Breast cancer axillary lymph node metastasis prediction method based on imageomics and domain adaptation

Abstract

The present application relates to the technical field of pathological digital image processing, and particularly relates to a breast cancer axillary lymph node metastasis prediction method based on imageomics and domain adaptation; the specific steps of the method are as follows: S1, selecting multicenter data and dividing; S2, preprocessing the data in S1; S3, extracting features from the data preprocessed in S2; S4, screening the features extracted in S3; S5, constructing a domain adaptation-imageomics model and verifying; the present application introduces a domain adaptation algorithm, reduces the feature distribution difference between the source domain and the target domain, and better reflects the real clinical practice environment by incorporating multicenter image data; the present application incorporates multiple sequence preoperative breast magnetic resonance images, and enriches the available information of imageomics; the present application provides a guarantee for improving the performance and generalization of imageomics in predicting breast cancer axillary lymph node metastasis, and provides technical support for precision medical services.

Description

Technical Field

[0001] This invention relates to the technical field of pathological digital image processing, specifically to a method for predicting axillary lymph node metastasis in breast cancer based on radiomics and domain adaptation. Background Technology

[0002] In the diagnosis and treatment of breast cancer, axillary lymph node metastasis (ALNM) is highly correlated with distant metastasis, recurrence, and survival rates. Accurate detection and effective assessment of ALNM are crucial for patient prognosis and treatment selection. Radiomics can reveal valuable insights and quantitative information in imaging data that may be invisible to the naked eye, helping medical professionals predict the axillary lymph node status in breast cancer patients. Among existing similar patents, CN113555115A discloses a method for constructing a lymph node metastasis prediction model for breast cancer patients based on radiomics. This method uses a random forest algorithm to screen image features, performs univariate analysis on clinical feature data to obtain key clinical features, and uses a support vector machine algorithm to establish a comprehensive lymph node metastasis prediction model based on key imaging and clinical features. This achieves the establishment of a prediction model based on the principle of structural risk minimization, aiding in lymph node prediction. CN116741390A also discloses a predictive model for axillary lymph node metastasis in breast cancer patients. This method involves performing three-dimensional reconstruction on two-dimensional enhanced CT images of the lungs, selecting more than five radiomics features from all radiomics features of each axillary lymph node to distinguish whether the breast cancer has metastasized to the axillary lymph nodes, and using random forest to construct a prediction model.

[0003] Although some patents have disclosed methods for intelligent prediction of axillary lymph node status in breast cancer using radiomics, their technical solutions still have some shortcomings: 1. They do not use multi-center datasets to validate the radiomics model, which cannot guarantee that the model will have good predictive performance in unfamiliar centers with different distributions, resulting in low model reliability; 2. Models that use multi-center datasets to validate their performance do not necessarily have good predictive ability in every center. When the distribution of the external validation set differs greatly from that of the training set, the model performs poorly.

[0004] Therefore, it is of great significance to develop a new method for predicting axillary lymph node metastasis in breast cancer based on radiomics and domain adaptation. Summary of the Invention

[0005] Purpose of the invention: The technical problem to be solved by the present invention is to address the shortcomings of the existing technology and provide a method for predicting axillary lymph node metastasis of breast cancer based on radiomics and domain adaptation, which is different from existing patents of the same type. It starts from the perspective of improving the performance of radiomics models in multi-center datasets by incorporating domain adaptation algorithms into the traditional radiomics process.

[0006] To address the aforementioned technical problems, this invention discloses a method for predicting axillary lymph node metastasis in breast cancer based on radiomics and domain adaptation. The specific steps are as follows:

[0007] S1. Select multicenter data and divide it;

[0008] S2. Preprocess the data in S1;

[0009] S3. Perform feature extraction on the data preprocessed in S2;

[0010] S4. Filter the features extracted in S3;

[0011] S5. Construct and validate a domain adaptation-radiomics model.

[0012] The multi-center data consists of multi-parameter MRI images; the multi-parameter MRI images are eight sequences from phases 0 to 5 of T2WI, nonfs-T1WI, and DCE-MRI.

[0013] The specific steps of S2 are as follows: segmenting the divided data and delineating the ROI; resampling and standardizing the delineated image;

[0014] Preferably, the size of the resampling is 1×1×1mm;

[0015] In S3, the features extracted during the feature extraction process include shape features, first-order features, texture features, and higher-order features obtained by various transformations of the original image. The texture features include gray-level co-occurrence matrix features, gray-level run-length matrix features, gray-level region size matrix features, and gray-level dependency matrix features. The transformations used in the higher-order features include square, square root, logarithm, exponential, gradient magnitude, local binary pattern, wavelet filtering, and Laplacian of Gaussian.

[0016] In step S4, feature selection is performed using ANOVA, mRMR, and LASSO regression.

[0017] Specifically, the domain adaptation-radiomics model is a model constructed by incorporating a domain adaptation algorithm into the model development step of the traditional radiomics workflow (image acquisition - image segmentation - radiomics feature extraction - feature selection - model development - model performance evaluation). First, the key features selected from the training set and the corresponding identical features from the external validation set are input into the Balanced Distribution Adaptation (BDA) domain adaptation algorithm. Then, the BDA algorithm achieves alignment between the two domains by minimizing the difference in feature distributions between the source domain (training set) and the target domain (external validation set). The objective formula of this algorithm is as follows:

[0018]

[0019] stA T XHX T A = I. 0 ≤ μ ≤ 1

[0020] Where X represents x s and x t The input data matrix consists of A, which represents the transformation matrix, μ, and M0 and M... c It is an MMD matrix, where c∈{1,2,···,C} is the category of the label, and λ is the regularization parameter. It is a Frobenius norm. .I∈R (m+n)×(m+n) ' is the identity matrix, and H is the central matrix, i.e., H = I - (1 / n)1.

[0021] The formula contains two terms: adaptation for marginal and conditional distributions with equilibrium factors (term 1) and a regularization term (term 2). Furthermore, the formula has two constraints: the first constraint ensures that the data after the distribution transformation (A) T X) The internal properties of the original data should be preserved. The second constraint specifies the range of the balancing factor μ.

[0022] The optimal way to adapt to the marginal and conditional distributions of the two domains is achieved by adjusting the parameter μ in the above formula. If the difference in the marginal distributions between the two domains is considered more important, the value of μ is made to approach 0; if the difference in the conditional distributions is considered more important, the value of μ is made to approach 1. Then, using the new distribution and the classifier embedded in the BDA algorithm, a domain adaptation-radiomics model is constructed on the training set. Finally, the domain adaptation-radiomics model predicts the labels for the target domain (i.e., the external validation set), obtains pseudo-labels, and iteratively refines them to generate output labels for performance verification.

[0023] The present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the present invention's method for predicting axillary lymph node metastasis in breast cancer based on radiomics and domain adaptation.

[0024] The present invention also provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of the present invention’s method for predicting axillary lymph node metastasis in breast cancer based on radiomics and domain adaptation.

[0025] Beneficial effects: Compared with traditional radiomics models, this invention introduces a domain adaptation algorithm, which can reduce the difference in feature distribution between the source and target domains and overcome the differences in magnetic resonance imaging data caused by different hospitals, different instruments and equipment, or different types of patients, thereby improving the model's predictive performance and generalization ability in multi-center studies. This invention incorporates multi-center imaging data, with different centers having different data distributions, which can better reflect the real clinical practice environment. This invention includes preoperative breast magnetic resonance images of breast cancer patients in eight sequences: T2WI, non-fsT1WI, and DCE-MRI stages 0-5, which can comprehensively extract and analyze the multidimensional radiomics features of tumors, enriching the information available for radiomics. This invention provides a guarantee for improving the performance and generalization of radiomics in predicting axillary lymph node metastasis in breast cancer and provides technical support for precision medicine services. Attached Figure Description

[0026] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments, and the advantages of the present invention in the above and / or other aspects will become clearer.

[0027] Figure 1 This is a flowchart of a breast cancer axillary lymph node metastasis prediction method based on radiomics and domain adaptation in an embodiment of the present invention;

[0028] Figure 2 This is the LASSO regression analysis cross-validation curve in an embodiment of the present invention;

[0029] Figure 3 This is a LASSO coefficient curve diagram in an embodiment of the present invention;

[0030] Figure 4 These are the ROC curves of all models in the training set in this embodiment of the invention;

[0031] Figure 5 The ROC curves are from the external validation set in this embodiment of the invention. Detailed Implementation

[0032] The following description, in conjunction with the embodiments and accompanying drawings, provides further details, but should not be construed as limiting the scope of protection of this invention.

[0033] Example 1:

[0034] 1. Experimental Data: A retrospective analysis was conducted, collecting preoperative multiparametric MRI images of 556 patients who underwent breast MRI examinations at four centers (Centers 1-4) between January 2014 and December 2023. These patients had surgical or biopsy-confirmed malignant lesions and possessed immunohistochemical results. Center 1 is the Second Affiliated Hospital of Nantong University, Center 2 is the Third Affiliated Hospital of Nantong University, Center 3 is Nantong Municipal Hospital of Traditional Chinese Medicine, and Center 4 is the Affiliated Hospital of Nantong University. After screening based on the patient inclusion criteria (① patients who underwent MRI examination 2 weeks prior to surgery; ② MRI results showing a breast mass on at least two consecutive slices; ③ positive cases confirmed by lymph node dissection or ultrasound fine-needle aspiration biopsy, and negative cases confirmed by lymph node dissection surgery; ④ no missing laboratory test results and clinical medical records) and exclusion criteria (① no enhancement of breast cancer lesions after contrast administration; ② history of puncture or radiotherapy / chemotherapy before breast MRI examination; ③ poor image quality), 506 patients remained. Among them, Center 1 included 365 patients, and 5 patients were included in bilateral breast lesion lesion categorization, resulting in a total of 370 lesions in Center 1. Centers 2, 3, and 4 included 25, 11, and 105 patients, respectively. The multiparametric MRI images mentioned included T2-weighted imaging (T2WI), non-fat-suppressed T1-weighted imaging (nonfs-T1WI), and dynamic contrast-enhanced MRI (DCE-MRI) with 8 sequences from phases 0 to 5.

[0035] 2. Dataset Splitting: Center 1 was used as the training set (n=370), and Centers 2, 3, and 4 were combined as the external validation set (n=141). Patients in both the training and external validation sets were divided into positive and negative groups based on whether axillary lymph node metastasis was present. For lesions positive for axillary lymph node metastasis (ALNM), confirmation was achieved through lymph node dissection or ultrasound-guided fine-needle aspiration biopsy. For ALNM negative lesions, confirmation was achieved solely through axillary lymph node dissection. The training set showed 169 lesions in the ALNM positive group and 201 lesions in the ALNM negative group, with an ALNM positivity rate of 45.7%. The external validation set showed 70 lesions in the ALNM positive group and 71 lesions in the ALNM negative group, with an ALNM positivity rate of 49.6%.

[0036] 3. Data Preprocessing: ROI Delineation: Eight sequences of images from all patients (T2WI, nonfs-T1WI, and DCE-MRI stages 0-5) were imported into the Deepwise Research Platform in DICOM format for semi-automatic ROI delineation. This involves first automatically segmenting the lesion area, followed by manual fine-tuning by two imaging experts with 15 and 8 years of MRI image interpretation experience, respectively. During ROI delineation, factors such as cystic degeneration, necrosis, and calcification within the tumor are not avoided.

[0037] Resampling: Eight sequences of images from T2WI, nonfs-T1WI, and DCE-MRI phases 0-5 were resampled to the same size of 1×1×1mm;

[0038] Standardization: The window width and window level of all images from 8 sequences of T2WI, nonfs-T1WI, and DCE-MRI stages 0-5 were standardized.

[0039] 4. Feature extraction: On the Deepwise Research Platform, a discretization operation was adopted to extract 1648 radiomics features for each lesion and each MR sequence (T2WI, non-fsT1WI, and 8 sequences of DCE-MRI stages 0-5) in the training set and external validation set.

[0040] The 1648 imagemic features include 14 shape features, 18 first-order features, 68 texture features, and 1548 higher-order features extracted from the original images through various transformations. The texture features include 22 gray-level co-occurrence matrix (GLCM) features, 16 gray-level run-length matrix (GLRLM) features, 16 gray-level region size matrix (GLSZM) features, and 14 gray-level dependency matrix (GLDM) features. The transformations used for the higher-order features include square (86), square root (86), logarithmic (86), exponential (86), gradient magnitude (86), and local binary pattern (86), as well as wavelet filtering (688) and Laplacian Gaussian (344). Each sequence yields 1648 features, resulting in a total of 13184 features extracted from the eight sequences.

[0041] 5. Feature Selection: First, ANOVA was used to select features with a variance threshold of 1e5, resulting in 2926 features. Then, mRMR was used to select features that maximized the correlation with the target category while minimizing redundant information among them, with a threshold of 130, resulting in 130 features. Finally, LASSO regression was used to select all features with non-zero coefficients, ultimately selecting 19 radiomics features for further analysis. Figure 2 The cross-validation curves of LASSO regression analysis for radiomics features are shown. The gray dashed line in the figure represents the optimal value of λ. Figure 3 This is a LASSO coefficient curve for 19 radiomics features. Table 1 lists the names and coefficient values ​​of these 19 features;

[0042] Table 1. Selected radiomics features and coefficients

[0043] T2WI_exponential_gldm_grayscale nonuniformity 0.05082014 T2WI_square root_glszm_grayscale variance -0.1003547 T2WI_l_logarithmic_glrlm_grayscale variance 0.07035194 nonfsT1WI_original_glszm_region size inhomogeneity 0.02143709 nonfsT1WI_log-σ-3-0-mm-3D_gldm_high-dependency high-grayscale emphasis -0.02576701 nonfsT1WI_wavelet-HLL_glcm_cluster shadow 0.008178519 DCE-0_Wavelet-LLL_gldm_High Dependency High Grayscale Emphasis -0.000731806 DCE-1_Index_glrlm_Stroke Non-uniformity 0.005392295 DCE-1_Logarithmic_First_Order_Energy -0.09972658 DCE-1_Logarithmic_First_Order_Total_Energy -7.22545E-16 DCE-2_log_gldm_dependent_uniformity -0.04126104 DCE-3_Original_Shape_Surface Area 0.06134264 DCE-3_index_glszm_region size non-uniformity 0.00252967 DCE-3_Index_glrlm_Stroke Non-uniformity 0.08836839 DCE-4_Square Root_glszm_Grayscale Variance -0.03344872 DCE-4_log-σ-2-0-mm-3D_gldm_High Dependency High Grayscale Emphasis -0.005703682 DCE-4_Wavelet-LLL_gldm_High Dependency High Grayscale Emphasis 0.08913544 DCE-5_log-σ-5-0-mm-3D_glszm_Region Size Inhomogeneity 0.05786319 DCE-5_Wavelet-LHL_glszm_Large Area High Gray Scale Emphasis -0.03250647

[0044] Where: T2WI, T2-weighted imaging; gldm, gray-level dependence matrix; glszm, gray-level region size matrix; grlm, gray-level run matrix; nonfsT1WI, non-fat-suppressed T1-weighted imaging; log, Laplacian of Gaussian; glcm, gray-level co-occurrence matrix; DCE, dynamic contrast enhancement.

[0045] 6. Domain Adaptation - Radiomics Model Construction and External Validation: In constructing the domain adaptation - radiomics model, the key features selected from the training set in S5 and their corresponding identical features in the external validation set are first input into the Balanced Distribution Adaptation (BDA) domain adaptation algorithm. Then, the BDA algorithm aligns the two domains by minimizing the difference in feature distributions between the source domain (training set) and the target domain (external validation set). The objective formula for this algorithm is as follows:

[0046]

[0047] stA T XHX T A = I. 0 ≤ μ ≤ 1

[0048] Where X represents x s and x t The input data matrix consists of A, which represents the transformation matrix, μ, and M0 and M... c It is an MMD matrix, where c∈{1,2,···,C} is the category of the label, and λ is the regularization parameter. It is the Frobenius norm. I∈R (m+n)×(m+n) ' is the identity matrix, and H is the central matrix, i.e., H = I - (1 / n)1.

[0049] The formula contains two terms: adaptation for marginal and conditional distributions with equilibrium factors (term 1) and a regularization term (term 2). Furthermore, the formula has two constraints: the first constraint ensures that the data after the distribution transformation (A) T X) The internal properties of the original data should be preserved. The second constraint specifies the range of the balancing factor μ.

[0050] The optimal way to adapt to the marginal and conditional distributions of the two domains is achieved by adjusting the parameter μ in the above formula. If the difference in the marginal distributions of the two domains is considered more important, the value of μ is made to approach 0; if the difference in the conditional distributions is considered more important, the value of μ is made to approach 1. Then, the new distribution is used to construct a domain adaptation-radiomics model on the training set using the classifier embedded in the BDA algorithm. Finally, the domain adaptation-radiomics model predicts the labels of the target domain, i.e., the external validation set, obtains pseudo-labels, and iteratively refines them to finally generate output labels for performance verification. Furthermore, to investigate the superiority of the domain adaptation algorithm, this embodiment also constructs six traditional radiomics models for performance comparison. The following six models are developed on the training set: Support Vector Machine (SVM), k-Nearest Neighbors (KNN), Logistic Regression (LR), Decision Tree (DT), Extreme Gradient Boosting (XGBoost), and Naive Bayes (NB), with selected key features as input and lymph node metastasis status as output.

[0051] All trained models predicted the axillary lymph node status for each patient in the external validation set. The performance of each model on both the training and external validation sets was evaluated using receiver operating characteristic (ROC) curves and the area under the ROC curve (AUC), and all models were compared. The ROC curves of all models on the training and external validation sets are shown below. Figure 3 As shown in Table 2, the AUC results (95% confidence interval) and the p-values ​​obtained from the Delong test are presented.

[0052] Table 2. AUC results of domain adaptation-radiomics model and traditional radiomics model in predicting ALNM.

[0053]

[0054]

[0055] Where: p-value represents the comparison between the BDA domain adaptation-radiomics model and other traditional radiomics models in the training queue and external validation queue. AUC: Area under the curve; BDA: Balanced Distribution Adaptation; SVM: Support Vector Machine; KNN: k-Nearest Neighbor; LR: Logistic Regression; DT: Decision Tree; XGBoost: Extreme Gradient Boosting; NB: Naive Bayes.

[0056] The results showed that the BDA domain-adaptive radiomics model had the best predictive performance, with an AUC of 0.978 on the training set. Among the six traditional radiomics models, DT performed best on the training set, with an AUC of only 0.866. On the other hand, the BDA domain-adaptive radiomics model had an AUC of 0.796 on the external validation set, higher than all traditional radiomics models, while XGBoost had the highest AUC among the traditional radiomics models, reaching only 0.698. This indicates that the domain-adaptive radiomics model can improve the low predictive performance and low generalization ability of traditional radiomics models in external centers, which is beneficial for the widespread application of this method in various clinical centers.

[0057] Example 2:

[0058] The computer-readable storage medium of this embodiment stores a computer program that, when executed by a processor, implements the steps in the breast cancer axillary lymph node metastasis prediction method based on radiomics and domain adaptation of Embodiment 1.

[0059] The computer-readable storage medium in this embodiment can be an internal storage unit of the terminal, such as the terminal's hard disk or memory; the computer-readable storage medium in this embodiment can also be an external storage device of the terminal, such as a plug-in hard disk, smart memory card, secure digital card, flash memory card, etc. equipped on the terminal; furthermore, the computer-readable storage medium can include both the terminal's internal storage unit and external storage devices.

[0060] The computer-readable storage medium of this embodiment is used to store computer programs and other programs and data required by the terminal. The computer-readable storage medium can also be used to temporarily store data that has been output or will be output.

[0061] Example 3:

[0062] The computer device of this embodiment includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the steps in the breast cancer axillary lymph node metastasis prediction method based on radiomics and domain adaptation of Embodiment 1.

[0063] In this embodiment, the processor can be a central processing unit, or other general-purpose processors, digital signal processors, application-specific integrated circuits, off-the-shelf programmable gate arrays or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor, etc. The memory can include read-only memory and random access memory, and provides instructions and data to the processor. A portion of the memory can also include non-volatile random access memory. For example, the memory can also store device type information.

[0064] Those skilled in the art will understand that the content disclosed in the embodiments can be provided as a method, system, or computer program product. Therefore, this solution can take the form of a hardware embodiment, a software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this solution can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage and optical storage) containing computer-usable program code.

[0065] This solution is described with reference to flowchart illustrations and / or block diagrams of methods and computer program products according to embodiments of this solution. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing device to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing device, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0066] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0067] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0068] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc.

[0069] This invention provides a method for predicting axillary lymph node metastasis in breast cancer based on radiomics and domain adaptation. Many methods and approaches exist for implementing this technical solution; the above description is merely a preferred embodiment of the invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of this invention, and these improvements and modifications should also be considered within the scope of protection of this invention. All components not explicitly stated in this embodiment can be implemented using existing technologies.

Claims

1. A method for predicting axillary lymph node metastasis in breast cancer based on radiomics and domain adaptation, characterized in that, The specific steps are as follows: S1. Select multicenter data and divide it; S2. Preprocess the data in S1; S3. Perform feature extraction on the data preprocessed in S2; S4. Filter the features extracted in S3; S5. Construct and validate a domain adaptation-radiomics model; The construction process of the domain adaptation-radiomics model is as follows: Input the selected key features and the corresponding identical features in the external validation set, use the balanced distribution adaptation domain adaptation algorithm to adjust the feature distribution of the source domain and the target domain to achieve alignment of the two domains, and use the classifier embedded in the balanced distribution adaptation domain adaptation algorithm to complete the construction of the domain adaptation-radiomics model on the training set using the new distribution. The objective formula for the balanced distribution adaptation domain adaptation algorithm is as follows: Where X represents the product of and The input data matrix consists of A, where A represents the transformation matrix. It is a balancing factor. and It is an MMD matrix, where c∈{1, 2, ..., C} is the category of the label. It is a regularization parameter. It is the Frobenius norm; I∈ ' is the identity matrix, and H is the central matrix, i.e., H = I - (1 / n)1.

2. The prediction method according to claim 1, characterized in that, The multicenter data consists of multiparametric MRI images; the multiparametric MRI images are eight sequences from phases 0 to 5 of T2WI, nonfs-T1WI, and DCE-MRI.

3. The prediction method according to claim 1, characterized in that, The specific steps of S2 are as follows: the divided data is segmented and ROI is delineated; the delineated image is resampled and standardized.

4. The prediction method according to claim 1, characterized in that, In S3, the features extracted during the feature extraction process are shape features, first-order features, texture features, and higher-order features obtained by various transformations of the original image; the texture features are gray-level co-occurrence matrix features, gray-level run-length matrix features, gray-level region size matrix features, and gray-level dependency matrix features; the transformations used in the higher-order features are square, square root, logarithm, exponential, gradient magnitude, local binary mode, wavelet filtering, and Laplacian of Gaussian.

5. The method according to claim 1, characterized in that, In step S4, feature selection is performed using ANOVA, mRMR, and LASSO regression.

6. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by the processor, it implements the steps in the method for predicting axillary lymph node metastasis of breast cancer based on radiomics and domain adaptation as described in any one of claims 1 to 5.

7. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps in the method for predicting axillary lymph node metastasis of breast cancer based on radiomics and domain adaptation as described in any one of claims 1 to 5.

Images