjuvant chemotherapy combined with immunotherapy. By combining CT images and voxel-level radiomic features, the model can more comprehensively capture the spatial heterogeneity and local changes inside the tumor, improving the accuracy of the system's prediction of the probability of complete pathological remission of esophageal cancer after neoadjuvant chemotherapy combined with immunotherapy; its model architecture has strong scalability. By adjusting the input voxel-level radiomic features and other medical imaging data, the model can be applied to other types of medical imaging tasks, and can be extended to the treatment of other tumor types or other clinical application fields.

[0041] (2) Modifications were made to the original Mamba model to adapt to multi-input three-dimensional medical imaging data, so that the model can process multiple types of medical imaging data inputs in parallel, thereby improving the processing efficiency and adaptability of the model. Through multi-input structural design and optimized model architecture, especially the introduction of the state-space model, the risk of overfitting of the model on small sample data sets is reduced, and the versatility and robustness of the model are improved; when processing high-dimensional data, the optimized 3D convolution operation and state-space model effectively reduce the complexity of calculations, and have higher computational efficiency at the same accuracy, which is suitable for operation in an environment with limited computing resources; in the case of uneven distribution of positive and negative samples, the BCEWithLogitsLoss loss function is used to deal with the problem of class imbalance, ensuring that the performance of the model on minority categories will not be ignored, thereby improving the reliability of predictions. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] In order to make the above-mentioned system and method for predicting complete pathological remission after neoadjuvant chemotherapy and immunization for esophageal cancer more obvious and easy to understand, the drawings required for use in the specific implementation of the present invention are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative work.

[0043] Figure 1 Schematic diagram showing the flow chart of the method for predicting pathological complete remission after neoadjuvant chemotherapy and immunotherapy for esophageal cancer. DETAILED DESCRIPTION

[0044] Embodiment 1: Provided are a system and method for predicting complete pathological remission of esophageal cancer after neoadjuvant chemotherapy and immunization, wherein the system for predicting complete pathological remission of esophageal cancer after neoadjuvant chemotherapy and immunization comprises: an image processing module, a feature extraction module, a feature screening module and a pathological prediction module; the method for predicting complete pathological remission of esophageal cancer after neoadjuvant chemotherapy and immunization is as follows: Figure 1 As shown, the details are as follows.

[0045] The image processing module is used to standardize the input image and segment the tumor area in the image.

[0046] The input image includes a CT image.

[0047] The image processing module standardizes the input image by resampling the input image into an isotropic voxel size of 1×1×5 mm.

[0048] The image processing module segments the tumor area according to the boundary of the tumor in the image.

[0049] The borders of the tumor are determined by various diagnostic modalities.

[0050] The diagnostic method includes at least one of PET-CT, esophageal angiography and esophagoscopy.

[0051] PET-CT is used to determine the metabolically active areas and anatomical boundaries of the tumor.

[0052] The invasion of the tumor into the esophageal cavity was obtained through esophageal contrast diagnosis.

[0053] The morphological characteristics of the tumor were obtained by esophagoscopy.

[0054] Before the tumor boundary is finally determined, it is reviewed and modified by a senior doctor with more than 25 years of industry experience.

[0055] The output end of the image processing module is connected to the input end of the feature extraction module, and the tumor area image segmented by the image processing module is transmitted to the feature extraction module to extract voxel-level radiomics features of the tumor area.

[0056] The feature extraction module is used to extract voxel-level radiomic features from the tumor region.

[0057] The voxel-level radiomics features are extracted in a voxel-level manner.

[0058] The voxel-level radiomics features in the feature extraction module include intensity features and texture features.

[0059] The voxel-level radiomic features of each tumor region included 17 intensity features and 73 texture features.

[0060] The pathology prediction module determines the spatial heterogeneity and local changes within the tumor based on the voxel-level radiomics features.

[0061] The feature screening module is used to screen out several voxel-level radiomic features with the most predictive ability.

[0062] The feature screening module screens voxel-level radiomic features through statistical analysis and machine learning methods, and selects X features with the highest predictive ability from 90 initial feature maps. The specific steps include: preliminary screening of all features through the training set, and evaluating the predictive ability of the features by calculating the correlation between each feature and the prediction target; then, further verifying the contribution of each feature to the prediction result through the cross-validation method, eliminating features with low contribution or redundant; finally, selecting features that show high predictive performance in multiple cross-validations to form the final X features.

[0063] The feature screening module determines the predictive ability of the feature by calculating the correlation between the voxel-level radiomics feature and the prediction target.

[0064] The feature screening module verifies the contribution of the voxel-level radiomics features to the system prediction results by a cross-validation method, and removes features with low contribution or redundancy according to the verification results.

[0065] The most predictive voxel-level radiomics features should show consistently high expression predictive ability in multiple cross-validations.

[0066] The pathology prediction module is used to generate a comprehensive feature vector based on the segmented input image and the screened voxel-level radiomics features, and predict the efficacy of neoadjuvant chemotherapy combined with immunotherapy based on the comprehensive feature vector.

[0067] The input data received by the pathology prediction module is a four-dimensional array with the shape of ,in is a voxel-level radiomics feature map, and the other dimension is the segmented tumor area image. The model preliminarily processes the four-dimensional array through a series of 3D convolution operations, and captures the long-range dependencies and global context information in medical images through the state space model. The final comprehensive feature vector is classified and predicted by the fully connected layer. Due to the imbalanced distribution of positive and negative samples in the prediction of pathological complete remission after neoadjuvant chemotherapy combined with immunotherapy, in order to improve the prediction ability of minority categories, the BCEWithLogitsLoss loss function is used in model training, combining the Sigmoid activation function and the binary cross entropy loss function, and adjusting the weight of positive samples to deal with the problem of class imbalance.

[0068] The pathology prediction module includes a Vision-Mamba model, which takes multidimensional data including voxel-level radiomics features and segmented tumor area images as input, and outputs a classification prediction of pathological complete remission after neoadjuvant chemotherapy combined with immunotherapy.

[0069] The Vision-Mamba model performs preliminary processing on multidimensional data through a series of 3D convolution operations, and captures long-range dependencies and global context information in medical images through a state-space model. The final comprehensive feature vector is classified and predicted through a fully connected layer, and the prediction result is optimized through the BCEWithLogitsLoss loss function to output the prediction result of whether the patient can achieve pathological complete remission after receiving neoadjuvant chemotherapy combined with immunotherapy.

[0070] Optimizations to make the Vision-Mamba model suitable for 3D medical imaging data include: ①Change the convolutional layer from nn.Conv2d to nn.Conv3d; ②Change the batch normalization layer from nn.BatchNorm2d to nn.BatchNorm3d; ③ Flatten the three-dimensional voxel data into a sequence form through the reshape operation; ④ Adjust the state update equation of the selective scanning mechanism according to the characteristics of the three-dimensional data.

[0071] The Vision-Mamba model is designed with an independent convolution kernel for each input channel. After multiple convolutions, pooling, and batch normalization, the features of each channel are concatenated in a fixed order before the fully connected layer.

[0072] The pathology prediction module preprocesses the received multi-dimensional data through the 3D convolution method. After convolution, the shape of the feature map becomes , by flattening and tokenizing the feature map, and splitting it into multiple parts for easy processing by the state-space model.

[0073] The pathology prediction module performs bidirectional scanning and long sequence dependency modeling on the spliced ​​features through a selective scanning mechanism and state-space sequence processing features to obtain the spatial-temporal dependency and global context information contained in the three-dimensional medical imaging data, and finally generates a comprehensive feature vector.

[0074] The pathology prediction module performs classification prediction by passing the generated comprehensive feature vector through a fully connected layer and outputs a prediction result.

[0075] The Vision-Mamba model adjusts the weights parameter in the BCEWithLogitsLoss loss function so that the model can better balance the learning weights of different categories, and gives higher weights to label parameters with fewer samples to alleviate the problem of data imbalance, thereby improving the model's ability to recognize minority category samples and achieving more balanced classification performance.

[0076] The training of the Vision-Mamba model combines the Sigmoid activation function and the binary cross entropy loss function on the basis of the BCEWithLogitsLoss loss function, and handles the problem of class imbalance by adjusting the weights of positive samples.

[0077] The interpretability analysis of the prediction results specifically includes calculating the contribution of each feature to the system's prediction results.

[0078] It will be appreciated by those skilled in the art that embodiments of the present invention may be provided as methods, systems or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable non-transient storage media containing computer-usable program code.

[0079] The present invention can provide computer program instructions to a processor of a general-purpose computer, a special-purpose computer, an embedded processor or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device produce means for implementing the system.

[0080] These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer-readable memory produce a product including an instruction device that implements the functions of the system.

[0081] These computer program instructions may also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, whereby the instructions executed on the computer or other programmable device provide steps for implementing the functions of the system.

Claims

1. A prediction system for pathological complete remission of esophageal cancer after neoadjuvant chemotherapy and immunization, characterized by: include: An image processing module, used for standardizing the input image and segmenting the tumor area in the image; A feature extraction module to extract voxel-level radiomic features from tumor regions; Feature screening module, used to screen out several most predictive voxel-level radiomic features; The pathology prediction module is used to generate a comprehensive feature vector based on the segmented input image and the screened voxel-level radiomics features, and predict the efficacy of neoadjuvant chemotherapy combined with immunotherapy based on the comprehensive feature vector.

2. The system according to claim 1, characterized in that: The image processing module standardizes the input image in a manner that includes resampling the input image to an isotropic voxel size having a specific value.

3. The system according to claim 1, characterized in that: The voxel-level radiomics features in the feature extraction module include intensity features and texture features; and / or The pathology prediction module determines the spatial heterogeneity and local changes within the tumor based on the voxel-level radiomics features.

4. The system according to claim 1, characterized in that: The feature screening module determines the predictive ability of the feature by calculating the correlation between the voxel-level radiomics feature and the prediction target, verifies the contribution of the feature to the system prediction result by a cross-validation method, and eliminates features with low contribution or redundant features based on the verification results.

5. The system according to claim 1, characterized in that: The pathology prediction module includes a Vision-Mamba model, which takes multidimensional data including voxel-level radiomics features and segmented tumor area images as input, and outputs a classification prediction of pathological complete remission after neoadjuvant chemotherapy combined with immunotherapy.

6. The system according to claim 5, characterized in that: The Vision-Mamba model performs preliminary processing on multidimensional data through a series of 3D convolution operations, and captures long-range dependencies and global context information in medical images through a state-space model. The final comprehensive feature vector is classified and predicted through a fully connected layer, and the prediction result is optimized through the BCEWithLogitsLoss loss function to output the prediction result of whether the patient can achieve pathological complete remission after receiving neoadjuvant chemotherapy combined with immunotherapy.

7. The system according to claim 6, characterized in that: The Vision-Mamba model adjusts the weights parameter in the BCEWithLogitsLoss loss function so that the model can better balance the learning weights of different categories, and alleviates the problem of data imbalance by giving higher weights to label parameters with fewer samples.

8. A method for predicting pathological complete remission of esophageal cancer after neoadjuvant chemotherapy and immunization, characterized by: The method is implemented based on the system according to any one of claims 1 to 7: The method comprises: Normalize the input image and segment the tumor area in the image; Extract voxel-level radiomic features from tumor regions; Several voxel-level radiomic features with the most predictive power were screened through statistical analysis and machine learning methods; Generate a comprehensive feature vector based on the segmented input image and the screened voxel-level radiomic features, and predict the efficacy of neoadjuvant chemotherapy combined with immunotherapy based on the comprehensive feature vector; Perform interpretability analysis on the prediction results and generate a SHAP value graph based on the analysis results.

9. A computer device comprising a processor and a memory, wherein the processor is connected to the memory, and the memory is used to store a computer program, wherein: The processor is configured to execute the computer program stored in the memory, so that the computer device performs at least one step of the method according to claim 8.

10. A computer-readable storage medium, wherein a computer program is stored in the computer-readable storage medium, characterized in that: The computer program implements at least one step of the method of claim 8 when executed.

Images