Methods, Systems, Devices, and Media for Predicting the Efficacy of Immunotherapy in TNBC Patients

Abstract

The present application relates to the field of medical technologies, and discloses a method, a system, a device and a medium for predicting the efficacy of immunotherapy for TNBC patients, including: obtaining sample data on TNBC patients; using the prognostic review information of TNBC patients in the retrospective cohort as the outcome variable, and screening to obtain the pathological omics features corresponding to the WSI baseline pathological images in the retrospective cohort; extracting baseline radiomics features, performing correlation analysis, and obtaining radiomics features with pathological interpretability; based on the early MRI images and the MRI baseline images, using the radiomics features with pathological interpretability as labels to train a model, and obtaining a trained prediction model; validating the prediction model based on the multi-temporal MRI images in the prospective cohort to obtain a target prediction model, so as to obtain a prediction result of the treatment efficacy. The beneficial effect is to more accurately evaluate the treatment effect, so as to be able to judge the treatment effect earlier and more accurately.

Description

Technical Field

[0001] This application relates to the field of medical technologies, and particularly to a method, system, device, and medium for predicting the efficacy of immunotherapy for TNBC patients. Background Art

[0002] With the rise of immunotherapy, which activates the immune system to combat tumors, it has become a potential breakthrough in the treatment of TNBC. Research shows that TNBC has a relatively high tumor mutation burden and strong immunogenicity. Therefore, compared with other cancer types, TNBC patients are more likely to benefit from immunotherapy. For example, in the KEYNOTE-355 clinical trial, the PD-1 inhibitor pembrolizumab combined with chemotherapy as a first-line treatment regimen can significantly extend the overall survival of advanced TNBC patients (23 months vs. 16.1 months). However, although immunotherapy has potential, it has problems such as high costs, a relatively low remission rate (about 52.7%), and immune-related adverse reactions. Therefore, evaluating the efficacy of immunotherapy has become the key to improving treatment effects and reducing side effects.

[0003] Currently, the evaluation of the efficacy of immunotherapy mainly relies on the "Immune Modification Response Evaluation Criteria" (iRECIST) proposed in 2017, which evaluates the efficacy by measuring tumor diameters. However, this criterion has certain limitations in the evaluation of immunotherapy. The process of immunotherapy activating the body's immune system to combat tumors is relatively slow, and imaging examinations are often lagging, and may lead to the appearance of the "pseudo-progression" phenomenon, which is easily confused with "true progression". Therefore, it is necessary to accurately distinguish these different types of progression, and misjudgment may lead to incorrect treatment decisions. Moreover, immunotherapy may change the tumor microenvironment, affecting the internal structure and metabolic activities of tumors. These changes often do not immediately reflect in tumor size but may significantly affect the treatment response.

[0004] Therefore, there is an urgent need for a more precise evaluation method to be able to judge the treatment effect earlier and more accurately, avoiding prematurely terminating effective treatment or delaying the replacement of treatment regimens with severe adverse reactions. Summary of the Invention

[0005] This application provides a method, system, device, and medium for predicting the efficacy of immunotherapy for TNBC patients, which solves the technical problem of inaccurate evaluation of the efficacy of immunotherapy and achieves the technical effect of more precisely evaluating the treatment effect so as to be able to judge the treatment effect earlier and more accurately.

[0006] To achieve the above object, the main technical solutions adopted in this application include:

[0007] In a first aspect, an embodiment of the present application provides a method for predicting the efficacy of immunotherapy for TNBC patients, the method comprising:

[0008] Obtain sample data regarding TNBC patients; wherein, the sample data includes a retrospective cohort as a training set and a prospective cohort as a validation set;

[0009] Using the prognostic review information of the TNBC patients in the retrospective cohort as the outcome variable, screen for the pathological omics features corresponding to the WSI baseline pathological images in the retrospective cohort;

[0010] Extract the baseline imaging omics features corresponding to the MRI baseline imaging images in the retrospective cohort, and perform a correlation analysis on the baseline imaging omics features and the pathological omics features to obtain imaging omics features with pathological interpretability;

[0011] Based on the MRI early imaging images and the MRI baseline imaging images in the retrospective cohort, use the imaging omics features with pathological interpretability as labels to train a model to obtain a trained prediction model;

[0012] Validate the prediction model based on the MRI multi-temporal imaging images in the prospective cohort to obtain a target prediction model for obtaining the efficacy prediction result.

[0013] The method for predicting the efficacy of immunotherapy for TNBC patients provided in this embodiment collects sample data regarding TNBC patients, including a retrospective cohort as a training set and a prospective cohort as a validation set, ensuring that the model can take into account the learning of historical data and the prediction and validation of future data, thereby improving the generalization ability of the model. Using the prognostic review information in the retrospective cohort as the outcome variable, relevant pathological omics features corresponding to the baseline pathological images are screened out, providing basic data for subsequent analysis. Further extract the imaging omics features corresponding to the MRI imaging images in the retrospective cohort and perform a correlation analysis to reveal the relationship between the imaging omics features and the pathological omics features, and then obtain imaging omics features with pathological interpretability. Using these imaging omics features as labels, combined with the early imaging images and baseline imaging images of MRI in the retrospective cohort for model training, finally a trained prediction model is obtained. Finally, by using the MRI multi-temporal imaging images in the prospective cohort to validate the prediction model, a target prediction model is obtained, improving the prediction accuracy of the model for the prognosis and efficacy of TNBC patients.

[0014] Optionally, the step of using the prognostic review information of the TNBC patients in the retrospective cohort as the outcome variable and screening for the pathological omics features corresponding to the WSI baseline pathological images in the retrospective cohort includes:

[0015] Perform nuclear segmentation on the WSI baseline pathology images in the retrospective cohort to obtain the position information of each nucleus;

[0016] Based on the position information, construct a multi-level pairwise relationship graph; wherein, each relationship graph characterizes the relationship between different nuclear types;

[0017] Determine the topological features of each nucleus in each relationship graph and construct the comprehensive topological features of each nucleus;

[0018] Using the prognostic review information of the TNBC patients in the retrospective cohort as the outcome variable, analyze the relationship between the comprehensive topological features and the prognostic review information to obtain the pathomics features corresponding to the WSI baseline pathology images in the retrospective cohort.

[0019] In this embodiment, nuclear segmentation is first performed on the WSI baseline pathology images in the retrospective cohort to obtain the position information of each nucleus, which can accurately identify the nuclei in the pathology images and provide basic data for subsequent analysis. Then, based on the obtained nuclear position information, a multi-level pairwise relationship graph is constructed, which can depict the interconnections between nuclei and reveal the association patterns of nuclear types. Next, the topological features of each nucleus in the relationship graph are determined, and the comprehensive topological features of each nucleus are constructed. Through topological analysis, the spatial structure features of the nuclei can be comprehensively characterized, providing in-depth data information for further analysis. Finally, using the prognostic review information of the TNBC patients in the retrospective cohort as the outcome variable, analyze the relationship between the comprehensive topological features and the prognostic information, and then obtain the pathomics features corresponding to the WSI baseline pathology images in the retrospective cohort, which can establish the association between the pathological features and the patient prognosis, thus providing more guiding pathomics information for clinical practice.

[0020] Optionally, the constructing a multi-level pairwise relationship graph based on the position information includes:

[0021] For any nucleus, based on the position information, determine the nearest neighbors of the any nucleus corresponding to different types of nuclei;

[0022] Under the condition of meeting the preset edge configuration condition, establish an edge between the any nucleus and the nearest neighbor;

[0023] Based on the type of the nucleus and the edge, construct the relationship graph.

[0024] In this embodiment, based on the position information of the cell nuclei, the nearest neighbor relationship between each cell nucleus and different types of cell nuclei is first determined, thereby revealing the relative layout of the cell nuclei in space. Then, according to the preset edge configuration conditions, the connection relationship between the cell nuclei is established, which helps to clarify the interaction patterns between different cell nucleus types. Finally, by combining the type information of the cell nuclei and the connection relationship of the edges, a complete relationship graph is constructed. This graph can comprehensively display the spatial associations and structural relationships between the cell nuclei, providing a basic framework for subsequent topological analysis and research on cell-cell interactions.

[0025] Optionally, the preset edge configuration conditions are as follows:

[0026]

[0027] where E is the set of all edges; is the type of cell nucleus V i ; is the type of cell nucleus V j ; D(V i , V j ) is the Euclidean distance between cell nucleus V i and cell nucleus V j ; T is the edge configuration threshold.

[0028] Optionally, at least one sub-feature is included in the topological features, and the sub-feature is determined in the following manner:

[0029] For any target cell nucleus among the respective cell nuclei, obtain the target relationship graph where the target cell nucleus is located;

[0030] In the target relationship graph, determine the distance information between each neighboring cell nucleus and the target cell nucleus, and integrate the determined distance information into a type distance sequence according to the types of the respective neighboring cell nuclei;

[0031] For any type distance data in the type distance sequence, query in the preset standard library whether there is a matching type distance template;

[0032] If there is a match, use the sequence feature corresponding to the matching type distance template as a sub-feature in the topological features of the target cell nucleus;

[0033] If there is no match, create a type distance template for the type distance data in the preset standard library, and use the sequence feature corresponding to the newly created type distance template as a sub-feature in the topological features of the target cell nucleus.

[0034] In this embodiment, by obtaining the relationship graph where the target cell nucleus is located, it is ensured that the target cell nucleus can be accurately located in the overall layout, laying a foundation for subsequent analysis. Then, the distances between the target cell nucleus and adjacent cell nuclei are determined, and integrated into a type-distance sequence according to the types of adjacent cell nuclei, revealing the spatial distribution characteristics of different types of cell nuclei. If a matching type-distance template is found in the preset standard library, the corresponding sequence features are used as the sub-features of the target cell nucleus, improving the analysis efficiency and enhancing the accuracy. If no matching template is found, a new template can be created in the standard library and used as part of the topological features of the target cell nucleus, ensuring the flexibility of the system and expanding the standard library. Overall, it can not only accurately describe the spatial relationship between cell nuclei, but also provide more comprehensive and accurate analysis for research in fields such as cell biology and pathology during the process of dynamically updating and expanding the standard library.

[0035] Optionally, the correlation analysis of the baseline radiomics features and the pathomics features to obtain radiomics features with pathological interpretability includes:

[0036] Stitch the baseline radiomics features and the pathomics features to obtain a fused feature set;

[0037] Eliminate redundant features in the fused feature set to obtain a target feature set;

[0038] Based on the target set, perform correlation analysis using the Spearman correlation coefficient to obtain the radiomics features with pathological interpretability.

[0039] In this embodiment, by stitching the baseline radiomics features and the pathomics features, a fused feature set is formed, providing a comprehensive data basis for subsequent analysis. Then, by eliminating redundant features in the fused feature set, a target feature set is obtained, thus simplifying the data and improving the efficiency and accuracy of the analysis. Finally, based on the target feature set, correlation analysis is performed using the Spearman correlation coefficient to screen out radiomics features with pathological interpretability, providing a valuable basis for further pathological research and clinical diagnosis.

[0040] Optionally, the eliminating redundant features in the fused feature set to obtain a target feature set includes:

[0041] Determine the correlation coefficient of each fused feature in the fused feature set;

[0042] Determine the fused features with correlation coefficients greater than a preset threshold as the redundant features;

[0043] Eliminate the redundant features from the fused feature set to obtain the target feature set.

[0044] In this embodiment, the similarity or correlation between features is evaluated by calculating the correlation coefficient of each fused feature, providing a basis for subsequent screening of redundant features. Then, the fused features with a correlation coefficient greater than a preset threshold are identified as redundant features. These features are repetitive or highly similar to other feature information and may have a negative impact on the analysis results. Finally, by removing these redundant features from the fused feature set, a target feature set is obtained, ensuring that the finally retained features are more independent and have high information content, thereby improving the accuracy and efficiency of subsequent analysis.

[0045] Optionally, training a model using the radiomics features with pathological interpretability as labels based on the early MRI images and the baseline MRI images in the retrospective cohort to obtain a trained prediction model includes:

[0046] Extracting the early radiomics features corresponding to the early MRI images in the retrospective cohort and extracting the baseline radiomics features corresponding to the baseline MRI images;

[0047] Inputting the early radiomics features and the baseline radiomics features into the UniFormer model for training, and training the UniFormer model using the radiomics features with pathological interpretability as labels to obtain the prediction model.

[0048] In this embodiment, the radiomics features of the early MRI images and the baseline MRI images in the retrospective cohort are extracted. These features contain important information reflecting the imaging data. The extracted early and baseline radiomics features are input into the UniFormer model for training. At the same time, using the radiomics features with pathological interpretability as labels further improves the accuracy of the model and the prediction ability for pathological changes. Through this process, a prediction model based on radiomics features is finally obtained, which can effectively identify and predict the progress and changes of the curative effect.

[0049] Optionally, the total loss function of the UniFormer model includes a decorrelation loss, and the acquisition method of the decorrelation loss is as follows:

[0050] Obtaining the deep features and radiological features after each iterative optimization of the UniFormer model;

[0051] Based on the deep features, the radiological features, and preset weight parameters, the decorrelation loss is determined using an exponentially weighted method.

[0052] In this embodiment, after each iterative optimization of the model, deep features and radiological features are obtained. These features respectively reflect the high-level semantic information and radiological features of the image, which helps the model to comprehensively understand the imaging data. By combining these deep features and radiological features and according to the preset weight parameters, the exponential weighting method is used to determine the decorrelation loss, thereby effectively reducing the redundant information between features and improving the independence of features. Through this processing, the model can more accurately capture the key information of the image, improve the prediction effect, and enhance the robustness of the model.

[0053] Optionally, the decorrelation loss is:

[0054]

[0055] where L corr is the decorrelation loss; w is the preset weight parameter, w < 1; N k is the total number of deep features and radiological features; Z i is the i-th deep feature; R i is the i-th radiological feature.

[0056] Optionally, the method further includes evaluating the prediction model:

[0057] The prediction model is evaluated using an ROC curve, a decision curve, or a calibration curve.

[0058] Optionally, the method further includes: analyzing the transcriptome data in the prospective cohort and the prospective radiomics features corresponding to the multi-temporal MRI images in the prospective cohort to obtain the interpretability analysis results of the radiomics features and genes, including:

[0059] Based on the prognostic prospective information of the TNBC patients in the prospective cohort, the TNBC patients are divided into treatment responders and treatment non-responders;

[0060] Based on the transcriptome data in the prospective cohort, differential expression analysis is performed on the treatment responders and the treatment non-responders to determine differentially expressed genes;

[0061] Based on the differentially expressed genes, enrichment analysis is performed to obtain the biological pathways of the differentially expressed genes;

[0062] Extract the prospective radiomics features corresponding to the multi-temporal MRI images in the prospective cohort;

[0063] Based on the prediction model and the pathways, determine the interpretability analysis results of the prospective radiomics features and genes.

[0064] In this embodiment, according to the prognostic information in the prospective cohort, TNBC patients are divided into treatment responders and non-responders, providing a clear grouping basis for subsequent analysis. Then, differential expression analysis is performed on these two groups of patients based on transcriptome data to identify key genes affecting treatment response. Enrichment analysis is performed on these differentially expressed genes to reveal the biological pathways they may be involved in. By extracting radiomics features from multi-temporal MRI images, the dimension of clinical information is further expanded, laying a foundation for subsequent multi-modal analysis. Finally, correlation analysis is performed between the radiomics features and gene expression data and their related pathways to obtain interpretable analysis results between imaging features and genes. This analysis not only helps to discover potential biomarkers for treatment response but also improves the predictive ability of radiomics features in tumor treatment.

[0065] In a second aspect, an embodiment of the present application provides a TNBC patient immunotherapy efficacy prediction system, which includes:

[0066] A data acquisition unit for acquiring sample data on TNBC patients; wherein the sample data includes a retrospective cohort as a training set and a prospective cohort as a validation set;

[0067] A pathomics feature acquisition unit for using the prognostic review information of the TNBC patients in the retrospective cohort as an outcome variable to screen out the pathomics features corresponding to the WSI baseline pathology images in the retrospective cohort;

[0068] An interpretable feature acquisition unit for extracting baseline radiomics features corresponding to the MRI baseline images in the retrospective cohort and performing correlation analysis on the baseline radiomics features and the pathomics features to obtain pathologically interpretable radiomics features;

[0069] A prediction model acquisition unit for training a model based on the MRI early images and the MRI baseline images in the retrospective cohort using the pathologically interpretable radiomics features as labels to obtain a trained prediction model;

[0070] A model verification unit for verifying the prediction model based on the MRI multi-temporal images in the prospective cohort to obtain a target prediction model for obtaining an efficacy prediction result.

[0071] In a third aspect, an embodiment of the present application provides a computer device, including:

[0072] A memory and a processor, which are communicatively connected to each other. The memory stores computer instructions, and the processor executes the computer instructions to execute the above-mentioned TNBC patient immunotherapy efficacy prediction method.

[0073] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, on which computer instructions are stored, and the computer instructions are used to cause a computer to execute the TNBC patient immunotherapy efficacy prediction method described above. Description of the Drawings

[0074] In order to more clearly illustrate the specific embodiments of the present application or the technical solutions in the prior art, the following will briefly introduce the drawings required for use in the description of the specific embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present application. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0075] Figure 1 It is a flowchart of a TNBC patient immunotherapy efficacy prediction method provided by an embodiment of the present application;

[0076] Figure 2 It is a flowchart of step S3 provided by an embodiment of the present application;

[0077] Figure 3 It is a flowchart of step S33 provided by an embodiment of the present application;

[0078] Figure 4 It is a flowchart of how sub-features are determined according to the following manner provided by an embodiment of the present application;

[0079] Figure 5 It is a flowchart of step S5 provided by an embodiment of the present application;

[0080] Figure 6 It is a flowchart of step S53 provided by an embodiment of the present application;

[0081] Figure 7 It is a flowchart of step S7 provided by an embodiment of the present application;

[0082] Figure 8 It is an architecture diagram of UniFormer provided by an embodiment of the present application;

[0083] Figure 9 It is a flowchart of the acquisition method of the decorrelation loss provided by an embodiment of the present application;

[0084] Figure 10 It is a flowchart of another TNBC patient immunotherapy efficacy prediction method provided by an embodiment of the present application;

[0085] Figure 11 It is a block diagram for describing the specific implementation provided by an embodiment of the present application;

[0086] Figure 12 Block diagram of an immune therapy efficacy prediction system provided by an embodiment of the present application;

[0087] Figure 13 Structural schematic diagram of a computer device provided by an embodiment of the present application. Detailed implementation manners

[0088] To make the objectives, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present application. Apparently, the described embodiments are some, but not all, of the embodiments of the present application. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present application without creative efforts shall fall within the protection scope of the present application.

[0089] Currently, the evaluation of immune therapy efficacy mainly relies on the "Immune-related Response Evaluation Criteria" (iRECIST) proposed in 2017, which evaluates the efficacy by measuring tumor diameters. However, this criterion has certain limitations in the evaluation of immune therapy. The process of activating the body's immune system to fight tumors by immune therapy is relatively slow, and imaging examinations are often lagging, and may lead to the appearance of the "pseudo-progression" phenomenon, which is easily confused with "true progression". Therefore, it is necessary to accurately distinguish these different types of progression, and misjudgment may lead to incorrect treatment decisions. Moreover, immune therapy may change the tumor microenvironment, affecting the internal structure and metabolic activities of tumors. These changes are often not immediately reflected in tumor size, but may significantly affect the treatment response.

[0090] The application of MRI functional imaging technology in tumor treatment is of great significance. It can provide key information about tumor hemodynamics, cell size, density, and metabolism, and can reflect the infiltration status of immune cells as well as related protein and gene phenotypes, thereby capturing the real-time changes of tumors and their microenvironments. The dynamic changes in the immune microenvironment are closely related to the efficacy of immunotherapy and the prognosis of patients. Therefore, MRI functional imaging has the potential in monitoring the efficacy of immunotherapy. However, existing technologies still face some limitations, especially the limited parameters derived from MRI, which cannot comprehensively reveal the complex ecological environment during tumor immunotherapy. For example, it cannot provide a complete picture of the dynamic changes of tumors during treatment, such as treatment response, changes in tumor growth rate, remodeling of the microenvironment, and development of drug resistance, resulting in limitations in the evaluation of immune responses. In addition, although machine learning has made some progress in tumor image analysis, most current models lack interpretability and are difficult to establish connections with disease-specific biological mechanisms, mainly due to their disconnection from potential pathophysiological mechanisms. Therefore, how to accurately reflect the changes in tumors and their microenvironments through multi-dimensional imaging data and machine learning models and form an effective treatment decision support tool remains a challenge to be solved urgently.

[0091] According to an embodiment of the present application, an embodiment of a method for predicting the efficacy of immunotherapy for TNBC patients is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although the logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in a different order than here.

[0092] In this embodiment, a method for predicting the efficacy of immunotherapy for TNBC patients is provided. Figure 1 The flowchart of a method for predicting the efficacy of immunotherapy for TNBC patients provided by an embodiment of the present application is as Figure 1 shown, and this process includes the following steps:

[0093] Step S1, obtaining sample data about TNBC patients; among them, the sample data includes a retrospective cohort as a training set and a prospective cohort as a validation set.

[0094] Specifically, the retrospective cohort refers to a set of sample data from the start of immunotherapy to the completion of immunotherapy within a past time period, including MRI baseline image, MRI early image, WSI pathology image, and prognostic retrospective information. The prospective cohort refers to sample data collected from the current time point for a future time period, including MRI multi-temporal images, prognostic prospective information, and transcriptome data.

[0095] The selection criteria for the retrospective cohort include inclusion criteria and exclusion criteria, which are shown as follows:

[0096] Inclusion criteria:

[0097] (1) Patients must be diagnosed with triple-negative breast cancer (TNBC) by pathological puncture biopsy;

[0098] (2) Patients with stage III breast cancer involving the skin, chest wall or extensive lymph node involvement or stage IV breast cancer with metastases;

[0099] (3) Patients receive a treatment regimen of immunotherapy combined with chemotherapy;

[0100] (4) According to the modified immune response evaluation criteria in solid tumors (iRECIST), patients have at least 1 measurable lesion;

[0101] (5) Patients have MRI imaging, WSI pathological images and prognostic information at the same time, and the collection methods are as follows:

[0102] MRI imaging: Standard breast multi-parametric MRI scans are performed before puncture, and the MRI baseline images before immunotherapy and the MRI early images in the early stage of immunotherapy (after the end of the first course) are collected;

[0103] WSI pathological images: The pathological images are scanned using a slide scanner, and a digital image WSI database is established;

[0104] Prognostic information: The overall survival and recurrence-free survival information are collected through electronic medical record data; among them, the overall survival refers to the time from the start of treatment of the patient to the last follow-up or the death of the patient; the recurrence-free survival refers to the time from the completion of the initial treatment of the patient to the first detection of a new local recurrence or distant metastasis by imaging examination.

[0105] Exclusion criteria:

[0106] (1) Incomplete clinicopathological data;

[0107] (2) Patients who discontinued treatment due to severe side effects;

[0108] (3) Artifacts, incomplete sequences, poor quality, etc. in MRI imaging or WSI pathological images.

[0109] For the prospective cohort selection criteria, including inclusion criteria and exclusion criteria are the same as those of the retrospective cohort. Among them, transcriptome data: Breast puncture biopsy tissue samples of advanced TNBC patients are collected for RNA quantitative detection and sequencing, etc. Prognostic prospective information: Regular imaging reexaminations and related examinations are performed on patients to evaluate prognostic survival information.

[0110] After the above screening process, patients meeting the criteria are included in the study cohort.

[0111] Step S3: Using the prognostic review information of TNBC patients in the retrospective cohort as the outcome variable, screen out the corresponding pathomics features of the WSI baseline pathology images in the retrospective cohort.

[0112] Specifically, collect the prognostic review information of TNBC patients in the retrospective cohort, including the patients' clinical data, treatment response, and follow-up results. Collect the WSI baseline pathology images corresponding to these patients, and extract pathomics features, including texture features, morphological features, edge gradient features, etc. Using statistical methods, for example, with the prognostic review information as the outcome variable, first perform univariate Cox regression analysis to screen out the pathomics features that may be related to prognosis. Incorporate the significant pathomics features in the univariate analysis and other clinical covariates (such as age, tumor size, lymph node status, etc.) into the multivariate Cox regression model to determine the independent factors affecting prognosis. Based on the results of the multivariate Cox regression analysis, construct a model to determine the independent pathomics features affecting the overall survival and recurrence-free survival of TNBC patients. Evaluate the discrimination and calibration of the model. The discrimination can be evaluated by the C-index and AUROC metrics, while the calibration can be evaluated by the calibration curve. In summary, the Cox proportional hazards regression model can be systematically used to analyze and screen the pathomics features related to the prognosis of TNBC patients, providing a scientific basis for clinical decision-making.

[0113] Step S5: Extract the baseline radiomics features corresponding to the MRI baseline images in the retrospective cohort, and perform a correlation analysis between the baseline radiomics features and the pathomics features to obtain radiomics features with pathological interpretability.

[0114] Specifically, use an automatic, semi-automatic, or manual method to delineate the region of interest (ROI) for the MRI baseline images, and use a radiomics toolbox such as Pyradiomics to extract the baseline radiomics features including shape, texture, signal intensity, etc. from the delineated ROI region. These features include but are not limited to first-order statistical features, texture features (such as GLCM, GLSZM, etc.), signal intensity features, etc. Then preprocess the extracted baseline radiomics features and pathomics features, including normalization to eliminate the differences between different patients. Use statistical methods, such as Pearson or Spearman correlation analysis, to evaluate the correlation between the baseline radiomics features and the pathomics features. Set the threshold of the correlation coefficient. Preferably, the features with a correlation coefficient less than 0.5 can be determined as radiomics features with pathological interpretability.

[0115] Step S7: Based on the early MRI images and baseline MRI images in the retrospective cohort, use radiomics features with pathological interpretability as labels to train a model, and obtain a trained prediction model.

[0116] Specifically, collect the early MRI images and baseline MRI images of TNBC patients from the retrospective cohort. Use the radiomics features extracted from the imaging images and combine them with the corresponding labels, namely radiomics features with pathological interpretability, to construct a training dataset. These labels contain prognostic information, etc. Select a suitable machine learning model, such as random forest, support vector machine, or deep learning model, to train how to predict the efficacy. Use, for example, the ROC curve to evaluate the prediction performance of the model, calculate the AUC value to measure the overall performance of the model, and optimize the model according to the evaluation results, which may include adjusting the model structure, parameter tuning, etc., to improve the prediction performance and generalization ability of the model.

[0117] Step S9: Validate the prediction model based on the multi-temporal MRI images in the prospective cohort to obtain a target prediction model for obtaining the efficacy prediction result.

[0118] Specifically, the prospective cohort refers to collecting data from the beginning of the study and continuously collecting new data over time. Different from the retrospective cohort, the purpose of the prospective cohort is to verify the prediction ability of the model through future data. The data of each patient includes multi-temporal MRI images at multiple time points and the corresponding clinical efficacy information (such as disease progression, treatment response, etc.). Organize these data into a validation set. Input the multi-temporal imaging features extracted from the prospective cohort into the trained prediction model for prediction analysis. These prediction results can be efficacy prediction, disease development prediction, prognosis evaluation, etc. Use different evaluation metrics to evaluate the performance of the model on the prospective cohort validation set. Common evaluation metrics include: Accuracy: The consistency between the prediction result and the actual clinical result. F1 score: The harmonic mean of precision and recall, comprehensively evaluating the model performance. AUC-ROC curve: Evaluating the performance of the model at different decision thresholds, especially suitable for binary classification problems. Time-dependent analysis: Evaluating the stability and prognostic ability of the model at different time points.

[0119] A method for predicting the efficacy of immunotherapy for TNBC patients provided in this embodiment collects sample data on TNBC patients, including a retrospective cohort as a training set and a prospective cohort as a validation set, ensuring that the model can take into account the learning of historical data and the prediction and verification of future data, thereby improving the generalization ability of the model. Using the prognostic review information in the retrospective cohort as the outcome variable, relevant pathological features corresponding to the baseline pathological images are screened out to provide basic data for subsequent analysis. Further extract the radiomics features corresponding to the MRI image in the retrospective cohort and perform correlation analysis to reveal the relationship between the radiomics features and the pathological features, and then obtain radiomics features with pathological interpretability. Using these radiomics features as labels, combined with the early MRI images and baseline MRI images in the retrospective cohort for model training, a trained prediction model is finally obtained. Finally, by using the multi-temporal MRI images in the prospective cohort to verify the prediction model, the target prediction model is obtained, and the prediction accuracy of the model for the prognosis and efficacy of TNBC patients is improved.

[0120] Figure 2 It is a flowchart of step S3 provided in an embodiment of the present application, and this process may include the following steps:

[0121] Step S31, perform nuclear segmentation on the WSI baseline pathological images in the retrospective cohort to obtain the position information of each nucleus.

[0122] Specifically, in order to ensure the accuracy of subsequent analysis, it is necessary to preprocess the WSI baseline pathological images. First, along the HE (hematoxylin-eosin) stained images, the WSI baseline pathological images are segmented into patches of a unified size, and the size of each patch is 512×512 pixels. Then, elastic transformation is applied to each patch for random augmentation to increase the diversity of the images and simulate different tissue deformations. The patch is flipped horizontally or vertically to increase the scale and diversity of the dataset. The patch is rotated at a random angle to further increase the diversity of the dataset. Then, these enhanced patches are input into a deep learning model such as CNN to capture the local features between the image patches. The patches after feature extraction are merged through a fusion technique to reconstruct larger image patches. It should be noted that this fusion can be through simple splicing or more complex fusion strategies, such as using an attention mechanism to weight different patches, so as to reconstruct an image containing more context information. Finally, the fused large image patches are further captured for global features through a multi-layer perceptron (MLP). The MLP can learn the complex relationships between features and extract deeper feature representations through non-linear transformations.

[0123] Next, use CellProfiler software to segment the nuclei in the preprocessed WSI baseline pathology images and obtain the location information of each nucleus, including the position coordinates of each nucleus and the morphological features of the nucleus, such as area, perimeter, major axis and minor axis lengths, roundness, etc. Further, deep learning techniques are used to extract features of different tumor microenvironments, including morphological, texture, and topological information. Morphological feature extraction includes area, perimeter, major axis and minor axis lengths, roundness, etc.; texture feature extraction involves gray-level co-occurrence matrix (GLCM) features, such as contrast, correlation, energy, and entropy, etc.; topological information extraction involves the spatial relationships of objects in the image, including object connectivity, adjacency, and inclusion, etc.

[0124] Step S33: Based on the location information, construct a multi-level pairwise relationship graph; where each relationship graph characterizes the relationship between different types of nuclei.

[0125] Specifically, for the location information of each pair of nuclei of different types, the Euclidean distance between them can be calculated. If the distance between the nuclei is very close, less than a preset distance threshold, it may indicate that there is a certain spatial relationship or biological connection between them, and an edge is drawn between the two. Using the nuclei as nodes, a multi-level pairwise relationship graph is constructed, and each relationship graph represents the relationship between different types of nuclei. For example, construct Graph_T-T (tumor cell - tumor cell), Graph_T-L (tumor cell - lymphocyte), Graph_T-S (tumor cell - stromal cell).

[0126] Step S35: Determine the topological features of each nucleus in each relationship graph and construct the comprehensive topological feature of each nucleus.

[0127] Specifically, for each nucleus, calculate its topological features in each relationship graph respectively, including: calculating the degree of each nucleus node, that is, the number of edges directly connected to the nucleus; calculating the clustering coefficient of each nucleus to measure the tightness of the connection between its neighbor nodes; calculating the betweenness centrality of each nucleus to measure the number of shortest paths passing through the nucleus; calculating the closeness centrality of each nucleus to measure the average distance from the nucleus to all other nodes in the graph; calculating the eigenvector centrality of each nucleus, which takes into account the importance of the nucleus neighbors. The topological features of each nucleus in each relationship graph are obtained, and these topological features are concatenated together in the form of a vector to form a comprehensive topological feature.

[0128] Here, in order to eliminate the influence of the dimension of different features, the comprehensive topological feature is normalized. If the feature vector is very long, dimensionality reduction techniques such as principal component analysis (PCA) can also be used to reduce the feature dimension.

[0129] Step S37: Using the prognostic review information of TNBC patients in the retrospective cohort as the outcome variable, analyze the relationship between the comprehensive topological features and the prognostic review information to obtain the pathomics features corresponding to the WSI baseline pathological images in the retrospective cohort.

[0130] Specifically, the prognostic review information includes overall survival and recurrence-free survival. Use univariate Cox regression analysis to analyze the relationship between the comprehensive topological features and the prognostic review information, and determine which features are significantly correlated with the prognosis. Among them, for the P value obtained from the univariate Cox regression analysis, the P value needs to be corrected using FDR (False Discovery Rate) correction or Bonferroni correction, select the differentially expressed genes with statistical significance, and finally obtain the corrected P value and the pathomics features with significant differences.

[0131] Compared with Figure 1 the embodiment shown, in this embodiment, first, nuclear segmentation is performed on the WSI baseline pathological images in the retrospective cohort to obtain the position information of each nucleus, which can accurately identify the nuclei in the pathological images and provide basic data for subsequent analysis. Then, based on the obtained nuclear position information, a multi-level pairwise relationship graph is constructed, which can depict the mutual connections between nuclei and reveal the association patterns of nuclear types. Next, determine the topological features of each nucleus in the relationship graph and construct the comprehensive topological features of each nucleus. Through topological analysis, the spatial structure features of the nuclei can be comprehensively characterized, providing in-depth data information for further analysis. Finally, using the prognostic review information of TNBC patients in the retrospective cohort as the outcome variable, analyze the relationship between the comprehensive topological features and the prognostic information, and then obtain the pathomics features corresponding to the WSI baseline pathological images in the retrospective cohort, which can establish the association between the pathological features and the patient's prognosis, thus providing more guiding pathomics information for clinical practice.

[0132] Figure 3 The flowchart of step S33 provided by the embodiment of the present application, this process may include the following steps:

[0133] Step S331: For any nucleus, based on the position information, determine the nearest neighbors of the any nucleus corresponding to different types of nuclei.

[0134] Specifically, calculate the Euclidean distance, Manhattan distance, etc. between any two cell nuclei to determine their spatial proximity, obtaining a distance matrix between the cell nuclei. For each cell nucleus, determine its k nearest neighbors according to the distance matrix, that is, the k cell nuclei with the closest distances, obtaining a list of k nearest neighbors for each cell nucleus. Screen out the nearest neighbors belonging to different types of cell nuclei from the list of k nearest neighbors. For example, if the current cell nucleus is a tumor cell T, then screen out the lymphocyte L and stromal cell S among its nearest neighbors, and finally obtain a list of nearest neighbors corresponding to each cell nucleus for different types of cell nuclei.

[0135] Step S333, when the preset edge configuration condition is satisfied, establish an edge between any cell nucleus and its nearest neighbor.

[0136] In some preferred embodiments, the preset edge configuration condition is:

[0137]

[0138] where E is the set of all edges; is the type of cell nucleus V i ; is the type of cell nucleus V j ; D(V i , V j ) is the Euclidean distance between cell nucleus V i and cell nucleus V j ; T is the edge configuration threshold.

[0139] Specifically, for each cell nucleus, check whether its k nearest neighbors satisfy the preset edge configuration condition, and create an edge for the cell nucleus pair (V i , V j ) that meets the condition. It should be noted that the types of the cell nucleus pair are different.

[0140] Step S335, based on the types of cell nuclei and the edges, construct a relationship graph.

[0141] Specifically, construct multiple graphs, each graph representing the relationship between different types of cells, such as Graph_T-L (the relationship graph between tumor cells and lymphocytes) and Graph_T-S (the relationship graph between tumor cells and stromal cells). Each obtained relationship graph shows the spatial relationship between specific two types of cells.

[0142] And Figure 2Compared with the shown embodiments, in this embodiment, first, based on the position information of the cell nuclei, the nearest neighbor relationship between each cell nucleus and different types of cell nuclei is determined, thereby revealing the relative layout of the cell nuclei in space. Then, according to the preset edge configuration conditions, the connection relationship between the cell nuclei is established, which helps to clarify the interaction patterns between different cell nucleus types. Finally, by combining the type information of the cell nuclei and the connection relationship of the edges, a complete relationship graph is constructed. This graph can comprehensively display the spatial associations and their structural relationships between the cell nuclei, providing a basic framework for subsequent topological analysis and the study of cell-cell interactions.

[0143] Figure 4 The topological features provided by the embodiments of the present application include at least one sub-feature, and the flowchart for determining the sub-feature is as follows. The process may include the following steps:

[0144] Step S351, for any target cell nucleus among all cell nuclei, obtain the target relationship graph where the target cell nucleus is located.

[0145] Specifically, select a specific target cell nucleus from all cell nuclei. For this target cell nucleus, obtain the corresponding relationship graph, which contains the positional relationship between the target cell nucleus and its neighboring cell nuclei.

[0146] Step S353, in the target relationship graph, determine the distance information between each neighboring cell nucleus and the target cell nucleus, and integrate the determined distance information into a type-distance sequence according to the types of the neighboring cell nuclei.

[0147] Specifically, in the target relationship graph, calculate the distance between the target cell nucleus and each neighboring cell nucleus, and record this distance information. According to the types of the neighboring cell nuclei, integrate the distance information into a type-distance sequence. For example, if the target cell nucleus is a tumor cell nucleus T1, and in the target relationship graph where the tumor cell nucleus T1 is located, there are neighboring lymphocyte nuclei L1, L2 and stromal cell nuclei S1, S2. Calculate the Euclidean distances between T1 and L1, L2, S1, S2. Suppose the distances are as follows: D(T1,L1)=5 μm, D(T1,L2)=8 μm, D(T1,S1)=3 μm, D(T1,S2)=10 μm, then the type-distance sequence is (L[5,8], S[3,10]).

[0148] Step S355, for any type-distance data in the type-distance sequence, query in the preset standard library whether there is a matching type-distance template.

[0149] Specifically, check whether there is a template in the preset standard library that matches the type-distance sequence. The template defines the upper limit of the number of distances that each type can be associated with and the coverage range of each distance. For example, the template defines that lymphocyte L can be associated with 2 distances, stromal cell S can also be associated with 2 distances, and the coverage range of each distance is 0 - 10 microns.

[0150] Step S357, if available, use the sequence feature corresponding to the matched type-distance template as a sub-feature in the topological feature of the target cell nucleus.

[0151] Specifically, if a matching template is found, use the sequence feature corresponding to the template as a sub-feature of the topological feature of the target cell nucleus. The sequence feature is the quantitative value of the type and the distance. For example, lymphocyte L type may be represented as 001, stromal cell S distance type may be represented as 002, and the quantitative value of the distance is set with the corresponding numerical number according to the interval where the value is located.

[0152] For example, if the distance of 0 - 15 microns is divided into 5 intervals, each interval is 3 microns: 0 - 3 microns: number 1; 4 - 6 microns: number 2; 7 - 9 microns: number 3; 10 - 12 microns: number 4; 13 - 15 microns: number 5. For lymphocyte L, if the distances are 5 microns and 8 microns, according to the above intervals, 5 microns belongs to the second interval (number 2), and 8 microns belongs to the third interval (number 3). For stromal cell S, if the distances are 3 microns and 10 microns, 3 microns belongs to the first interval (number 1), and 10 microns belongs to the fourth interval (number 4). The obtained sub-feature is (001[2,3], 002[1,4]).

[0153] Step S359, if not available, create a type-distance template for the type-distance data in the preset standard library, and use the sequence feature corresponding to the newly created type-distance template as a sub-feature of the topological feature of the target cell nucleus.

[0154] Specifically, if no matching template is found, create a new type-distance template in the standard library, and determine the corresponding sequence feature as a sub-feature of the topological feature of the target cell nucleus. In the newly created template, the number of distances associated with the cell nucleus type should be greater than or equal to the number of distances in the type-distance data, and the coverage range of each distance in the template should include the corresponding distance value in the type-distance data.

[0155] For example, the type-distance sequence is (L[10, 20, 30, 40]). The newly created template will cover at least four distances and divide the distances into the following intervals: 0 - 10 microns: number 1; 11 - 20 microns: number 2; 20 - 30 microns: number 3; 30 - 40 microns: number 4. The obtained sub-feature is 001[1, 2, 3, 4].

[0156] Finally, all sub-features can be concatenated in the order of arrangement in the type-distance sequence to form the overall topological feature of the target cell nucleus.

[0157] Compared with Figure 2 the embodiment shown, in this embodiment, by obtaining the relationship graph where the target cell nucleus is located, it is ensured that the target cell nucleus can be accurately positioned in the overall layout, laying a foundation for subsequent analysis. Then, the distances between the target cell nucleus and adjacent cell nuclei are determined, and the type-distance sequence is integrated according to the types of adjacent cell nuclei, revealing the spatial distribution characteristics of different types of cell nuclei. If a matching type-distance template is found in the preset standard library, the corresponding sequence feature is used as the sub-feature of the target cell nucleus, improving the analysis efficiency and enhancing the accuracy. If no matching template is found, a new template can be created in the standard library and used as part of the topological feature of the target cell nucleus, ensuring the flexibility of the system and expanding the standard library. Overall, it can not only accurately describe the spatial relationship between cell nuclei, but also provide a more comprehensive and accurate analysis for research in fields such as cell biology and pathology during the process of dynamically updating and expanding the standard library.

[0158] Figure 5 This is the flowchart of step S5 provided by the embodiment of the present application. This process may include the following steps:

[0159] Step S51, splice the baseline radiomics features and pathomics features to obtain a fused feature set.

[0160] Step S53, eliminate redundant features in the fused feature set to obtain a target feature set.

[0161] Step S55, based on the target set, perform correlation analysis using the Spearman correlation coefficient to obtain radiomics features with pathological interpretability.

[0162] Specifically, ensure that the baseline radiomics features and pathomics features correspond to the same group of patients. Horizontally concatenate the baseline radiomics features and pathomics features to form a row of feature vectors to obtain a fused feature set. Then, feature selection methods (such as recursive feature elimination RFE, LASSO regression, etc.) can be used to eliminate redundant features and retain the features that contribute more, obtaining a target feature set. Use the Spearman correlation coefficient to analyze the correlation between the features in the target feature set and the prognostic review information. According to the results of the Spearman correlation coefficient, select the radiomics features that have a significant correlation with the prognostic review information, and these features have pathological interpretability.

[0163] Compared with Figure 1 the embodiment shown, in this embodiment, by concatenating the baseline radiomics features and pathomics features, a fused feature set is formed, providing a comprehensive data basis for subsequent analysis. Then, by eliminating the redundant features in the fused feature set, a target feature set is obtained, thereby simplifying the data and improving the efficiency and accuracy of the analysis. Finally, based on the target feature set, the Spearman correlation coefficient is used for correlation analysis to screen out the radiomics features with pathological interpretability, thus providing a valuable basis for further pathological research and clinical diagnosis.

[0164] Figure 6 The flowchart of step S53 provided by the embodiment of the present application may include the following steps:

[0165] Step S531, determine the correlation coefficient of each fused feature in the fused feature set.

[0166] In some preferred embodiments, the correlation coefficient is:

[0167]

[0168] where ρ is the correlation coefficient; x ik is the k-th element in the i-th fused feature vector in the fused feature set; x jk is the k-th element in the j-th fused feature vector in the fused feature set; is the mean of all elements in the i-th fused feature vector in the fused feature set; is the mean of all elements in the j-th fused feature vector in the fused feature set.

[0169] Step S533, determine the fused features with a correlation coefficient greater than the preset threshold as redundant features.

[0170] Step S535, eliminate the redundant features from the fused feature set to obtain a target feature set.

[0171] Specifically, the correlation coefficient is calculated using the formula for the Spearman correlation coefficient. The preset threshold here is preferably 0.9. Identify those features with a correlation coefficient greater than this threshold, mark these features as redundant features, and remove the marked redundant features from the fused feature set. The remaining features constitute the target feature set.

[0172] Compared with Figure 5 the embodiment shown, in this embodiment, by calculating the correlation coefficient of each fused feature, the similarity or correlation between features is evaluated, providing a basis for subsequent screening of redundant features. Then, the fused features with a correlation coefficient greater than the preset threshold are identified as redundant features. These features are repeated or highly similar to other feature information and may have a negative impact on the analysis result. Finally, by removing these redundant features from the fused feature set, the target feature set is obtained, ensuring that the finally retained features are more independent and have high information content, thereby improving the accuracy and efficiency of subsequent analysis.

[0173] Figure 7 The flowchart of step S7 provided by the embodiment of the present application is as follows. This process may include the following steps:

[0174] Step S71, extract the early radiomics features corresponding to the MRI early imaging images in the retrospective cohort, and extract the baseline radiomics features corresponding to the MRI baseline imaging images.

[0175] Specifically, use ITK-SNAP to perform three-dimensional ROI labeling on the tumors in the MRI early imaging images and MRI baseline imaging images, including regions such as intralesional hemorrhage and necrosis. For tumors with multicentric lesions, select the largest tumor as the research object to ensure the consistency and comparability of the research. The ROI delineation is independently performed by two radiologists with breast cancer diagnosis experience to test the reproducibility within observers, that is, whether the segmentation results for the same lesion are consistent among different doctors, which is crucial for the reliability of subsequent feature extraction. Apply PyRadiomics to extract a large number of quantitative radiomics features from medical images, including multi-dimensional features such as shape, texture, grayscale, and volume. These features help to reveal the microstructure and biological characteristics of the tumor. After extracting a large number of features, screen out the features with a consistency greater than 0.75 between observers to ensure the stability and reliability of the selected features. Features with high consistency are more likely to remain consistent under different observers and different scanning conditions, thus having better pathological interpretability.

[0176] Step S73, input the early radiomics features and the baseline radiomics features into the UniFormer model for training, and use the radiomics features with pathological interpretability as labels to train the UniFormer model to obtain a prediction model.

[0177] Specifically, referring to Figure 8 , Figure 8 which is the architecture diagram of UniFormer provided by the embodiment of the present application. The block in Uniformer consists of three parts: DPE: Dynamic Position Encoding; MHRA: Multi-Head Relationship Aggregation; FFN: Feed-Forward Neural Network. The early radiomics features and baseline radiomics features are normalized, and the normalized features are input into the DPE module in the UniFormer model. The DPE module performs lightweight feature extraction through depthwise separable convolution (DWConv) with zero-padding, and zero-padding is added at both ends to retain the key information at the image boundary. This module introduces dynamic position encoding for each token, enabling the model to dynamically adjust the position information according to the specific content of the two MRI images, so as to better capture spatio-temporal features. After the DPE module, the features of the image patches are further non-linearly transformed through a feed-forward network (FFN). The FFN consists of two linear layers and a non-linear activation function (GELU). It enhances the features in each UniFormer block, allowing the model to learn more complex feature representations, thereby improving the model's sensitivity to subtle changes while maintaining computational efficiency. The UniFormer model adopts global and local MHRA strategies to process the temporal information of the two MRI images. Global MHRA is responsible for establishing long-term dependencies in the feature space in the deep layers of the network, and determines the affinity between tokens through similarity comparison of global contexts. Local MHRA, on the other hand, focuses on local neighborhood contexts in the shallow layers of the network, effectively solving the local redundancy problem. In the shallow feature processing, the design of local MHRA enables it to efficiently capture local spatio-temporal structures, complementing the design concept of convolutional operations. The local relationship aggregator defines the affinity through a learnable parameter matrix, ensuring effective aggregation of local neighborhood context information in the shallow layer, thereby enhancing the model's ability to capture local features. The UniFormer model constructs the network by hierarchically stacking UniFormer blocks, which can effectively integrate local and global spatio-temporal information at different levels, enhancing the overall performance of the model.

[0178] The UniFormer model is trained with radiomics features with pathological interpretability as labels, thereby reducing overfitting and better capturing local changes between radiomics features. The output of the model is a probability distribution, representing the probability of efficacy prediction or the probability of prognosis prediction. An appropriate optimizer (such as Adam) and learning rate scheduling strategy are used to train the model, monitor the performance on the validation set, and use early stopping to avoid overfitting. The trained prediction models are an efficacy prediction model and a prognosis prediction model, which are obtained by training based on setting different labels.

[0179] With Figure 1Compared with the embodiments shown, in this embodiment, radiomics features of early MRI images and baseline images in a retrospective cohort are extracted, and these features contain important information reflecting the imaging data. The extracted early and baseline radiomics features are input into the UniFormer model for training. At the same time, radiomics features with pathological interpretability are used as labels to further improve the accuracy of the model and its prediction ability for pathological changes. Through this process, a prediction model based on radiomics features is finally obtained, which can effectively identify and predict the progression and changes of the therapeutic effect.

[0180] Figure 9 The total loss function of the UniFormer model provided by the embodiments of this application includes a decorrelation loss. The flowchart of the acquisition method of the decorrelation loss may include the following steps:

[0181] Step S701, obtain the deep features and radiological features after each iteration and optimization of the UniFormer model.

[0182] Step S703, based on the deep features, radiological features and preset weight parameters, use the exponential weighting method to determine the decorrelation loss.

[0183] In some preferred embodiments, the decorrelation loss is:

[0184]

[0185] where L corr is the decorrelation loss; w is the preset weight parameter, w < 1; N k is the total number of deep features and radiological features; Z i is the i-th deep feature; R i is the i-th radiological feature.

[0186] Specifically, deep features refer to high-dimensional features that can capture complex patterns and structural information in data. In medical image analysis, deep features can include: texture information, such as the texture features of the tumor region, which may be related to the invasiveness or grading of the tumor; shape information, such as the edge shape of the tumor, which may be related to the malignancy of the tumor; location information, the location of the tumor in the image may provide clues for diagnosis; context information, the structure and state of the surrounding tissues, which may be related to the biological behavior of the tumor. Radiomic features, also known as imaging biomarker features, are high-throughput features extracted from medical images and can include: shape features, such as the volume, surface area, shape index, etc. of the tumor; texture features, such as the signal intensity distribution inside the tumor, which may reflect the heterogeneity of the tissue; signal intensity features, such as the signal intensity of the tumor in different imaging sequences; signal texture features, such as the statistical characteristics of the signals inside the tumor, such as mean, variance, skewness, kurtosis, etc.; signal attenuation features, in CT scans, the degree of attenuation of the tumor to X-rays; kinetic features, in dynamic contrast-enhanced scans, the absorption rate and pattern of the tumor to the contrast agent.

[0187] Radiomic features with pathological interpretability are used as prior information to enhance the feature representation ability of the UniFormer model, reduce overfitting, and better capture local variations between features. In addition, by introducing a decorrelation loss, it is ensured that UniFormer features and radiomic features can complement each other's information. The library storing UniFormer features and radiomic features adopts a first-in-first-out (FIFO) strategy and saves at most N k features. In the initial stage, the model warms up by learning the data in the feature library to stabilize the learning process. The exponential weighted average method is used to calculate the correlation between samples, and the weight parameter w satisfies w < 1 to emphasize the correlation of the most recent samples.

[0188] In the calculation of the decorrelation loss, the exponential weighting method is used to emphasize the correlation between the most recent feature pairs. Through the above formula, the weighted correlation between each pair of features can be calculated, and the overall decorrelation loss is measured by the sum of absolute values, which helps to reduce redundant information between features and improve the generalization ability of the model. Through this method, the UniFormer model can optimize deep features and radiomic features in each iteration, reduce the correlation between features, and thus improve the prediction performance and stability of the model.

[0189] Compared with Figure 8Compared with the embodiments shown, in this embodiment, after each iterative optimization of the model, deep features and radiological features are obtained. These features respectively reflect the high-level semantic information and radiological features of the image, which helps the model to comprehensively understand the imaging data. By combining these deep features and radiological features, and according to the preset weight parameters, an exponential weighting method is used to determine the decorrelation loss, thereby effectively reducing the redundant information between features and improving the independence of features. Through this processing, the model can more accurately capture the key information of the image, improve the prediction effect, and enhance the robustness of the model.

[0190] In some preferred embodiments, it further includes evaluating the prediction model:

[0191] The prediction model is evaluated by using the ROC curve, decision curve or calibration curve.

[0192] Specifically, taking the recurrence-free survival period in the retrospective cohort as the outcome variable of the prognostic review information, the tumor size at different time points after treatment is recorded. And the iRECIST standard is applied for efficacy evaluation according to the results of each examination. By plotting the ROC curve, the prediction performance of the constructed model is evaluated, and the AUC value is calculated to measure its overall performance. For different thresholds, the true positive rate (TPR) and false positive rate (FPR) of the model are calculated. The closer the AUC value is to 1, the stronger the discrimination ability of the model. The decision curve evaluates the practicality of the model by comparing the net benefit of the model prediction and no prediction. The calibration curve is used to evaluate the consistency between the predicted probability of the model and the actual occurrence probability, that is, the calibration degree of the model.

[0193] Figure 10 It is a flowchart of another method for predicting the efficacy of immunotherapy for TNBC patients provided by the embodiments of the present application. This process may include the following steps:

[0194] Step S101, based on the prognostic forward information of TNBC patients in the prospective cohort, divide the TNBC patients into treatment responders and treatment non-responders.

[0195] Specifically, collect the prognostic forward information of TNBC patients in the prospective cohort, evaluate the change in tumor size by comparing the multi-temporal imaging genomics images of MRI before and after treatment, judge the treatment response of the patients, and divide the patients into treatment responders (effective treatment, tumor shrinkage or stability) and treatment non-responders (ineffective treatment, tumor progression or enlargement).

[0196] Step S103, based on the transcriptome data in the prospective cohort, perform differential expression analysis on treatment responders and treatment non-responders to determine differentially expressed genes.

[0197] Specifically, preprocessing steps such as quality control, removal of low-expression genes, and normalization are performed on the transcriptome data of treatment responders and non-responders. Tools such as DESeq2 or edgeR are used to perform differential expression analysis on the two groups of samples to determine genes with significantly different expression levels. The P-values are corrected, such as using FDR (False Discovery Rate) correction, and differentially expressed genes with statistical significance are selected.

[0198] Step S105, perform enrichment analysis based on the differentially expressed genes to obtain the biological pathways of the differentially expressed genes.

[0199] Specifically, enrichment analysis is performed on the differentially expressed genes, such as GO analysis, KEGG pathway analysis, etc., to understand the biological functions and pathway participation of these genes. The analysis results are visually displayed through graphs such as volcano plots and heatmaps.

[0200] Step S107, extract the prospective radiomics features corresponding to the multi-temporal MRI images in the prospective cohort.

[0201] Specifically, the method for extracting the prospective radiomics features corresponding to the multi-temporal MRI images in the prospective cohort here is the same as that in step S71, which will not be elaborated here.

[0202] Step S109, determine the interpretability analysis results of the prospective radiomics features and genes based on the prediction model and the pathway.

[0203] Specifically, the prospective radiomics features are input into the prediction model, and the follow-up information of the patients is recorded. Through Pearson or Spearman correlation analysis, the relationship between the prospective radiomics features and the differentially expressed genes is further determined, further improving the interpretability of the images and helping to understand how gene expression affects the imaging features of tumors. Among them, a correlation coefficient less than 0.5 obtained in the correlation analysis indicates that the pathway is significantly correlated with the prospective radiomics features.

[0204] And Figure 1Compared with the embodiments shown, in this embodiment, according to the prognostic information in the prospective cohort, TNBC patients are divided into treatment responders and non-responders, providing a clear grouping basis for subsequent analysis. Then, differential expression analysis is performed on these two groups of patients based on transcriptome data to identify key genes affecting treatment response. Enrichment analysis is performed on these differentially expressed genes to reveal the biological pathways they may be involved in. By extracting radiomics features from multi-temporal MRI images, the dimension of clinical information is further expanded, laying a foundation for subsequent multi-modal analysis. Finally, correlation analysis is performed between radiomics features and gene expression data and their related pathways to obtain interpretable analysis results between imaging features and genes. This analysis not only helps to discover potential biomarkers for treatment response but also improves the predictive ability of radiomics features in tumor treatment.

[0205] The following combines Figure 11 to describe the specific implementation of the present invention. Refer to Figure 11 ,

[0206] Step S1001, collect retrospective cohort data and prospective cohort data.

[0207] Step S1002, after preprocessing the MRI baseline image and the WSI baseline pathology image, perform feature extraction respectively, and use Pearson / Spearman correlation analysis to obtain radiomics features with pathological interpretability; generate a visualization heatmap to clearly display the radiomics features with pathological interpretability.

[0208] Step S1003, preprocess the MRI baseline image and the MRI early image and input them into UniFormer, and use the radiomics features with pathological interpretability as labels for supervised learning to obtain a prediction model, where the prediction model includes a treatment effect prediction model and a prognosis prediction model; a prospective cohort can be used to verify and optimize the model, which can capture dynamic adjustment positions according to MRI time-series data, thereby improving the accuracy of the model.

[0209] Step S1004, input the MRI multi-temporal image into the prediction model for feature extraction and prediction, and perform differential expression analysis on the corresponding transcriptome data, and perform enrichment analysis on the obtained differentially expressed genes to obtain pathways.

[0210] Step S1005, analyze the differentially expressed genes and imaging features to obtain biologically interpretable analysis results at the molecular level, and the analysis results can be intuitively displayed through graphs such as volcano plots and heatmaps.

[0211] Correspondingly, please refer to Figure 12A block diagram of an immune therapy efficacy prediction system for TNBC patients provided by an embodiment of the present application. The system includes:

[0212] A data acquisition unit S111, configured to acquire sample data about TNBC patients; wherein the sample data includes a retrospective cohort as a training set and a prospective cohort as a validation set;

[0213] A pathological omics feature acquisition unit S113, configured to use the prognostic review information of TNBC patients in the retrospective cohort as an outcome variable, and screen out the pathological omics features corresponding to the WSI baseline pathological images in the retrospective cohort;

[0214] An interpretable feature acquisition unit S115, configured to extract the baseline imaging omics features corresponding to the MRI baseline imaging images in the retrospective cohort, and perform a correlation analysis on the baseline imaging omics features and the pathological omics features to obtain imaging omics features with pathological interpretability;

[0215] A prediction model acquisition unit S117, configured to train a model based on the MRI early imaging images and the MRI baseline imaging images in the retrospective cohort, using the imaging omics features with pathological interpretability as labels, to obtain a trained prediction model;

[0216] A model validation unit S119, configured to validate the prediction model based on the MRI multi-temporal imaging images in the prospective cohort to obtain a target prediction model, so as to obtain an efficacy prediction result.

[0217] In some preferred embodiments, the pathological omics feature acquisition unit S113 includes:

[0218] Perform nuclear segmentation on the WSI baseline pathological images in the retrospective cohort to obtain the position information of each nucleus;

[0219] Based on the position information, construct a multi-level pairwise relationship graph; wherein each relationship graph characterizes the relationship between different nuclear types;

[0220] Determine the topological features of each nucleus in each relationship graph, and construct the comprehensive topological features of each nucleus;

[0221] Use the prognostic review information of TNBC patients in the retrospective cohort as an outcome variable, analyze the relationship between the comprehensive topological features and the prognostic review information, and obtain the pathological omics features corresponding to the WSI baseline pathological images in the retrospective cohort.

[0222] In some preferred embodiments, constructing a multi-level pairwise relationship graph based on the position information includes:

[0223] For any nucleus, based on the position information, determine the nearest neighbors of the any nucleus corresponding to different types of nuclei;

[0224] When the preset edge configuration condition is satisfied, establish any cell nucleus and its nearest neighbor edge.

[0225] Construct a relationship graph based on the type of the cell nucleus and the edge.

[0226] In some preferred embodiments, the preset edge configuration condition is:

[0227]

[0228] where E is the set of all edges; is the type of cell nucleus V i ; is the type of cell nucleus V j ; D(V i , V j ) is the Euclidean distance between cell nucleus V i and cell nucleus V j ; T is the edge configuration threshold.

[0229] In some preferred embodiments, the topological features include at least one sub-feature, and the sub-feature is determined in the following manner:

[0230] For any target cell nucleus in each cell nucleus, obtain the target relationship graph where the target cell nucleus is located;

[0231] In the target relationship graph, determine the distance information between each neighboring cell nucleus and the target cell nucleus, and integrate the determined distance information into a type-distance sequence according to the type of each neighboring cell nucleus;

[0232] For any type-distance data in the type-distance sequence, query in the preset standard library whether there is a matching type-distance template;

[0233] If there is a match, use the sequence feature corresponding to the matching type-distance template as a sub-feature in the topological features of the target cell nucleus;

[0234] If there is no match, create a type-distance template for the type-distance data in the preset standard library, and use the sequence feature corresponding to the newly created type-distance template as a sub-feature of the topological features of the target cell nucleus.

[0235] In some preferred embodiments, perform a correlation analysis on the baseline radiomics features and the pathomics features to obtain radiomics features with pathological interpretability, including:

[0236] Concatenate the baseline radiomics features and the pathomics features to obtain a fused feature set;

[0237] Eliminate redundant features in the fused feature set to obtain the target feature set;

[0238] Based on the target set, perform correlation analysis using the Spearman correlation coefficient to obtain radiomics features with pathological interpretability.

[0239] In some preferred embodiments, eliminating redundant features in the fused feature set to obtain the target feature set includes:

[0240] Determine the correlation coefficient of each fused feature in the fused feature set;

[0241] Determine the fused features with correlation coefficients greater than the preset threshold as redundant features;

[0242] Eliminate redundant features from the fused feature set to obtain the target feature set.

[0243] In some preferred embodiments, the prediction model acquisition unit S117 includes:

[0244] Extract the early radiomics features corresponding to the MRI early imaging images in the retrospective cohort, and extract the baseline radiomics features corresponding to the MRI baseline imaging images;

[0245] Input the early radiomics features and the baseline radiomics features into the UniFormer model for training, and use the radiomics features with pathological interpretability as labels to train the UniFormer model to obtain the prediction model.

[0246] In some preferred embodiments, the total loss function of the UniFormer model includes a decorrelation loss, and the acquisition method of the decorrelation loss is as follows:

[0247] Obtain the deep features and radiological features after each iteration optimization of the UniFormer model;

[0248] Based on the deep features, radiological features, and preset weight parameters, use the exponential weighting method to determine the decorrelation loss.

[0249] In some preferred embodiments, the decorrelation loss is:

[0250]

[0251] Among them, L corr is the decorrelation loss; w is the preset weight parameter, w < 1; N k is the total number of deep features and radiological features; Z i is the i-th deep feature; R i is the i-th radiological feature.

[0252] In some preferred embodiments, the method further includes evaluating the prediction model:

[0253] The prediction model is evaluated using an ROC curve, a decision curve, or a calibration curve.

[0254] In some preferred embodiments, the method further includes: analyzing the transcriptomic data in the prospective cohort and the prospective radiomics features corresponding to the multi-temporal MRI images in the prospective cohort to obtain the interpretability analysis results of the radiomics features and genes, including:

[0255] Based on the prognostic prospective information of TNBC patients in the prospective cohort, the TNBC patients are divided into treatment responders and treatment non-responders;

[0256] Based on the transcriptomic data in the prospective cohort, differential expression analysis is performed on treatment responders and treatment non-responders to determine differentially expressed genes;

[0257] Based on the differentially expressed genes, enrichment analysis is performed to obtain the biological pathways of the differentially expressed genes;

[0258] Extract the prospective radiomics features corresponding to the multi-temporal MRI images in the prospective cohort;

[0259] Based on the prediction model and the pathway, determine the interpretability analysis results of the prospective radiomics features and genes.

[0260] The further function descriptions of the above-mentioned various modules and units are the same as those in the corresponding embodiments above, and will not be elaborated here.

[0261] A TNBC patient immunotherapy efficacy prediction system in this embodiment is presented in the form of functional units. Here, the unit refers to an ASIC (Application Specific Integrated Circuit) circuit, a processor and a memory that execute one or more software or fixed programs, and / or other devices that can provide the above functions.

[0262] Please refer to Figure 13 , Figure 13 which is a schematic structural diagram of a computer device provided by an embodiment of the present application. As Figure 13As shown, the computer device includes: one or more processors 10, a memory 20, and interfaces for connecting the various components, including a high-speed interface and a low-speed interface. The various components communicate with each other using different buses and can be installed on a common motherboard or in other ways as needed. The processor can process instructions executed within the computer device, including instructions stored in the memory or on the memory to display graphical information of a GUI on an external input / output device (such as a display device coupled to the interface). In some alternative embodiments, multiple processors and / or multiple buses can be used together with multiple memories and multiple memories if needed. Similarly, multiple computer devices can be connected, and each device provides some necessary operations (such as a server array, a set of blade servers, or a multi-processor system). Figure 13 In the figure, one processor 10 is taken as an example.

[0263] The processor 10 can be a central processing unit, a network processor, or a combination thereof. Among them, the processor 10 can further include a hardware chip. The above hardware chip can be an application-specific integrated circuit, a programmable logic device, or a combination thereof. The above programmable logic device can be a complex programmable logic device, a field-programmable gate array, a generic array logic, or any combination thereof.

[0264] Among them, the memory 20 stores instructions executable by at least one processor 10, so that the at least one processor 10 executes the method shown in the above embodiments.

[0265] The memory 20 can include a program storage area and a data storage area. Among them, the program storage area can store an operating system and application programs required for at least one function; the data storage area can store data created according to the use of the computer device and the like. In addition, the memory 20 can include a high-speed random access memory and can also include a non-transitory memory, such as at least one disk storage device, a flash memory device, or other non-transitory solid-state storage devices. In some alternative embodiments, the memory 20 can optionally include a memory remotely set relative to the processor 10, and these remote memories can be connected to the computer device through a network. Examples of the above network include but are not limited to the Internet, an enterprise intranet, a local area network, a mobile communication network, and combinations thereof.

[0266] The memory 20 can include a volatile memory, such as a random access memory; the memory can also include a non-volatile memory, such as a flash memory, a hard disk, or a solid-state drive; the memory 20 can also include a combination of the above types of memories.

[0267] The computer device further includes a communication interface 30 for the computer device to communicate with other devices or communication networks.

[0268] Embodiments of the present application also provide a computer-readable storage medium. The methods according to the embodiments of the present application can be implemented in hardware, firmware, or be implemented as computer code that can be recorded on a storage medium, or be implemented as computer code originally stored in a remote storage medium or a non-transitory machine-readable storage medium and to be downloaded through a network and stored in a local storage medium, so that the methods described herein can be stored as such software processing on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. Among them, the storage medium can be a magnetic disk, an optical disk, a read-only memory, a random access memory, a flash memory, a hard disk, or a solid-state drive, etc.; further, the storage medium can also include a combination of the above-mentioned types of memories. It can be understood that a computer, a processor, a microprocessor controller, or programmable hardware includes a storage component that can store or receive software or computer code. When the software or computer code is accessed and executed by the computer, the processor, or the hardware, the methods shown in the above embodiments are implemented.

[0269] The systems and units illustrated in the above embodiments can be specifically implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer. Specifically, the computer can be, for example, a personal computer, a laptop computer, a cellular phone, a camera phone, a smart phone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or a combination of any of these devices.

[0270] For the convenience of description, when describing the above devices, they are described separately as various units according to their functions. Of course, when implementing the present application, the functions of each unit can be implemented in the same or multiple software and / or hardware.

[0271] Those skilled in the art should understand that the embodiments of the present application can be provided as methods and systems. Therefore, the present application can adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0272] The present application is described with reference to the flowcharts and / or block diagrams of methods and systems according to the embodiments of the present application. It should be understood that each process and / or block in the flowchart and / or block diagram, and the combination of processes and / or blocks in the flowchart and / or block diagram can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a dedicated computer, an embedded processor, or other programmable data processing devices to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing devices generate for implementing the processFigure 1 one or more processes and / or blocks Figure 1 a device for the functions specified in one or more blocks

[0273] These computer program instructions can 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, such that the instructions stored in the computer-readable memory produce a manufactured article including an instruction device that implements the functions in the process Figure 1 one or more processes and / or blocks Figure 1 specified in one or more blocks

[0274] These computer program instructions can also be loaded onto a computer or other programmable data processing device, such that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, and thus the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in the process Figure 1 one or more processes and / or blocks Figure 1 specified in one or more blocks

[0275] It should also be noted that the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, such that a process, method, commodity or device including a series of elements not only includes those elements, but also includes other elements not expressly listed, or elements inherent to such process, method, commodity or device. Without further limitation, an element defined by the statement "including one..." does not exclude the presence of additional identical elements in the process, method, commodity or device including the said element

[0276] Each embodiment in this specification is described in a progressive manner. The same or similar parts among the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for system embodiments, since they are basically similar to method embodiments, the description is relatively simple, and the relevant parts can be referred to the description of the method embodiments

[0277] The above description is only for the embodiments of the present application and is not used to limit the present application. For those skilled in the art, the present application can have various changes and modifications. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included within the scope of the claims of the present application

[0278] Although the embodiments of the present application are described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of the present application, and such modifications and variations fall within the scope defined by the appended claims

Claims

1. A method for predicting the efficacy of immunotherapy in TNBC patients, characterized in that, The method includes: Obtaining sample data regarding TNBC patients; wherein, the sample data includes a retrospective cohort as a training set and a prospective cohort as a validation set; Using the prognostic retrospective information of the TNBC patients in the retrospective cohort as an outcome variable, screening to obtain the pathomics features corresponding to the WSI baseline pathology images in the retrospective cohort; Wherein, using the prognostic retrospective information of the TNBC patients in the retrospective cohort as an outcome variable and screening to obtain the pathomics features corresponding to the WSI baseline pathology images in the retrospective cohort includes: Performing nucleus segmentation on the WSI baseline pathology images in the retrospective cohort to obtain the position information of each nucleus; Based on the position information, constructing a multi-level pairwise relationship graph; wherein, each relationship graph characterizes the relationship between different nucleus types; Determining the topological features of each nucleus in each relationship graph and constructing the comprehensive topological features of each nucleus; Using the prognostic retrospective information of the TNBC patients in the retrospective cohort as an outcome variable, analyzing the relationship between the comprehensive topological features and the prognostic retrospective information to obtain the pathomics features corresponding to the WSI baseline pathology images in the retrospective cohort; Extracting the baseline radiomics features corresponding to the MRI baseline imaging images in the retrospective cohort, and performing correlation analysis on the baseline radiomics features and the pathomics features to obtain radiomics features with pathological interpretability; Based on the MRI early imaging images and the MRI baseline imaging images in the retrospective cohort, using the radiomics features with pathological interpretability as labels to train a model to obtain a trained prediction model; Validating the prediction model based on the MRI multi-temporal imaging images in the prospective cohort to obtain a target prediction model for obtaining a curative effect prediction result; Analyzing the transcriptome data in the prospective cohort and the prospective radiomics features corresponding to the MRI multi-temporal imaging images in the prospective cohort to obtain an interpretable analysis result of the radiomics features and genes, including: Based on the prognostic prospective information of the TNBC patients in the prospective cohort, classifying the TNBC patients into treatment responders and treatment non-responders; Based on the transcriptome data in the prospective cohort, performing differential expression analysis on the treatment responders and the treatment non-responders to determine differentially expressed genes; Performing enrichment analysis based on the differentially expressed genes to obtain the biological pathways of the differentially expressed genes; Extracting the prospective radiomics features corresponding to the MRI multi-temporal imaging images in the prospective cohort; Based on the prediction model and the pathways, determining the interpretable analysis result of the prospective radiomics features and genes.

2. The method according to claim 1, wherein The constructing a multi-level pairwise relationship graph based on the position information includes: For any nucleus, based on the position information, determining the nearest neighbors of the any nucleus corresponding to different types of nuclei; Establishing an edge between the any nucleus and the nearest neighbors when meeting a preset edge configuration condition. The relationship graph is constructed based on the type of the cell nucleus and the edge.

3. The method according to claim 1, wherein At least one sub - feature is included in the topological feature, and the sub - feature is determined in the following manner: For any target cell nucleus among all the cell nuclei, obtain the target relationship graph where the target cell nucleus is located; In the target relationship graph, determine the distance information between each neighboring cell nucleus and the target cell nucleus, and integrate the determined distance information into a type - distance sequence according to the type of each neighboring cell nucleus; For any type - distance data in the type - distance sequence, query whether there is a matching type - distance template in the preset standard library; If there is a match, use the sequence feature corresponding to the matching type - distance template as a sub - feature in the topological feature of the target cell nucleus; If there is no match, create a type - distance template for the type - distance data in the preset standard library, and use the sequence feature corresponding to the newly created type - distance template as a sub - feature in the topological feature of the target cell nucleus.

4. The method according to claim 1, wherein The correlation analysis of the baseline radiomics features and the pathomics features to obtain radiomics features with pathological interpretability includes: Concatenate the baseline radiomics features and the pathomics features to obtain a fused feature set; Eliminate the redundant features in the fused feature set to obtain a target feature set; Based on the target set, perform correlation analysis using the Spearman correlation coefficient to obtain the radiomics features with pathological interpretability.

5. The method according to claim 4, characterized in that, The elimination of the redundant features in the fused feature set to obtain a target feature set includes: Determine the correlation coefficient of each fused feature in the fused feature set; Determine the fused features with correlation coefficients greater than the preset threshold as the redundant features; Eliminate the redundant features from the fused feature set to obtain the target feature set.

6. The method according to claim 1, characterized in that, Based on the MRI early - stage image and the MRI baseline image in the retrospective cohort, using the radiomics features with pathological interpretability as labels to train a model to obtain a trained prediction model, includes: Extract the early - stage radiomics features corresponding to the MRI early - stage image in the retrospective cohort, and extract the baseline radiomics features corresponding to the MRI baseline image; Input the early - stage radiomics features and the baseline radiomics features into the UniFormer model for training, and use the radiomics features with pathological interpretability as labels to train the UniFormer model to obtain the prediction model.

7. The method according to claim 6, wherein The total loss function of the UniFormer model includes a decorrelation loss, and the acquisition method of the decorrelation loss is as follows: Obtain the deep features and radiological features after each iteration and optimization of the UniFormer model; Based on the deep features, the radiological features and the preset weight parameters, use the exponential weighting method to determine the decorrelation loss.

8. The method according to claim 7, wherein The decorrelation loss is: Among them, L corr is the decorrelation loss; w is a preset weight parameter, w < 1; N k is the total number of depth features and radiological features; Z i is the i-th depth feature; R i is the i-th radiological feature.

9. The method according to claim 1, characterized in that, The method further includes evaluating the prediction model: The prediction model is evaluated using an ROC curve, a decision curve, or a calibration curve.

10. A prediction system for predicting the efficacy of immunotherapy for TNBC patients according to any one of claims 1-9, characterized in that, The system includes: a data acquisition unit configured to acquire sample data regarding TNBC patients; wherein the sample data includes a retrospective cohort as a training set and a prospective cohort as a validation set; a pathomics feature acquisition unit configured to screen out pathomics features corresponding to the WSI baseline pathological images in the retrospective cohort, with the prognostic review information of the TNBC patients in the retrospective cohort as the outcome variable; an interpretable feature acquisition unit configured to extract baseline radiomics features corresponding to the MRI baseline imaging images in the retrospective cohort, and perform a correlation analysis on the baseline radiomics features and the pathomics features to obtain radiomics features with pathological interpretability; a prediction model acquisition unit configured to train a model using the radiomics features with pathological interpretability as labels based on the MRI early imaging images and the MRI baseline imaging images in the retrospective cohort to obtain a trained prediction model; a model validation unit configured to validate the prediction model based on the MRI multi-temporal imaging images in the prospective cohort to obtain a target prediction model for obtaining a curative effect prediction result.

11. A computer device, characterized in that, including: a memory and a processor, which are communicatively connected to each other, wherein the memory stores computer instructions, and the processor executes the computer instructions to execute the TNBC patient immunotherapy curative effect prediction method according to any one of claims 1 to 9.

12. A computer-readable storage medium, characterized in that, Computer instructions are stored on the computer-readable storage medium, and the computer instructions are used to cause a computer to execute the TNBC patient immunotherapy curative effect prediction method according to any one of claims 1 to 9.

Images