Prediction system and prediction method for lymphatic accumulation state of breast cancer

Abstract

The invention belongs to the technical field of biotechnology, and particularly relates to a breast cancer lymphatic accumulation state prediction system and method, and the system comprises a database which is used for storing related data of a breast cancer patient; the pathological omics image processing module and the iconomics image processing module are respectively used for obtaining pathological omics depth features and iconomics features; the pathological omics score, radiomics score and immunity calculation module is used for calculating a pathological omics score, a radiomics score and an immunity score respectively; the regression analysis module is used for carrying out statistical analysis on all scores; the model construction module is used for constructing a combined column graph model; and the output module is used for displaying and outputting a prediction result of the combined column graph model. According to the prediction system and the prediction method, the use of traditional invasive operations such as armpit lymph node cleaning and sentinel lymph node biopsy is avoided, the pain and complication risk of a patient are reduced, and a non-invasive and safe evaluation means is provided for clinic.

Description

Technical Field

[0001] This application belongs to the field of biotechnology, and specifically relates to a breast cancer lymph involvement status prediction system and a prediction method. Background Art

[0002] Breast cancer is one of the most common malignant tumors in women. Its lymph involvement status, especially axillary lymph node metastasis, is an important factor in evaluating the clinical stage of patients, formulating personalized treatment plans, and assessing prognosis. Traditional assessment of lymph nodes relies on pathological examinations such as axillary lymph node dissection (ALND) and sentinel lymph node biopsy (SLNB), but these methods are invasive and may cause additional pain and complications to patients.

[0003] Currently, with the rapid development of medical imaging technology and high-throughput sequencing technology, emerging fields such as pathomics, radiomics, and immuno-oncology have provided new ideas for the precise diagnosis and treatment of breast cancer. However, single-omics methods often have problems such as incomplete information and limited prediction accuracy. Summary of the Invention

[0004] To solve at least one technical problem in the prior art, this application provides a breast cancer lymph involvement status prediction system and a prediction method.

[0005] In a first aspect, this application discloses a breast cancer lymph involvement status prediction system, including: A database for storing FFPE section image data, MRI image data, and gene expression data of breast cancer patients; A pathomics image processing module for extracting the FFPE section image data of patients in the database, preprocessing and feature extracting the FFPE section image data to obtain the deep pathomics features of patients; A radiomics image processing module for extracting the MRI image data of patients in the database, preprocessing and feature extracting the MRI image data to obtain the radiomics features of patients; A pathomics score calculation module for calculating the pathomics score of patients based on the deep pathomics features; A radiomics score calculation module for calculating the radiomics score of patients based on the radiomics features; An immune score calculation module for extracting the gene expression data of patients in the database and calculating the immune score of patients based on the gene expression data; A regression analysis module for statistically analyzing the pathomics score, radiomics score, and immune score of patients; A model construction module, configured to construct a combined nomogram model integrating comprehensive pathology, radiomics, and immune scores based on the statistical analysis results of the regression analysis module, where the combined nomogram model is used to predict the lymph node involvement status of breast cancer patients; An output module, configured to display and output the prediction results of the combined nomogram model.

[0006] Optionally, the pathology image processing module includes: A first extraction unit, configured to extract the FFPE section image data from the database; A first preprocessing unit, configured to perform preprocessing on the image including color correction, blank image deletion, and format conversion using a deep learning framework, extract the region of interest in the FFPE section image data, and remove holes; A first feature extraction unit, configured to divide the FFPE section image data processed by the first preprocessing unit into multiple small blocks, and use a pre-trained deep learning model to extract the deep learning features of each small block; A feature processing unit, configured to integrate and average the deep learning features of all small blocks divided from multiple sections of the same patient to obtain the pathology deep features of the patient.

[0007] Optionally, the pathology score calculation module applies the Lasso regression algorithm to screen the pathology deep features with non-zero regression coefficients, and calculates the pathology score of the patient using the screened features and corresponding weights.

[0008] Optionally, the radiomics image processing module includes: A second extraction unit, configured to extract the MRI image data from the database; A second preprocessing unit, configured to perform format conversion, bias field correction, normalization, and resampling preprocessing on the MRI image data; A second feature extraction unit, configured to extract radiomics features of the region of interest from the image data processed by the second preprocessing unit, where the radiomics features include but are not limited to first-order statistical features, two-dimensional and three-dimensional shape features, and gray-scale features.

[0009] Optionally, the radiomics score calculation module applies the Lasso regression algorithm to screen the radiomics features with non-zero regression coefficients, and calculates the radiomics score of each patient using the screened features and corresponding weights.

[0010] Optionally, the immune score calculation module estimates the stromal cell and immune cell content in tumor samples based on the gene expression data of patients using the ESTIMATE algorithm, and calculates the immune score, which includes four scores: stromal score, immune score, ESTIMATE score, and tumor purity.

[0011] Optionally, the breast cancer lymph node involvement status prediction system further includes: A normalization unit for performing Z-score normalization processing on the pathological omics score, radiomics score, and immune score before the regression analysis module.

[0012] In a second aspect, the present application also discloses a method for predicting the lymph node involvement status of breast cancer, including the following steps: Step 1: Preprocess and extract features from the FFPE section image data of patients in the database to obtain the deep pathological omics features of the patients; Step 2: Preprocess and extract features from the MRI image data of patients in the database to obtain the radiomics features of the patients; Step 3: Calculate the pathological omics score of the patient based on the deep pathological omics features, calculate the radiomics score of the patient based on the radiomics features, and calculate the immune score of the patient based on the gene expression data in the database; Step 4: Perform statistical analysis on the pathological omics score, radiomics score, and immune score of the patient; Step 5: Construct a combined nomogram model integrating pathological omics, radiomics, and immune scores based on the statistical analysis results of Step 4; Step 6: Use the combined nomogram model to predict the lymph node involvement status of breast cancer patients, and display and output the prediction results.

[0013] Optionally, in Step 1, the preprocessing and feature extraction of the FFPE section image data of patients in the database specifically include the following steps: Step 1.1: Extract the FFPE section image data from the database; Step 1.2: Use a deep learning framework to preprocess the image, including color correction, blank image deletion, and format conversion, and extract the region of interest in the FFPE section image data and remove holes; Step 1.3: Cut the FFPE section image data processed in Step 1.2 into multiple small pieces, and use a pre-trained deep learning model to extract the deep learning features of each small piece; Step 1.4: Integrate and average the deep learning features of all small pieces cut from multiple sections of the same patient to obtain the deep pathological omics features of the patient; Further, in the third step, calculating the pathological omics score of the patient according to the deep pathological omics features includes: Applying the Lasso regression algorithm to screen the deep pathological omics features with non-zero regression coefficients, and calculating the pathological omics score of the patient by using the screened features and corresponding weights.

[0014] Optionally, in the second step, the preprocessing and feature extraction of the MRI image data of the patients in the database specifically include the following steps: Step 2.1: Extract the MRI image data from the database; Step 2.2: Perform preprocessing including format conversion, bias field correction, normalization, and resampling on the MRI image data; Step 2.3: Extract the radiomics features of the region of interest from the image data processed in Step 2.2; Further, in the third step, calculating the radiomics score of the patient according to the radiomics features includes: Applying the Lasso regression algorithm to screen the radiomics features with non-zero regression coefficients, and calculating the radiomics score of each patient by using the screened features and corresponding weights.

[0015] The present application has at least the following beneficial technical effects: 1) Non-invasive assessment method: The present application proposes a novel combined nomogram model that combines pathological omics, radiomics, and immune scores for predicting lymph node involvement in breast cancer; this system and method avoid the use of traditional invasive surgeries such as axillary lymph node dissection (ALND) and sentinel lymph node biopsy (SLNB), reducing the pain and complication risks of patients, and providing a non-invasive and safe assessment means for clinical practice.

[0016] 2) Multi-omics data fusion: By integrating multi-source data of pathological omics, radiomics, and immune scores, the present application can more comprehensively evaluate the pathological status and tumor microenvironment of breast cancer patients, improving the accuracy and reliability of predicting lymph node involvement; this method of multi-omics combined analysis provides richer information support for personalized treatment of breast cancer.

[0017] 3) Application of deep learning and machine learning algorithms: Using deep learning frameworks (such as ResNet50) and machine learning algorithms (such as Lasso regression) to extract and screen features from pathological images and image data, effectively extracting high-level abstract features, which have high sensitivity and specificity for predicting lymph node involvement; this automated feature extraction and screening process greatly improves the efficiency and accuracy of data analysis.

[0018] 4) Intuitiveness and practicality of the nomogram model: The constructed nomogram model can intuitively display the comprehensive impact of pathological, imaging, and immune scores on lymphatic involvement, providing a convenient and intuitive evaluation tool for clinicians; clinicians can formulate more personalized treatment plans based on the specific conditions of patients and the prediction results given by the nomogram model.

[0019] 5) Improving the efficiency of clinical decision-making: By predicting lymphatic involvement in advance, the systems and methods of the present application help clinicians identify high-risk patients earlier, so as to take timely intervention measures, such as strengthening postoperative adjuvant treatment, etc., to improve the prognosis of patients; at the same time, the systems and methods also provide new perspectives and ideas for clinical research, contributing to the progress and development of breast cancer diagnosis and treatment technologies.

[0020] 6) Optimization strategy for sample size dependence: Although the sample size is one of the limiting conditions for the accuracy of the model, the present application minimizes the impact of the sample size on the prediction results through strict preprocessing and normalization steps, as well as the joint analysis of multi-omics data; in the future, with the continuous accumulation of data resources and the further optimization of technologies, the prediction accuracy and generalization ability of this model are expected to be further improved.

[0021] Term definitions or explanations: 1) Lymphatic Involvement Definition: Refers to the process by which tumor cells spread to lymph nodes or other lymphatic tissues through the lymphatic system, especially the metastasis of axillary lymph nodes, which is crucial for the clinical staging and treatment plan formulation of breast cancer.

[0022] Application: Evaluating the lymphatic involvement status is the key to formulating personalized treatment plans and predicting the prognosis of patients.

[0023] 2) Immune Score Definition: A score that estimates the content of immune cells in the tumor microenvironment based on gene expression data through a specific algorithm.

[0024] Application: In the present application, the ESTIMATE algorithm is used to calculate the immune score, but it is found that its predictive effect on lymphatic involvement is not significant.

[0025] 3) Residual Network (ResNet) Definition: A network architecture that solves the problems of gradient disappearance and degradation in the training of deep networks by introducing residual blocks.

[0026] Application: The ResNet50 model is used for deep learning feature extraction of pathological images, improving the abstract level of features and prediction accuracy.

[0027] 4) Gray-Level Co-occurrence Matrix (GLCM) Definition: A statistical method used to describe the texture characteristics of an image, which reflects the texture features of the image by calculating the gray-level dependence between pixel pairs.

[0028] Application: In the construction of radiomics scores, GLCM is used to extract the texture features of MRI images for predicting lymph node involvement.

[0029] 5) Gray-Level Run-Length Matrix (GLRLM) Definition: A statistical method used to analyze the gray-level run lengths in an image, which reflects the non-uniformity of the gray-level distribution in the image.

[0030] Application: In the construction of radiomics scores, GLRLM is used to extract the gray-level distribution features of MRI images to evaluate the lymph node involvement situation.

[0031] 6) ESTIMATE algorithm Definition: An algorithm based on gene expression data to estimate the content of stromal cells and immune cells in the tumor microenvironment.

[0032] Application: Used to calculate the immune score of breast cancer patients, but no significant predictive effect on lymph node involvement was found in this application.

[0033] 7) Z-score Normalization Definition: A method to convert data into dimensionless values with a mean of 0 and a standard deviation of 1, which is used to eliminate the influence of the dimensions and ranges of different scoring systems.

[0034] Application: When integrating multi-omics data, Z-score normalization is used to process pathological, imaging, and immune scores for comprehensive analysis. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 It is a composition diagram of the breast cancer lymph node involvement status prediction system of this application; Figure 2 It is a composition diagram of the pathological omics image processing module in the breast cancer lymph node involvement status prediction system of this application; Figure 3 It is a composition diagram of the radiomics image processing module in the breast cancer lymph node involvement status prediction system of this application; Figure 4A It is the pathological picture after being processed by the first preprocessing unit; Figure 4BIt is the distribution diagram of lambda coefficients of lasso regression in the pathological omics scoring step; Figure 4C It is the distribution diagram of mean squared error in lasso cross-validation in the pathological omics scoring step; Figure 4D It is the AUC curve diagram of the lasso model in the pathological omics scoring step; Figure 4E It is the weight ranking diagram of pathological omics depth features; Figure 4F It is the box plot of pathological scores of two types of patients with or without lymph node involvement; Figure 5A It is the distribution diagram of coefficients of lasso regression in the radiomics scoring step; Figure 5B It is the distribution diagram of mean squared error in lasso cross-validation in the radiomics scoring step; Figure 5C It is the AUC curve diagram of the lasso model in the radiomics scoring step; Figure 5D It is the weight ranking diagram of radiomics features; Figure 5E It is the box plot of radiomics scores of two types of patients in the radiomics scoring step; Figure 6A It is the distribution diagram of Stromal score, Immue score, and Estimate score of TCGA-BRCA patients evaluated by the estimate algorithm; Figure 6B It is the box plot of immune scores of two groups of patients; Figure 7 The nomogram after combining pathological imaging and immune scores. Specific implementation manner

[0036] To make the purpose, technical solutions, and advantages of the implementation of this application clearer, the technical solutions in the embodiments of this application will be described in more detail below in conjunction with the accompanying drawings in the embodiments of this application.

[0037] In the first aspect, this application discloses a breast cancer lymph node involvement status prediction system, as Figure 1 shown. This prediction system at least includes a database 11, a pathological omics image processing module 12, a radiomics image processing module 13, a pathological omics scoring calculation module 14, a radiomics scoring calculation module 15, an immune scoring calculation module 16, a normalization unit 17, a regression analysis module 18, a model construction module 19, and an output module 20.

[0038] Among them, the database 11 is the TCGA-BRCA database and the TCIA database. The database 11 stores the pathological section images (i.e., FFPE section image data) and clinical data (including MRI image data and gene expression data) of breast cancer patients.

[0039] The pathological omics image processing module 12 is used to extract the FFPE section image data of 1,133 patients, a total of 97 patients in this embodiment, from the database 11, and randomly select 70% of the section image data for preprocessing and feature extraction to obtain the deep pathological omics features of the patients.

[0040] Specifically, as Figure 2 shown, the pathological omics image processing module 12 may include a first extraction unit 121, a first preprocessing unit 122, a first feature extraction unit 123, and a feature processing unit 124.

[0041] Among them, the first extraction unit 121 is used to extract FFPE section image data from the database 11.

[0042] The first preprocessing unit 122 is used to preprocess the selected image data using a deep learning framework (CLAM framework). Here, the preprocessing at least includes color correction, blank image deletion, and format conversion processing, and uses preset parameters optimized for TCGA to extract the region of interest (ROI, such as the tumor region) in the FFPE section image data and remove holes (see Figure 4A ).

[0043] The first feature extraction unit 123 is used to divide the FFPE section image data processed by the first preprocessing unit 122 into multiple small blocks. In this embodiment, each section is divided into blocks (tiles) of 512×512 size, and a deep learning model pre-trained on Imagenet (such as the ResNet50 model) is used to extract the deep learning features of each small block. Among them, the integrated channel output of multiple tiles corresponding to each slide (section) is a tensor of dimensions (tile level tensors), and the mean value of the tensors of the same dimension of multiple tiles is taken to obtain the tensor (slide level tensors) corresponding to a single slide.

[0044] The feature processing unit 124 is used to integrate and average the deep learning features of all small blocks divided from multiple sections of the same patient to obtain the deep pathological omics features of the patient. Among them, the mean value of the tensors of the same dimension of multiple sections of the same patient is taken to obtain the pathological omics deep feature of a single patient with a dimension of .

[0045] Further, the radiomics image processing module 13 is used to extract the MRI image data of the patients in the database 11, and perform preprocessing and feature extraction on the MRI image data to obtain the radiomics features of the patients.

[0046] Specifically, as Figure 3 shown, the radiomics image processing module 13 may include a second extraction unit 131, a second preprocessing unit 132, and a second feature extraction unit 133.

[0047] Among them, the second extraction unit 131 is used to extract MRI image data from the database 11 (TCIA database). In this embodiment, 137 MRI image data corresponding to 104 patients are preferably extracted, and 96 remain after filtering in combination with the scanner model and clinical information. The MRI image data is a 2D image in DICOM format.

[0048] The second preprocessing unit 132 is used to convert the MRI image data into 3D NIfTI format through SimpleITK, perform N4 bias field correction to correct the image brightness difference caused by magnetic field inhomogeneity, and then normalize and resample the image intensity to eliminate the difference in spatial resolution of different platforms and reduce image inconsistency.

[0049] The second feature extraction unit 133 is used to extract radiomics features of the region of interest (which can be manually circled by a medical imaging expert for the ROI of the image) from the processed image data above. Among them, the radiomics features include but are not limited to first-order statistical features, two-dimensional and three-dimensional shape features, and gray-scale features. In this embodiment, 100 radiomics features are preferably extracted.

[0050] Further, the pathologyomics score calculation module 14 applies the Lasso regression algorithm in scikit-learn to screen the pathologyomics depth features with non-zero regression coefficients (see Figure 4B 、 Figure 4C ), and calculates the pathologyomics score of the patient using the screened features and corresponding weights.

[0051] In this embodiment, the Lasso regression algorithm is used to screen the pathology features extracted by deep learning, and 17 pathologyomics depth features are obtained (see Figure 4E ).

[0052] The performance of the Lasso classification model is evaluated using the AUC curve, and the AUC reaches 0.67 (see Figure 4D ). This result shows that the performance of the Lasso classification model can achieve a good effect in distinguishing lymph node involvement.

[0053] In deep learning, the model learns through multiple levels of weights and extracts the feature representations of data. These features gradually abstract from low-level edges, textures, grayscale, etc. to high-level objects.

[0054] In this application, the features of different deep learning are the results of the abstract normalization of the pixel values of pathological images, which are directly related to the learning ability, prediction accuracy, and generalization ability of the model. This application calculates the pathological depth score of each patient using the lasso regression coefficient w and the quantitative value f of the deep features. The specific formula is as follows: ; Then, after calculating the pathological score corresponding to each patient, the patients are grouped according to whether the lymph nodes are involved. The results show that the pathological scores of patients with lymph node involvement are significantly higher (see Figure 4F ).

[0055] Furthermore, the radiomics score calculation module 15 uses the N lymph node involvement event in TNM as the dependent variable, extracts 100 quantitative radiomics features from the ROI using pyradiomics software, and then applies the Lasso regression algorithm to screen (see Figure 5A , Figure 5B ) the radiomics features with non-zero regression coefficients, and calculates the radiomics score of each patient using the screened features and their corresponding weights.

[0056] In this embodiment, the following 4 key radiomics features are screened out using the Lasso algorithm: "original_shape_Sphericity"; "original_glcm_ClusterShade"; "orignal_glcm_Idmn"; "original_glrlm_GrayLevelNonUniformityNormalized"; Among them, original_shape_Sphericity reflects the sphericity of the lesion and helps to evaluate its morphological features; original_glcm_ClusterShade and original_glcm_Idmn are features of the gray-level co-occurrence matrix (GLCM) and are used to describe the texture characteristics of the image.

[0057] original_glrlm_GrayLevelNonUniformityNormalized is a feature of the gray-level run length matrix (GLRLM) and reflects the non-uniformity of the gray-level distribution in the image. Then, the radiomics score RAD of each patient is calculated using the lasso regression coefficient w and the quantitative value f of the image features. The specific formula is as follows: ; Meanwhile, the performance of the lasso classification model was evaluated, and the performance reached 0.61 (see Figure 5C ). Sorting was performed according to the weights of the features in the model, see Figure 5D , and the radiomics feature with the highest weight was original_glrlm_GrayLevelNonUniformityNormalize. After calculating the corresponding radiomic scores for each patient, the patients were grouped according to whether the lymph nodes were involved. The results showed that the radiomic scores of the patients with lymph node involvement were significantly higher (see Figure 5E ).

[0058] Furthermore, the immune score calculation module 16 is used to extract the gene expression data of the patients in the database 11, and use the ESTIMATE algorithm to estimate the stromal cell and immune cell contents in the tumor samples based on the gene expression data of the patients, and calculate the immune score.

[0059] Among them, ESTIMATE is a single-sample gene set enrichment analysis algorithm based on gene expression data, which is used to estimate the contents of stromal cells and immune cells in the tumor microenvironment. This algorithm generates four scores by calculating specific signals in the gene expression pattern: stromal score, immune score, ESTIMATE score, and tumor purity (see Figure 6A ). The ESTIMATE package was used to evaluate the immune scores of TCGA-BRCA. The gene expression profiles of the patients with TCGA-BRCA were downloaded from the UCSC database and corresponded to the patients in the pathomics. The immune scores of each patient were calculated using the estimate algorithm for the expression matrix after normalizing log2(FPKM). The immune scores of the two groups of patients with or without lymph node involvement in TCGA-BRCA were evaluated, and the results showed that there was no significant difference between the two groups (see Figure 6B ).

[0060] Furthermore, the normalization unit 17 is used to perform Z-score (standard score) normalization on the above-obtained (96 patients) pathomics scores, radiomics scores, and immune scores, so that all data are converted into dimensionless values with a mean of 0 and a standard deviation of 1.

[0061] Further, the regression analysis module 18 is used to perform statistical analysis on the normalized pathological omics score, radiomics score, and immune score using a statistical regression model (such as RMS regression analysis), so as to evaluate the effects of the three scores and their interaction terms on the lymph node involvement outcome. Among them, RSM is a statistical and mathematical method used to model and analyze the effects of multiple variables on the response variable, and is particularly suitable for dealing with complex multi-factor interactions.

[0062] Further, the model construction module 19 is used to construct a combined nomogram model integrating pathological omics, radiomics, and immune scores according to the statistical analysis results of the regression analysis module 18. The combined nomogram model is used to predict the lymph node involvement status of breast cancer patients. Among them, the results of this embodiment show that the imaging score has the most significant effect on lymph node involvement (see Figure 7 ), and this result emphasizes the core position of imaging in the assessment of lymph node status.

[0063] Further, the output module 20 is used to display and output the prediction results of the combined nomogram model.

[0064] In summary, this application integrates multi-omics data to establish a new breast cancer lymph node involvement prediction model; this model combines the advantages of pathological, imaging, and gene expression data, and provides a non-invasive lymph node status assessment system and method; this multi-modal integrated system and method is expected to improve the accuracy of breast cancer lymph node involvement prediction and provide more comprehensive information support for clinical decision-making.

[0065] In the second aspect, this application also discloses a method for predicting the lymph node involvement status of breast cancer, including the following steps: Step 1: Preprocess and extract features from the FFPE section image data of patients in the database to obtain the deep pathological omics features of the patients.

[0066] Specifically, preprocessing and feature extraction of the FFPE section image data of patients in the database specifically include the following steps: Step 1.1: Extract the FFPE section image data from the database; Step 1.2: Use a deep learning framework to preprocess the image, including color correction, deletion of blank images, and format conversion, and extract the region of interest in the FFPE section image data and remove holes; Step 1.3: Cut the FFPE section image data processed in Step 1.2 into multiple small pieces, and use a pre-trained deep learning model to extract the deep learning features of each small piece; Step 1.4: Integrate and average the deep learning features of all small pieces cut from multiple sections of the same patient to obtain the deep pathological omics features of the patient.

[0067] Step 2: Preprocess and extract features from the MRI image data of patients in the database to obtain the radiomics features of the patients.

[0068] Specifically, the preprocessing and feature extraction of the MRI image data of patients in the database specifically include the following steps: Step 2.1: Extract the MRI image data from the database; Step 2.2: Perform preprocessing including format conversion, bias field correction, normalization, and resampling on the MRI image data; Step 2.3: Extract the radiomics features of the region of interest from the image data processed in Step 2.2.

[0069] Step 3: Calculate the pathomics score of the patient according to the deep pathomics features, calculate the radiomics score of the patient according to the radiomics features, and calculate the immune score of the patient according to the gene expression data in the database.

[0070] In this step, the pathomics score is specifically to apply the Lasso regression algorithm to screen the deep pathomics features with non-zero regression coefficients, and use the screened features and corresponding weights to calculate the pathomics score of the patient.

[0071] The radiomics score is specifically to apply the Lasso regression algorithm to screen the radiomics features with non-zero regression coefficients, and use the screened features and corresponding weights to calculate the radiomics score of each patient.

[0072] The immune score is to use the ESTIMATE algorithm to estimate the content of stromal cells and immune cells in tumor samples based on the gene expression data of the patient, so as to calculate the immune score.

[0073] Step 4: Conduct statistical analysis on the pathomics score, radiomics score, and immune score of the patient. The specific method of this step can be referred to the corresponding part in the above prediction system and will not be elaborated here.

[0074] Step 5: Construct a combined nomogram model integrating pathomics, radiomics, and immune scores according to the statistical analysis results in Step 4. Similarly, the specific method of this step can be referred to the corresponding part in the above prediction system and will not be elaborated here.

[0075] Step 6: Use the combined nomogram model to predict the lymph node involvement status of breast cancer patients, and display and output the prediction results.

[0076] As described above, it is only the specific implementation manner of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by those skilled in the art within the technical scope disclosed in the present application should be covered within the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the protection scope of the claimed rights.

Claims

1. A breast cancer lymph node involvement status prediction system, characterized in that: include: Database for storing FFPE slice image data, MRI imaging data, and gene expression data of breast cancer patients; A pathology genomics image processing module is used to extract the FFPE slice image data of the patient in the database, and perform preprocessing and feature extraction on the FFPE slice image data to obtain the pathology genomics depth features of the patient; An imaging omics image processing module is used to extract the MRI image data of the patient in the database, and perform preprocessing and feature extraction on the MRI image data to obtain the imaging omics features of the patient; A pathology omics score calculation module, used to calculate the patient's pathology omics score according to the pathology omics depth features; A radiomics score calculation module, used to calculate the patient's radiomics score according to the radiomics features; An immune score calculation module, used to extract the gene expression data of the patient in the database and calculate the immune score of the patient based on the gene expression data; Regression analysis module, used to perform statistical analysis on the patient's pathological scores, imaging scores, and immune scores; A model building module, used for building a combined nomogram model of comprehensive pathological omics, imaging omics and immune score according to the statistical analysis results of the regression analysis module, wherein the combined nomogram model is used for predicting the lymph node involvement status of breast cancer patients; The output module is used to display and output the prediction results of the combined nomogram model.

2. The breast cancer lymph node involvement status prediction system according to claim 1, characterized in that: The pathology genomics image processing module comprises: A first extraction unit, used for extracting the FFPE slice image data from the database; A first preprocessing unit is used to perform preprocessing including color correction, blank image deletion and format conversion on the image using a deep learning framework, and extract the region of interest in the FFPE slice image data and remove holes; A first feature extraction unit, configured to divide the FFPE slice image data processed by the first preprocessing unit into a plurality of small blocks, and extract a deep learning feature of each small block using a pre-trained deep learning model; The feature processing unit is used to integrate and average the deep learning features of all small blocks cut from multiple slices of the same patient to obtain the patient's pathological genomics deep features.

3. The breast cancer lymph node involvement status prediction system according to claim 2, characterized in that: The pathology omics score calculation module uses the Lasso regression algorithm to screen the pathology omics deep features whose regression coefficients are not zero, and uses the screened features and corresponding weights to calculate the patient's pathology omics score.

4. The breast cancer lymph node involvement status prediction system according to claim 1, characterized in that: The radiomics image processing module comprises: A second extraction unit, used for extracting the MRI image data from the database; A second preprocessing unit, used for performing format conversion, bias field correction, normalization and resampling preprocessing on the MRI image data; The second feature extraction unit extracts the radiomics features of the region of interest from the image data processed by the second preprocessing unit, wherein the radiomics features include but are not limited to first-order statistical features, two-dimensional and three-dimensional shape features, and grayscale features.

5. The breast cancer lymph node involvement status prediction system according to claim 4, characterized in that: The radiomics score calculation module uses the Lasso regression algorithm to screen radiomics features with non-zero regression coefficients, and uses the screened features and corresponding weights to calculate the radiomics score of each patient.

6. The breast cancer lymph node involvement status prediction system according to claim 4, characterized in that: The immune score calculation module uses the ESTIMATE algorithm to estimate the content of stromal cells and immune cells in the tumor sample based on the patient's gene expression data, and calculates the immune score. The immune score includes four scores: stromal score, immune score, ESTIMATE score and tumor purity.

7. The breast cancer lymph node involvement status prediction system according to claim 1, characterized in that: Also includes: The normalization unit is used to perform Z-score normalization processing on the pathological omics score, the imaging omics score and the immune score before the regression analysis module.

8. A method for predicting the lymph node involvement status of breast cancer, characterized in that: The steps include: Step 1: Preprocess and extract features of the FFPE slice image data of the patients in the database to obtain the pathological genomics deep features of the patients; Step 2: Preprocess and extract features of the MRI image data of the patients in the database to obtain the imaging features of the patients; Step 3, calculating the patient's pathological omics score according to the pathological omics depth features, calculating the patient's imaging omics score according to the imaging omics features, and calculating the patient's immune score according to the gene expression data in the database; Step 4: Statistical analysis of the patients' pathological scores, imaging scores, and immune scores; Step 5: construct a combined nomogram model of comprehensive pathological omics, imaging omics and immune score based on the statistical analysis results of step 4; Step six: Use the combined nomogram model to predict the lymph node involvement status of breast cancer patients, and display and output the prediction results.

9. The method for predicting lymph node involvement status of breast cancer according to claim 8, characterized in that: In the step 1, preprocessing and feature extraction of FFPE slice image data of patients in the database specifically includes the following steps: Step 1.1, extracting the FFPE slice image data from the database; Step 1.2, using a deep learning framework to preprocess the image including color correction, blank image deletion, and format conversion, and extracting the region of interest in the FFPE slice image data and removing holes; Step 1.3, dividing the FFPE slice image data processed in step 1.2 into multiple small blocks, and using the pre-trained deep learning model to extract deep learning features of each small block; Step 1.4: Integrate and average the deep learning features of all small blocks of multiple slices of the same patient to obtain the patient's pathological omics deep features; Further, in the step three, calculating the pathology score of the patient according to the pathology depth feature includes: The Lasso regression algorithm was used to screen the pathological omics deep features with non-zero regression coefficients, and the screened features and corresponding weights were used to calculate the patient's pathological omics score.

10. The method for predicting lymph node involvement status of breast cancer according to claim 8, characterized in that: In step 2, preprocessing and feature extraction of the MRI image data of patients in the database specifically includes the following steps: Step 2.1, extracting the MRI image data from the database; Step 2.2, performing format conversion, bias field correction, normalization and resampling preprocessing on the MRI image data; Step 2.3, extracting radiomic features of the region of interest from the image data processed in step 2.2; Further, in the step three, calculating the patient's radiomics score according to the radiomics features includes: The Lasso regression algorithm was used to screen radiomic features with non-zero regression coefficients, and the radiomics score of each patient was calculated using the screened features and the corresponding weights.

Images