Magnetic resonance image breast cancer lump vessel tumor embolus and histological grading method

Abstract

The invention discloses a magnetic resonance image breast cancer lump vessel tumor embolus and histological grading method and system. The grading method comprises the steps that S1, a magnetic resonance DCE-MRI sequence image is collected; s2, carrying out preprocessing and standardization on the magnetic resonance DCE-MRI sequence image; s3, time-signal intensity curve parameters are extracted; s4, extracting a blood vessel network image around the breast mass; s5, performing multi-dimensional feature quantitative analysis on the blood vessels around the tumor embolus lumps according to the blood vessel network image to obtain a blood vessel multi-dimensional feature analysis result; s6, on the basis of the time-signal intensity curve parameters, constructing a tumor embolus histological grading prediction model, and predicting tumor histological grading; and S7, constructing a cancer thrombus pathology prediction model, and predicting a vascular cancer thrombus pathology result. The pathological result of the vascular cancer embolus and the histological grading of the tumor can be accurately predicted.

Description

Technical Field

[0001] This invention relates to the field of medical image processing and analysis, and in particular to a method for identifying vascular tumor emboli and histological grading of breast cancer masses in magnetic resonance imaging. Background Technology

[0002] Breast cancer is one of the most common malignant tumors in women. The presence or absence of vascular tumor thrombi and the histological grade of the tumor are key factors affecting patient prognosis and treatment options. Traditional diagnostic methods mainly rely on biopsy and pathological examination, which have disadvantages such as being invasive, having large sampling errors, and being time-consuming. Magnetic resonance imaging (MRI) plays an important role in the diagnosis of breast cancer as a non-invasive examination method, but current technology still has many limitations in the detection of vascular tumor thrombi and histological grading.

[0003] Currently, in clinical practice, doctors mainly rely on subjective experience to perform qualitative analysis of MRI images, lacking objective and quantitative evaluation standards, which can lead to significant individual differences in diagnostic results. Furthermore, existing imaging methods have limited ability to detect breast cancer vascular emboli, especially microemboli, and it is difficult to accurately predict the histological grade of tumors using imaging techniques.

[0004] With the development of artificial intelligence and medical imaging technology, quantitative analysis methods based on magnetic resonance imaging have been gradually applied, but there is still a lack of systematic and standardized solutions, especially specialized algorithms and models for the detection and histological grading of breast cancer vascular tumor thrombi. Summary of the Invention

[0005] In view of the above problems, the present invention is proposed to provide a magnetic resonance imaging method for vascular tumor thrombus and histological grading of breast cancer masses that overcomes or at least partially solves the above problems.

[0006] According to one aspect of the present invention, a method for grading vascular tumor thrombi in breast cancer masses via magnetic resonance imaging and histological grading is provided, the grading method comprising:

[0007] Step S1: Acquire DCE-MRI sequence images;

[0008] Step S2: Preprocess and standardize the magnetic resonance DCE-MRI sequence images;

[0009] Step S3: Perform quantitative analysis of dynamic contrast enhancement curves based on the preprocessed magnetic resonance DCE-MRI sequence images, and extract time-signal intensity curve parameters;

[0010] Step S4: Extract the image of the vascular network surrounding the breast mass;

[0011] Step S5: Quantitatively analyze the multidimensional features of the blood vessels surrounding the tumor thrombus based on the vascular network image to obtain the results of the multidimensional vascular feature analysis.

[0012] Step S6: Based on the time-signal intensity curve parameters, construct a tumor thrombus histological grading prediction model to predict the tumor histological grade;

[0013] Step S7: Based on the results of the multidimensional vascular feature analysis, construct a tumor thrombus pathology prediction model to predict the pathological results of vascular tumor thrombus.

[0014] Optionally, step S1: acquiring magnetic resonance DCE-MRI sequence images specifically includes:

[0015] Acquire baseline T1-weighted images of the patient, and images before contrast agent injection;

[0016] After the contrast agent was injected, T1-weighted images were continuously acquired at 7 time points to form a dynamic enhancement sequence;

[0017] Obtain the 1+7step dataset;

[0018] The captured images are saved in DICOM format.

[0019] Optionally, step S2: preprocessing and standardizing the magnetic resonance DCE-MRI sequence images specifically includes:

[0020] Motion correction: A non-rigid registration algorithm is used to spatially align the enhanced images at 7 time points with the baseline image to eliminate motion artifacts.

[0021] Signal intensity standardization: Using the ipsilateral pectoralis major muscle as a reference region, the relative signal intensity value of each voxel was calculated.

[0022] RSI = SI_voxel / SI_muscle, where SI_voxel is the signal intensity of the voxel and SI_muscle is the average signal intensity of the reference muscle region;

[0023] Noise removal: Gaussian filtering or median filtering is used to remove noise from the image, and the size of the filter kernel is dynamically adjusted according to the image resolution.

[0024] Image interpolation resampling: Trilinear interpolation resampling is performed on all images, with a uniform voxel size of 1mm×1mm×1mm;

[0025] Optionally, step S3: performing quantitative analysis of dynamic contrast enhancement curves based on the preprocessed magnetic resonance DCE-MRI sequence images, and extracting time-signal intensity curve parameters specifically includes:

[0026] Time-signal intensity curve generation: For each voxel, the signal intensity values ​​at the baseline and 7 enhancement time points are extracted to construct a time-signal intensity curve; for the tumor region, the region of interest (ROI) is determined by segmentation method, and the average curve of all voxels in the ROI is extracted;

[0027] Curve parameter calculation: Based on the time-signal intensity curve, calculate quantitative parameters:

[0028] Relative enhancement rate (RE): RE = (SI_t - SI_0) / SI_0 × 100%, where SI_t is the signal strength at time t, and SI_0 is the baseline signal strength; Maximum enhancement rate (MER): max(RE); Time to peak (TTP): Time it takes for the signal strength to reach its peak value; Wash-in slope (WIR): Rate of change of signal strength from baseline to peak value; Wash-out slope (WOR): Rate of change of signal strength from peak value to the last time point; Area under the curve (AUC): Area under the time-signal strength curve;

[0029] Curve morphology classification: Based on the morphological characteristics of the time-signal intensity curve, curves are classified into three types:

[0030] Type I, continuously rising type: signal strength continues to increase, with no obvious washout;

[0031] Type II, Platform Type: Signal strength increases first and then tends to stabilize;

[0032] Type III, wash-out type: signal strength first increases and then decreases significantly;

[0033] Obtain a set of parameters that reflect tumor angiogenesis and permeability characteristics.

[0034] Optionally, step S4: extracting the vascular network image around the breast mass specifically includes:

[0035] Optimal enhancement phase selection: Select the magnetic resonance imaging at the moment when the contrast agent reaches its peak from 7 enhancement time points;

[0036] Subtraction processing: The selected enhanced temporal image is subtracted from the baseline image to obtain a subtracted image, which highlights the contrast-filled vascular structures;

[0037] Maximum Intensity Projection (MIP): The subtracted 3D data is processed by maximum intensity projection to generate a 2D vascular projection image. The maximum pixel value on each ray is searched along a preset direction, and the maximum value is projected onto a 2D plane to form an MIP image of the vascular network.

[0038] Image enhancement processing: The MIP image is post-processed by contrast enhancement and sharpening. An adaptive histogram equalization method is used to enhance image contrast, and a nonlinear enhancement algorithm is used to highlight the vascular structure.

[0039] Obtain two-dimensional MIP images and three-dimensional reconstructed models that clearly show the vascular network surrounding a breast mass.

[0040] Optionally, step S5: Quantitative analysis of multi-dimensional features of blood vessels surrounding the tumor thrombus based on the vascular network image to obtain the multi-dimensional feature analysis results of blood vessels specifically includes:

[0041] ROI region determination: With the breast mass as the center, circular or spherical ROI regions with radii of 2cm and 5cm are delineated respectively, and the internal breast region is marked as a third ROI;

[0042] Blood vessel segmentation: Within each ROI region, threshold segmentation combined with morphological operations is used to extract blood vessel structures and generate blood vessel segmentation masks;

[0043] Vascular morphological feature extraction: Based on the segmentation results, morphological features are calculated: vessel density, vessel diameter, vessel length, vessel tortuosity, branch point density, vessel direction, and vessel connectivity.

[0044] Vascular texture feature extraction: Calculate the gray-level co-occurrence matrix features, gray-level run length matrix features, and wavelet transform features of the vascular network within the ROI;

[0045] Feature standardization: Perform Z-score standardization on all extracted features;

[0046] Obtain a multidimensional feature set describing the characteristics of the vascular network surrounding the tumor embolus mass.

[0047] Optionally, step S6: constructing a tumor thrombus histological grading prediction model based on the time-signal intensity curve parameters, specifically includes predicting tumor histological grading:

[0048] Feature selection: From the obtained dynamic contrast enhancement curve parameters, a subset of features is selected using LASSO regression, recursive feature elimination, or chi-square test methods;

[0049] Dataset splitting: Patient data were randomly divided into training and test sets in a 7:3 ratio;

[0050] Model training: Based on the training set data, construct a histological grading prediction model;

[0051] Model evaluation: Evaluate model performance using test set data;

[0052] Model optimization: Based on the evaluation results, ensemble learning methods are used to optimize model performance;

[0053] Model Interpretation: Using SHAP values ​​or feature importance analysis, the basis for the model's prediction results is explained, and the features most influential on histological grading prediction are identified.

[0054] A model was constructed to predict the histological grade of breast cancer based on DCE-MRI sequence images, achieving a three-level classification of low, intermediate, and high grades.

[0055] Optionally, step S7: constructing a tumor thrombus pathology prediction model based on the results of the multidimensional vascular feature analysis, and predicting the pathological results of vascular tumor thrombus, specifically includes:

[0056] Feature selection: From the vascular morphological and texture features obtained in Step 105, LASSO regression, recursive feature elimination, or chi-square test were used to select a subset of features that were relevant to the pathological results of tumor thrombus.

[0057] Dataset splitting: Patient data were randomly divided into training and test sets in a 7:3 ratio;

[0058] Model training: Construct a pathological prediction model for tumor thrombi based on the training set data;

[0059] Model evaluation: Evaluate model performance using test set data;

[0060] Model optimization: Based on the evaluation results, ensemble learning methods are used to optimize model performance; SMOTE oversampling technique or class weights are adjusted.

[0061] Model Interpretation: Using SHAP values ​​or feature importance analysis, the basis for the model's prediction results is explained, and the features most influential on the pathological prediction of tumor thrombi are identified.

[0062] To construct a model that can accurately predict the pathological results of vascular tumor thrombi based on vascular network characteristics.

[0063] A magnetic resonance imaging (MRI) system for classifying vascular tumor emboli and histological grading of breast cancer masses, using the MRI system for classifying vascular tumor emboli and histological grading of breast cancer masses as described in any one of claims 1-8, characterized in that the system comprises:

[0064] The image acquisition module is used to acquire DCE-MRI sequence images.

[0065] The preprocessing module is used to preprocess and standardize the magnetic resonance DCE-MRI sequence images;

[0066] The DCE-MRI curve analysis module is used to perform quantitative analysis of dynamic contrast enhancement curves based on the preprocessed magnetic resonance DCE-MRI sequence images and extract time-signal intensity curve parameters.

[0067] The vascular imaging generation module is used to extract images of the vascular network surrounding a breast mass;

[0068] The feature extraction module is used to perform quantitative analysis of the multidimensional features of blood vessels around the tumor thrombus mass based on the vascular network image, and obtain the results of multidimensional vascular feature analysis.

[0069] The model prediction module is used to construct a tumor thrombus histological grading prediction model based on the time-signal intensity curve parameters to predict the tumor histological grade.

[0070] The decision support module is used to construct a tumor thrombus pathology prediction model based on the results of the multidimensional vascular feature analysis, and to predict the pathological results of vascular tumor thrombus.

[0071] A computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the magnetic resonance imaging method for identifying vascular tumor thrombi and histological grading of breast cancer masses according to any one of claims 1-8.

[0072] This invention provides a method and system for grading vascular tumor thrombi and histological classification of breast cancer masses using magnetic resonance imaging. The grading method includes: Step S1: acquiring DCE-MRI sequence images; Step S2: preprocessing and standardizing the DCE-MRI sequence images; Step S3: performing quantitative analysis of dynamic contrast enhancement curves based on the preprocessed DCE-MRI sequence images, and extracting time-signal intensity curve parameters; Step S4: extracting vascular network images surrounding the breast mass; Step S5: performing quantitative analysis of multi-dimensional features of the vessels surrounding the tumor thrombus mass based on the vascular network images, and obtaining multi-dimensional vascular feature analysis results; Step S6: constructing a tumor thrombus histological grading prediction model based on the time-signal intensity curve parameters to predict the tumor histological grade; Step S7: constructing a tumor thrombus pathology prediction model based on the multi-dimensional vascular feature analysis results to predict the pathological results of vascular tumor thrombi. This method can accurately predict the pathological results of vascular tumor thrombi and the histological grade of tumors.

[0073] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it in accordance with the contents of the specification, and in order to make the above and other objects, features and advantages of the present invention more apparent and understandable, specific embodiments of the present invention are described below. Attached Figure Description

[0074] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0075] Figure 1 A flowchart of a method for identifying vascular tumor emboli and histological grading of breast cancer masses using magnetic resonance imaging, provided in an embodiment of the present invention;

[0076] Figure 2 This is a detailed flowchart illustrating a method for identifying vascular tumor thrombi and histological grading of breast cancer masses using magnetic resonance imaging, as provided in an embodiment of the present invention.

[0077] Figure 3 This is a block diagram of a magnetic resonance imaging (MRI) system for classifying vascular tumor thrombi and histological grading of breast cancer masses, provided as an embodiment of the present invention. Detailed Implementation

[0078] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0079] The terms "comprising" and "having," and any variations thereof, in the specification, embodiments, claims, and drawings of this invention are intended to cover non-exclusive inclusion, such as including a series of steps or units.

[0080] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments.

[0081] like Figure 1 As shown, a method and system for grading vascular tumor emboli and histological classification of breast cancer masses using magnetic resonance imaging, wherein the grading method includes:

[0082] Step S1: Acquire DCE-MRI sequence images;

[0083] Step S2: Preprocess and standardize the magnetic resonance DCE-MRI sequence images;

[0084] Step S3: Perform quantitative analysis of dynamic contrast enhancement curves based on the preprocessed magnetic resonance DCE-MRI sequence images, and extract time-signal intensity curve parameters;

[0085] Step S4: Extract the image of the vascular network surrounding the breast mass;

[0086] Step S5: Quantitatively analyze the multidimensional features of the blood vessels surrounding the tumor thrombus based on the vascular network image to obtain the results of the multidimensional vascular feature analysis.

[0087] Step S6: Based on the time-signal intensity curve parameters, construct a tumor thrombus histological grading prediction model to predict the tumor histological grade;

[0088] Step S7: Based on the results of the multidimensional vascular feature analysis, construct a tumor thrombus pathology prediction model to predict the pathological results of vascular tumor thrombus.

[0089] like Figure 2 As shown, a method for identifying vascular tumor emboli and histological grading in breast cancer masses using magnetic resonance imaging includes:

[0090] Step 101: DCE-MRI Sequence Image Acquisition

[0091] We acquired magnetic resonance imaging (DCE-MRI) sequences of breast cancer patients to form a "1+7step" dataset. Specifically, we first acquired baseline T1-weighted images of the patients, i.e., images before contrast agent injection, as the first step; then, after contrast agent injection, we continuously acquired T1-weighted images at 7 time points to form a complete dynamic enhancement sequence.

[0092] During the acquisition process, a dedicated breast coil was used. The patient was positioned prone, with the breast naturally suspended within the coil to minimize the impact of respiratory movements on image quality. Scanning parameters were set as follows: TR / TE = 4.5ms / 2.2ms, flip angle = 15°, slice thickness = 1mm, FOV = 320mm × 320mm, matrix = 320 × 320. Fat suppression technology was employed to improve the contrast of lesion visualization.

[0093] The contrast agent used was gadopentetate dimeglumine, administered via the antecubital vein at a dose of 0.2 mmol / kg body weight at a rate of 2.0 ml / s, followed by flushing the tubing with 20 ml of normal saline. Acquired images were saved in DICOM format for subsequent processing and analysis.

[0094] Step 102 Magnetic Resonance Image Preprocessing and Standardization

[0095] The acquired DCE-MRI sequences are preprocessed and standardized to improve the accuracy of subsequent analysis. This includes the following sub-steps:

[0096] (1) Motion correction: Since there may be slight body movements of the patient during the acquisition process, there may be slight spatial displacement between the images at different time points, which requires image registration. A non-rigid registration algorithm is adopted, using the baseline image as a reference, to spatially align the enhanced images at 7 time points with the baseline image to eliminate motion artifacts.

[0097] (2) Signal Intensity Standardization: Due to the influence of various factors on magnetic resonance imaging, there may be differences in signal intensity between different patients, different devices, and even the same patient at different times. Signal intensity normalization is required. The specific method is to select the ipsilateral pectoralis major muscle as the reference area and calculate the relative signal intensity value of each voxel: RSI = SI_voxel / SI_muscle, where SI_voxel is the signal intensity of the voxel and SI_muscle is the average signal intensity of the reference muscle area.

[0098] (3) Noise Removal: Gaussian filtering or median filtering is used to reduce noise in the image and improve image quality. The size of the filter kernel is dynamically adjusted according to the image resolution, and is usually set to 3×3×3 voxels.

[0099] (4) Image interpolation resampling: In order to ensure that the spatial resolution of images of different patients is consistent, all images are resampled by trilinear interpolation with a uniform voxel size of 1mm×1mm×1mm.

[0100] Through the above preprocessing steps, high-quality DCE-MRI sequence images with spatial alignment, signal normalization, and noise reduction are obtained, laying the foundation for subsequent quantitative analysis.

[0101] Step 103 Quantitative Analysis of Dynamic Contrast Enhancement Curve

[0102] Quantitative analysis of dynamic contrast enhancement curves was performed on preprocessed DCE-MRI sequences to extract time-signal intensity curve parameters. This included the following sub-steps:

[0103] (1) Time-signal intensity curve generation: For each voxel, the signal intensity values ​​at the baseline and 7 enhancement time points are extracted to construct a time-signal intensity curve. For the tumor region, the region of interest (ROI) can be determined by manual or automatic segmentation methods, and the average curve of all voxels within the ROI is extracted.

[0104] (2) Curve parameter calculation: Based on the time-signal intensity curve, the following quantitative parameters are calculated:

[0105] Relative enhancement rate (RE): RE = (SI_t - SI_0) / SI_0 × 100%, where SI_t is the signal strength at time t and SI_0 is the baseline signal strength;

[0106] Maximum Enhancement Rate (MER): max(RE);

[0107] Time to Peak (TTP): The time it takes for the signal strength to reach its peak value;

[0108] Wash-in slope (WIR): The rate of change of signal intensity from baseline to peak value;

[0109] Washout slope (WOR): The rate of change of signal strength from peak value to the last time point;

[0110] Area under the curve (AUC): The integral area under the time-signal strength curve.

[0111] (3) Curve morphology classification: Based on the morphological characteristics of the time-signal intensity curve, the curves are divided into three types:

[0112] Type I (Continuously Rising): Signal strength continues to increase, with no obvious washout;

[0113] Type II (Platform Type): Signal strength increases first and then tends to stabilize;

[0114] Type III (washed-out type): Signal strength first increases and then decreases significantly.

[0115] Through the above quantitative analysis, a set of parameters reflecting tumor angiogenesis and permeability characteristics was obtained. These parameters will be used for the subsequent construction of histological grading prediction models.

[0116] Step 104: Generate images of blood vessels surrounding the breast lump.

[0117] Imaging the vascular network surrounding a breast mass provides a basis for the detection of vascular tumor emboli. This includes the following sub-steps:

[0118] (1) Optimal enhancement phase selection: Select magnetic resonance images at the time when the contrast agent reaches its peak (usually the 2nd or 3rd time point) from 7 enhancement time points;

[0119] (2) Subtraction processing: Subtract the baseline image from the selected enhanced temporal image to obtain the subtracted image, highlighting the contrast-filled vascular structure;

[0120] (3) Maximum Intensity Projection (MIP): The subtracted 3D data is processed by maximum intensity projection to generate a 2D vascular projection image. The specific method is to search for the maximum pixel value on each ray along a preset direction (usually the anterior-posterior direction, the left-right direction, and the head-to-foot direction), and project these maximum values ​​onto a 2D plane to form a MIP image of the vascular network;

[0121] (4) Image enhancement processing: Post-processing such as contrast enhancement and sharpening is performed on the MIP image to improve the clarity of blood vessel display. Adaptive histogram equalization is used to enhance image contrast, and nonlinear enhancement algorithms are used to highlight blood vessel structures;

[0122] Through the above steps, a two-dimensional MIP image and a three-dimensional reconstruction model that clearly show the vascular network around the breast mass are obtained, laying the foundation for subsequent quantitative vascular analysis.

[0123] Step 105: Quantitative analysis of blood vessels in the ROI region surrounding the tumor thrombus / mass.

[0124] For the generated vascular images, a quantitative analysis of the multidimensional features of the blood vessels surrounding the tumor embolus is performed. This includes the following sub-steps:

[0125] (1) Determining the ROI region: With the breast mass as the center, delineate circular or spherical ROI regions with radii of 2cm and 5cm respectively, and mark the inner breast region as the third ROI;

[0126] (2) Blood vessel segmentation: In each ROI region, threshold segmentation combined with morphological operations is used to extract blood vessel structures and generate blood vessel segmentation masks;

[0127] (3) Extraction of vascular morphological features: Based on the segmentation results, the following morphological features are calculated:

[0128] Vascular density: The ratio of the number of vascular voxels within a region of interest (ROI) to the total number of voxels in the ROI.

[0129] Blood vessel diameter: The average diameter and diameter distribution of blood vessels are calculated based on a distance transformation algorithm;

[0130] Blood vessel length: The centerline of the blood vessel is extracted using a skeletonization algorithm, and the total length and average length of the blood vessel are calculated.

[0131] Vascular tortuosity: the ratio of the actual length of a blood vessel to the straight-line distance between its two ends;

[0132] Branch point density: The ratio of the number of vascular branch points within a region of interest (ROI) to the volume of the ROI;

[0133] Blood vessel course direction: Analysis of the main course directions and distribution of blood vessels based on the Hessian matrix;

[0134] Vascular connectivity: The connectivity of vascular networks is assessed through connected component analysis.

[0135] (4) Extraction of vascular texture features: Calculate the gray-level co-occurrence matrix features (such as contrast, energy, entropy, correlation, etc.), gray-level run length matrix features, and wavelet transform features of the vascular network within the ROI;

[0136] (5) Feature standardization: Z-score standardization is performed on all extracted features to eliminate the influence of differences in the dimensions of different features.

[0137] The above analysis yielded a multidimensional feature set describing the characteristics of the vascular network surrounding the tumor thrombus. These features will be used in the subsequent construction of a tumor thrombus pathology prediction model.

[0138] Step 106: Construction of a Predictive Model for Histological Grading of Tumor Thrombus

[0139] Based on the quantitative analysis results of dynamic contrast-enhanced curves, a histological grading prediction model for tumor thrombi was constructed. This specifically includes the following sub-steps:

[0140] (1) Feature selection: From the dynamic contrast enhancement curve parameters obtained in Step 103, LASSO regression, recursive feature elimination or chi-square test are used to select a subset of features that are highly correlated with histological grading.

[0141] (2) Data set division: The patient data was randomly divided into a training set and a test set in a 7:3 ratio to ensure that the distribution ratio of each histological grade (low, medium and high) was consistent in the two datasets.

[0142] (3) Model Training: Based on the training set data, a histological grading prediction model is constructed. Available machine learning algorithms include random forest, support vector machine, and XGBoost. For each algorithm, the optimal hyperparameters are determined through grid search combined with k-fold cross-validation.

[0143] (4) Model evaluation: Use test set data to evaluate model performance, calculate metrics such as accuracy, precision, recall, and F1 score, plot confusion matrix and ROC curve, and calculate AUC value;

[0144] (5) Model optimization: Based on the evaluation results, ensemble learning methods such as voting and stacking are used to optimize model performance;

[0145] (6) Model interpretation: Using SHAP value or feature importance analysis, the basis for the model prediction results is explained, and the features that have the greatest influence on histological grading prediction are identified.

[0146] Through the above steps, a model is constructed that can accurately predict the histological grade of breast cancer based on DCE-MRI sequence images, achieving a three-level classification of low (highly differentiated), moderate (moderately differentiated), and high (lowly differentiated).

[0147] Step 107: Construction of a Tumor Thrombus Pathological Prediction Model

[0148] This step constructs a pathological prediction model for tumor thrombi based on the quantitative analysis results of blood vessels in the region of origin (ROI) surrounding the tumor thrombus. Specifically, it includes the following sub-steps:

[0149] (1) Feature selection: From the vascular morphology and texture features obtained in Step 105, LASSO regression, recursive feature elimination or chi-square test are used to select a subset of features that are highly correlated with the pathological results of tumor thrombus.

[0150] (2) Data set division: The patient data was randomly divided into training set and test set in a ratio of 7:3 to ensure that the distribution ratio of negative and positive samples in the two datasets is consistent;

[0151] (3) Model Training: Based on the training set data, a pathological prediction model for tumor thrombi is constructed. Optional machine learning algorithms include random forest, support vector machine, XGBoost, and multilayer perceptron. For each algorithm, the optimal hyperparameters are determined through grid search combined with k-fold cross-validation.

[0152] (4) Model evaluation: Use test set data to evaluate model performance, calculate metrics such as accuracy, precision, recall, and F1 score, plot confusion matrix and ROC curve, and calculate AUC value;

[0153] (5) Model optimization: Based on the evaluation results, ensemble learning methods such as voting and stacking are used to optimize model performance; to address the class imbalance problem, oversampling techniques such as SMOTE or class weights are used or the class weights are adjusted.

[0154] (6) Model interpretation: The basis for the model prediction results is explained by using SHAP value or feature importance analysis, and the features that have the greatest influence on the pathological prediction of tumor thrombus are identified.

[0155] Through the above steps, a model is constructed that can accurately predict the pathological results (negative / positive) of vascular tumor thrombi based on vascular network characteristics.

[0156] Step 108 Clinical Visualization Interpretation and Optimization

[0157] Develop a clinical visualization and interpretation interface to present prediction results to clinicians in an intuitive and easy-to-understand manner. This includes the following sub-steps:

[0158] (1) Results visualization: Design a graphical user interface to simultaneously display the original magnetic resonance images, enhanced vascular images, ROI regions and prediction results, including pathological prediction results of vascular tumor thrombus (negative / positive and probability values) and histological grade prediction results (low / medium / high and probability values);

[0159] (2) Feature contribution analysis: Visualize the contribution of key features to the prediction results through radar charts, heat maps and other methods to help doctors understand the basis for prediction;

[0160] (3) Correlation analysis between prediction results and clinical parameters: Correlation analysis is performed between prediction results and patients' clinical parameters (such as age, tumor size, molecular subtype, etc.) to provide a comprehensive risk assessment;

[0161] (4) Interactive exploration: Provides an interactive interface that allows doctors to adjust the location and size of the ROI and view the changes in feature parameters and prediction results of different areas in real time;

[0162] (5) Results report generation: Automatically generate standardized examination reports, including imaging findings, quantitative analysis results, prediction results and recommended follow-up treatment suggestions;

[0163] Through the above steps, the predicted results can be presented in a clinically friendly manner, improving doctors' understanding and acceptance of AI-assisted diagnostic results and promoting the application of this technology in clinical practice.

[0164] like Figure 3 As shown, the magnetic resonance imaging system for breast cancer mass vascular tumor thrombus and histological grading provided by the present invention includes:

[0165] Image Acquisition Module: Responsible for acquiring DCE-MRI sequences of breast cancer patients, including baseline images and continuous image sequences at seven time points after contrast agent injection. This module connects to the MRI equipment and controls imaging parameter settings, acquisition procedures, and image storage. The module includes a scan parameter configuration interface, a patient information management interface, and an image acquisition control interface, and supports the import and export of DICOM format images.

[0166] Image preprocessing module: Responsible for preprocessing and standardizing the acquired magnetic resonance images, including motion correction, registration, denoising, and signal intensity standardization. This module integrates multiple image processing algorithms, supports batch processing and real-time preview, and has an automatic parameter optimization function. It can adapt to input images of different qualities to ensure standardized, high-quality output images.

[0167] The Dynamic Curve Analysis Module is responsible for performing quantitative analysis of dynamic contrast-enhanced curves on standardized DCE-MRI sequence images, extracting time-signal intensity curve parameters. This module supports automatic and manual ROI plotting, generates color parameter maps to visually display the spatial distribution of each parameter, and provides curve classification and parameter statistics functions.

[0168] The vascular imaging generation module is responsible for extracting the MRI image at the peak contrast agent concentration, subtracting the baseline image, and performing maximum intensity projection to generate enhanced images of the blood vessels surrounding the breast mass. This module provides multi-directional MIP projection and 3D vascular reconstruction functions, supporting interactive operation and real-time rendering.

[0169] The vascular feature analysis module is responsible for quantitative analysis of multi-dimensional features of blood vessels within the ROI region surrounding the tumor thrombus, including morphology, curvature, course, and branching characteristics. This module integrates a library of vascular segmentation and feature extraction algorithms, enabling it to automatically calculate various vascular morphological parameters and texture features, and generate feature data tables.

[0170] The histological grading prediction module is responsible for building and applying histological grading prediction models based on the quantitative analysis results of dynamic contrast-enhanced curves. This module includes feature selection tools, a model training platform, and a prediction engine, supports various machine learning algorithms, and provides model performance evaluation and comparison functions.

[0171] The tumor thrombus pathology prediction module is responsible for constructing and applying a tumor thrombus pathology prediction model based on the quantitative analysis results of vascular characteristics in the ROI region. Similar to the histological grading prediction module, this module focuses on binary classification problems (negative / positive) and also provides complete model training, optimization, and application functions.

[0172] Clinical Visualization Interface Module: Responsible for developing the clinical visualization and interpretation interface, displaying prediction results and key influencing factors. This module provides an intuitive graphical user interface, integrates various visualization tools, supports multi-window simultaneous display, interactive operation, and report generation. The system adopts a client / server architecture, with the server side responsible for image processing and model calculation, and the client providing a user-friendly interface that supports remote access via network.

[0173] In addition, the system includes a data management module and a system management module. The former is responsible for the storage, retrieval, and backup of patient data, while the latter is responsible for user access control, system configuration, and log recording. The system adopts a modular design, with each functional module communicating through standard interfaces, facilitating upgrades and expansion.

[0174] The validation results of this embodiment show that combining vascular features with dynamic enhancement curve features can significantly improve prediction accuracy. In the tumor thrombus pathology prediction task, the model's AUC value reached 0.89, sensitivity was 0.85, and specificity was 0.87; in the histological grading prediction task, the model's overall accuracy reached 0.81, and the accuracy rates for high, medium, and low grades were 0.83, 0.78, and 0.82, respectively.

[0175] Beneficial effects:

[0176] High-precision prediction. This invention constructs a high-precision prediction model through multi-dimensional feature extraction and advanced machine learning algorithms, capable of accurately predicting the pathological results of vascular tumor thrombi and the histological grade of tumors, providing an objective and reliable basis for clinical decision-making. Compared with traditional qualitative diagnosis, the accuracy of this method is significantly improved, especially in its greater sensitivity for detecting micro-thrombi.

[0177] Clinically friendly. The visual explanation interface developed in this invention intuitively displays the prediction results and their basis, helping doctors understand the process and principles of AI-assisted diagnosis, thus improving the system's transparency and credibility. The system's operation flow conforms to clinical work habits, making it easy for doctors to accept and use.

[0178] Multi-task integration. This invention simultaneously achieves two important clinical tasks: pathological prediction of vascular tumor thrombi and histological grading prediction, providing physicians with a comprehensive assessment of tumor characteristics and assisting in the development of individualized treatment plans. The system can be flexibly applied to multiple clinical scenarios such as preoperative assessment, treatment effect monitoring, and prognostic prediction.

[0179] Scalability and Adaptability. This invention adopts a modular design and open architecture, facilitating the expansion of system functions and the updating and iteration of algorithms. The system has good adaptability and can be customized according to the equipment conditions and clinical needs of different hospitals.

[0180] It can significantly improve the accuracy, efficiency, and standardization of diagnosis and histological grading of breast cancer vascular emboli, helping clinicians to develop more precise individualized treatment plans, improve patient prognosis, and reduce unnecessary invasive examinations and lower medical costs.

[0181] The above specific embodiments further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for identifying vascular tumor emboli and histological grading in breast cancer masses using magnetic resonance imaging, characterized in that... The grading method includes: Step S1: Acquire DCE-MRI sequence images; Step S2: Preprocess and standardize the magnetic resonance DCE-MRI sequence images; Step S3: Perform quantitative analysis of dynamic contrast enhancement curves based on the preprocessed magnetic resonance DCE-MRI sequence images, and extract time-signal intensity curve parameters; Step S4: Extract the image of the vascular network surrounding the breast mass; Step S5: Quantitatively analyze the multidimensional features of blood vessels in the area surrounding the tumor thrombus based on the vascular network image to obtain the results of the multidimensional vascular feature analysis. Step S6: Based on the time-signal intensity curve parameters, construct a tumor thrombus histological grading prediction model to predict the tumor histological grade; Step S7: Based on the results of the multidimensional vascular feature analysis, construct a tumor thrombus pathology prediction model to predict the pathological results of vascular tumor thrombus.

2. The method for identifying vascular tumor emboli and histological grading of breast cancer masses using magnetic resonance imaging according to claim 1, characterized in that, Step S1: Acquiring DCE-MRI sequence images specifically includes: Acquire baseline T1-weighted images of the patient, and images before contrast agent injection; After the contrast agent was injected, T1-weighted images were continuously acquired at 7 time points to form a dynamic enhancement sequence; Obtain the 1+7step dataset, and save the collected images in DICOM format.

3. The method for identifying vascular tumor emboli and histological grading of breast cancer masses using magnetic resonance imaging according to claim 1, characterized in that, Step S2: Preprocessing and standardizing the magnetic resonance DCE-MRI sequence images specifically includes: Motion correction: A non-rigid registration algorithm is used to spatially align the enhanced images at 7 time points with the baseline image to eliminate motion artifacts. Signal intensity standardization: Using the ipsilateral pectoralis major muscle as a reference region, the relative signal intensity value of each voxel was calculated. RSI = SI_voxel / SI_muscle, where SI_voxel is the signal intensity of the voxel and SI_muscle is the average signal intensity of the reference muscle region; Noise removal: Gaussian filtering or median filtering is used to remove noise from the image, and the size of the filter kernel is dynamically adjusted according to the image resolution. Image interpolation resampling: Trilinear interpolation resampling is performed on all images, with a uniform voxel size of 1mm×1mm×1mm.

4. The method for identifying vascular tumor emboli and histological grading of breast cancer masses using magnetic resonance imaging according to claim 1, characterized in that, Step S3: Quantitative analysis of dynamic contrast enhancement curves is performed based on the preprocessed DCE-MRI sequence images, and the extraction of time-signal intensity curve parameters specifically includes: Time-signal intensity curve generation: For each voxel, the signal intensity values ​​at the baseline and 7 enhancement time points are extracted to construct a time-signal intensity curve; for the tumor region, the region of interest (ROI) is determined by segmentation method, and the average curve of all voxels in the ROI is extracted; Curve parameter calculation: Based on the time-signal intensity curve, calculate quantitative parameters: Relative enhancement rate (RE): RE = (SI_t - SI_0) / SI_0 × 100%, where SI_t is the signal strength at time t, and SI_0 is the baseline signal strength; Maximum enhancement rate (MER): max(RE); Time to peak (TTP): Time it takes for the signal strength to reach its peak value; Wash-in slope (WIR): Rate of change of signal strength from baseline to peak value; Wash-out slope (WOR): Rate of change of signal strength from peak value to the last time point; Area under the curve (AUC): Area under the time-signal strength curve; Curve morphology classification: Based on the morphological characteristics of the time-signal intensity curve, curves are classified into three types: Type I, continuously rising type: signal strength continues to increase, with no obvious washout; Type II, Platform Type: Signal strength increases first and then tends to stabilize; Type III, wash-out type: signal strength first increases and then decreases significantly; Obtain a set of parameters that reflect tumor angiogenesis and permeability characteristics.

5. The method for identifying vascular tumor emboli and histological grading of breast cancer masses using magnetic resonance imaging according to claim 1, characterized in that, Step S4: Extracting the vascular network image around the breast mass specifically includes: Optimal enhancement phase selection: Select the magnetic resonance imaging at the moment when the contrast agent reaches its peak from 7 enhancement time points; Subtraction processing: The selected enhanced temporal image is subtracted from the baseline image to obtain a subtracted image, which highlights the contrast-filled vascular structures; Maximum Intensity Projection (MIP): The subtracted 3D data is processed by maximum intensity projection to generate a 2D vascular projection image. The maximum pixel value on each ray is searched along a preset direction, and the maximum value is projected onto a 2D plane to form an MIP image of the vascular network. Image enhancement processing: The MIP image is post-processed by contrast enhancement and sharpening. An adaptive histogram equalization method is used to enhance image contrast, and a nonlinear enhancement algorithm is used to highlight the vascular structure. Obtain two-dimensional MIP images and three-dimensional reconstructed models that clearly show the vascular network surrounding a breast mass.

6. The method for identifying vascular tumor emboli and histological grading of breast cancer masses using magnetic resonance imaging according to claim 1, characterized in that, Step S5: Based on the vascular network image, perform quantitative analysis of the multidimensional features of the blood vessels surrounding the tumor thrombus mass to obtain the multidimensional feature analysis results of the blood vessels, specifically including: ROI region determination: With the breast mass as the center, circular or spherical ROI regions with radii of 2cm and 5cm are delineated respectively, and the internal breast region is marked as a third ROI; Blood vessel segmentation: Within each ROI region, threshold segmentation combined with morphological operations is used to extract blood vessel structures and generate blood vessel segmentation masks; Vascular morphological feature extraction: Based on the segmentation results, morphological features are calculated: vessel density, vessel diameter, vessel length, vessel tortuosity, branch point density, vessel direction, and vessel connectivity. Vascular texture feature extraction: Calculate the gray-level co-occurrence matrix features, gray-level run length matrix features, and wavelet transform features of the vascular network within the ROI; Feature standardization: Perform Z-score standardization on all extracted features; Obtain a multidimensional feature set describing the characteristics of the vascular network surrounding the tumor embolus mass.

7. The method for identifying vascular tumor emboli and histological grading of breast cancer masses using magnetic resonance imaging according to claim 1, characterized in that, Step S6: Based on the time-signal intensity curve parameters, construct a tumor thrombus histological grading prediction model. Predicting tumor histological grading specifically includes: Feature selection: From the obtained dynamic contrast enhancement curve parameters, a feature subset is selected using LASSO regression, recursive feature elimination, or chi-square test methods; Dataset splitting: Patient data were randomly divided into training and test sets in a 7:3 ratio; Model training: Based on the training set data, construct a histological grading prediction model; Model evaluation: Evaluate model performance using test set data; Model optimization: Based on the evaluation results, ensemble learning methods are used to optimize model performance; Model Interpretation: Using SHAP values ​​or feature importance analysis, the basis for the model's prediction results is explained, and the features most influential on histological grading prediction are identified. A model was constructed to predict the histological grade of breast cancer based on DCE-MRI sequence images, achieving a three-level classification of low, intermediate, and high grades.

8. The method for identifying vascular tumor emboli and histological grading of breast cancer masses using magnetic resonance imaging according to claim 1, characterized in that, Step S7: Based on the results of the multidimensional vascular feature analysis, construct a tumor thrombus pathology prediction model to predict the pathology results of vascular tumor thrombi. Specifically, this includes: Feature selection: From the vascular morphological and texture features obtained in Step 105, LASSO regression, recursive feature elimination, or chi-square test were used to select a subset of features that were relevant to the pathological results of tumor thrombus. Dataset splitting: Patient data were randomly divided into training and test sets in a 7:3 ratio; Model training: Construct a pathological prediction model for tumor thrombi based on the training set data; Model evaluation: Evaluate model performance using test set data; Model optimization: Based on the evaluation results, ensemble learning methods are used to optimize model performance; SMOTE oversampling technique or class weights are adjusted. Model Interpretation: Using SHAP values ​​or feature importance analysis, the basis for the model's prediction results is explained, and the features most influential on the pathological prediction of tumor thrombi are identified. To construct a model that can accurately predict the pathological results of vascular tumor thrombi based on vascular network characteristics.

9. A magnetic resonance imaging (MRI) system for classifying vascular tumor emboli and histological grading of breast cancer masses, comprising the MRI system for classifying vascular tumor emboli and histological grading of breast cancer masses as described in any one of claims 1-8, characterized in that, The system includes: The image acquisition module is used to acquire DCE-MRI sequence images. The preprocessing module is used to preprocess and standardize the magnetic resonance DCE-MRI sequence images; The DCE-MRI curve analysis module is used to perform quantitative analysis of dynamic contrast enhancement curves based on the preprocessed magnetic resonance DCE-MRI sequence images and extract time-signal intensity curve parameters. The vascular imaging generation module is used to extract images of the vascular network surrounding a breast mass; The feature extraction module is used to perform quantitative analysis of the multidimensional features of blood vessels around the tumor thrombus mass based on the vascular network image, and obtain the results of multidimensional vascular feature analysis. The model prediction module is used to construct a tumor thrombus histological grading prediction model based on the time-signal intensity curve parameters to predict the tumor histological grade. The decision support module is used to construct a tumor thrombus pathology prediction model based on the results of the multidimensional vascular feature analysis, and to predict the pathological results of vascular tumor thrombus.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the magnetic resonance imaging method for identifying vascular tumor thrombi and histological grading of breast cancer masses as described in any one of claims 1-8.

Images