Method and system for visually identifying hemodynamic spatial-temporal heterogeneity in breast tumor

Abstract

The invention discloses a method and a system for visually identifying hemodynamic spatial-temporal heterogeneity in breast tumors, and belongs to the technical field of medical image acquisition and auxiliary detection. The method for visually recognizing the hemodynamic spatial-temporal heterogeneity in the breast tumor comprises the steps that S1, DCE-MRI images of breast benign and malignant tumors are obtained, and region-of-interest sketching is carried out; s2, carrying out TIC automatic classification and coding on each voxel in the tumor by adopting 19 types of TIC curves to obtain a TIC color coding graph; s3, calculating the TIC type and proportion of each focus, and extracting a TIC color coding map to reflect radiomics characteristics of tumor internal heterogeneity; s4, screening the image omics characteristics of the TIC color coding image, and establishing three models including 19 types of TIC curve fusion models; according to the method, the difference of hemodynamics time and space heterogeneity in the breast tumor of the patient can be visually judged and compared, and benign and malignant and biological behaviors of the tumor of the patient are judged.

Description

Technical Field

[0001] The present invention belongs to the technical field of medical image acquisition and auxiliary detection, and in particular relates to a method and system for visually identifying the spatiotemporal heterogeneity of hemodynamics inside a breast tumor. Background Art

[0002] Breast cancer is a tumor with high biological behavior heterogeneity, with different histological subtypes, grades, tissue biological markers, proliferation states, gene mutations and immune microenvironments. Tumor microenvironment heterogeneity is an important cause of tumor treatment resistance and poor patient prognosis. The main driver of inter-tumor and intra-tumor heterogeneity is the potential heterogeneity of microvessels. Specifically, immature neovascularization and various pathological changes such as hemorrhage, cystic degeneration, necrosis or calcification lead to hemodynamic inhomogeneity, which requires a non-invasive tool to quantify and visualize the heterogeneity of "whole tumor" microvasculature.

[0003] Magnetic resonance imaging (MRI) has the advantages of being non-invasive, having high soft tissue resolution, and multi-directional and multi-sequence imaging. It plays an increasingly important role in the diagnosis and treatment of breast cancer. In particular, dynamic contrast enhanced MRI (DCE-MRI) has been widely studied and applied in the diagnosis, molecular typing, and prognosis of breast cancer. DCE-MRI reflects the hemodynamics and other information of the tumor by continuously monitoring the changes in magnetic resonance signals caused by the contrast agent entering and exiting the tumor. Previous studies mainly used qualitative analysis, semi-quantitative analysis, or model-based quantitative analysis to extract hemodynamic information from DCE-MRI images to reflect hemodynamic heterogeneity. These methods all have a limitation. Qualitative analysis cannot analyze the entire tumor, but can only analyze the local area of ​​the tumor. It is easily affected by the randomness of the region of interest (ROI) selection, the average effect of the signal, and the inconsistency between observers. It also cannot reflect the spatial heterogeneity of hemodynamics within the tumor. Semi-quantitative analysis is also easily affected by ROI selection bias, the average effect of the signal, and the inconsistency of the definition. In addition, model-based quantitative analysis methods can obtain quantitative physiological information of tissues and lesions, but current analyses are based on ideal models (such as the two-chamber model), and due to the complexity of the model, it is easily affected by MRI scanning parameters, baseline T1 values ​​of tissues before contrast agent injection, arterial input functions, and various artifacts, noise, deformation, etc., making it difficult to apply in clinical practice. Summary of the invention

[0004] The embodiment of the present invention provides a method and system for visually identifying the spatiotemporal heterogeneity of hemodynamics inside breast tumors, aiming to solve the problem that the qualitative analysis of dynamic enhanced magnetic resonance imaging in the diagnosis, molecular typing and prognosis of breast cancer cannot analyze the entire tumor, is easily affected by the randomness of the selection of the region of interest, the average effect of the signal, the inconsistency between observers, and cannot reflect the spatial heterogeneity of hemodynamics inside the tumor. The present invention realizes non-invasive, convenient and comprehensive visualization of the spatiotemporal heterogeneity of hemodynamics inside tumors through dynamic enhanced magnetic resonance imaging combined with artificial intelligence machine learning. With this method, the temporal and spatial differences of hemodynamics inside the patient's tumor can be effectively evaluated, so as to more accurately judge the benign and malignant nature and biological behavior of the tumor, and provide a scientific basis for prognosis evaluation and the formulation of treatment plans.

[0005] The technical solution of the present invention is achieved in this way:

[0006] A method for visually identifying the spatiotemporal heterogeneity of hemodynamics within breast tumors, comprising:

[0007] S1. Obtain DCE-MRI images of benign and malignant breast tumors and delineate the region of interest on the dynamic enhancement phase 3 images;

[0008] S2. Using voxels as units, 19 types of TIC (time-intensity-curve) curves are used to automatically classify and encode each voxel in the tumor, generate a 3D color coding map containing TIC information, and obtain a TIC color coding map;

[0009] S3. Based on the TIC color coding map, the types and proportions of TICs in each lesion were calculated. The percentage of voxels with the same TIC in all voxels (composition ratio) represented a feature set, and the imaging omics features of the TIC color coding map reflecting the internal heterogeneity of the tumor were extracted;

[0010] S4. Screen the radiomics features of TIC color coding images, and construct 19 types of TIC curve traditional feature models, 19 types of TIC curve radiomics models, and 19 types of TIC curve fusion models based on the screened features.

[0011] The dynamic enhanced phase 3 radiomics model, 19-type TIC curve traditional feature model, 19-type TIC curve radiomics model, and 19-type TIC curve fusion model obtained by the present invention can judge the differences in the time and spatial heterogeneity of internal hemodynamics of breast tumors of patients, and judge the benign or malignant tumor and biological behavior of the patient.

[0012] Further, in step S2,

[0013] The 19 types of TIC curves are as follows: according to the inflow rate (4 types), outflow rate (3 types) and outflow stability (2 types), the TIC spectrum is divided into 19 different subtypes [i.e., 3 (slow, medium, fast) × 3 (continuous, plateau, declining) × 2 (stable outflow, unstable) + 1 (no enhancement)] to cover all 19 types of TIC morphological types.

[0014] The inflow rates are classified into four categories: no intensification, slow, moderate and rapid; among which: I0 and I p represent the signal intensity value before enhancement and the enhancement peak value within the three phases, T P It is the time required for the signal intensity to reach the peak within the three phases; in terms of classification, when the inflow rate is <0.1, it is no enhancement; when the inflow rate = 0.1-0.5, it is slow; when the inflow rate = 0.5-1.0, it is medium speed; when the inflow rate >1.0, it is fast.

[0015] The outflow rate is divided into three categories: continuous, plateau and declining; among them, I last and I p They represent the signal intensity value of the last period of enhancement and the peak value of enhancement within the three periods respectively. In terms of classification, when the outflow rate>0.05, it is continuous; when the outflow rate=-0.05~0.05, it is a platform; when the outflow rate<-0.05, it is a decrease.

[0016] The outflow stability is divided into two categories: stable and unstable. Among them, the outflow stability uses the residual sum of squares to calculate the outflow fluctuation degree. I∈{p,…,n}, p is T P The corresponding period number, n is the number of the last period, I i is the signal strength of the i-th phase, f(I i ) is I i The linear prediction signal strength value; in terms of category classification, when the outflow stability is <0.1, it is stable; when the outflow stability is ≥0.1, it is unstable.

[0017] In step S2, the generation of a 3D color coding map containing TIC information includes the following steps: normalizing the images before and after enhancement on a voxel-by-voxel basis, generating a TIC subtype spatial mapping 3D coding map for each 3D lesion target area in units of voxels, and obtaining a TIC color coding map.

[0018] Preferably, in step S2, the software used for the normalization process is preferably Python software.

[0019] Furthermore, in step S3, it also includes:

[0020] S301, extracting radiomic features of 3D coding maps of spatial mapping of 19 TIC subtypes and dynamically enhanced radiomic features of phase 3, 1130 radiomic features were obtained, including 107 original features, 279 logarithmic features and 744 wavelet features. Among them, the original features were further divided into 14 shape features, 18 first-order features and 75 texture features; the logarithmic features were obtained by calculating the sigma value (1.0mm, 3.0mm, 5mm) of 18 first-order features and 75 texture features; the wavelet features were obtained from the first-order features and texture features by combining high-pass filters and low-pass filters in each of the three dimensions (93×8=744).

[0021] Preferably, step S301 is implemented on a python platform, using Pyradiomics package 11 to perform the extraction operation.

[0022] Further, in step S4,

[0023] The screening TIC color-coded image radiomics features include:

[0024] S401, using the hold-out scheme, the data set was divided into a training set and a test set in a ratio of 7:3; independent sample student-t test was used to screen out the features with statistical differences between benign and malignant tumors (P<0.05);

[0025] S402, using the least absolute shrinkage and selection operator (LASSO) regression algorithm to eliminate irrelevant features to prevent overfitting and enhance interpretability; the Lasso algorithm determines the feature with the most information based on the lambda (λ) value (tuning parameter) with the smallest deviation; the Lasso coefficient is used to evaluate the impact of each feature on the classification performance. The Lasso coefficient is an independent filtering model to avoid interference with the classification learning algorithm.

[0026] In step S4, it also includes:

[0027] S411, using support vector machine (SVM) to train machine learning models based on different feature sets;

[0028] S412, a 10-fold cross-validation method was used to ensure that model performance could be compared between different models;

[0029] S413, Receiver operative characteristics (ROC) curve analysis was used to evaluate the model performance, and the DeLong test was used to compare the area under the curve (AUC).

[0030] In step S4, the 19-type TIC curve fusion model is a fusion model of the 19-type TIC curve traditional features and the screened 19-type TIC curve radiomics features.

[0031] Furthermore, in step S4, it also includes: extracting radiomics features directly based on the target area delineated in the third phase of dynamic enhancement, and constructing an enhanced third phase radiomics model.

[0032] A system for visually identifying the spatiotemporal heterogeneity of hemodynamics within breast tumors, comprising:

[0033] An image acquisition module, used for acquiring DCE-MRI images;

[0034] A delineation module is used to perform delineation processing on the image acquired by the image acquisition module;

[0035] The classification and coding module uses 19 types of TIC curves to automatically classify and encode each voxel in the tumor image and generate a TIC color coding map;

[0036] The statistical extraction module calculates the type and proportion of TIC for each lesion based on the TIC color coding map, counts the percentage of voxels with the same TIC in all voxels to represent a feature set, and extracts the imaging omics features of the TIC color coding map that reflect the internal heterogeneity of the tumor;

[0037] Screening module, which screens the radiomics features of TIC color-coded images;

[0038] Model building module, based on the screened features, 19 types of TIC curve traditional feature models, 19 types of TIC curve imaging omics models, and 19 types of TIC curve fusion models were constructed;

[0039] The visualization module is used to display images and model results of internal heterogeneity of breast tumors.

[0040] Compared with the prior art, the technical solution of the present invention has the following beneficial effects:

[0041] The present invention provides a method and system for visually identifying the spatiotemporal heterogeneity of hemodynamics within a breast tumor, including the extraction of imaging genomics features related to tumor hemodynamic heterogeneity, interpretation and verification of the tumor hemodynamic heterogeneity phenotype based on imaging genomics, and analysis and establishment of the clinical value of the imaging genomics heterogeneity phenotype, thereby achieving non-invasive, convenient, and comprehensive visual identification of the temporal and spatial heterogeneity of hemodynamics within the tumor. The present invention can determine the difference in the temporal and spatial heterogeneity of hemodynamics within the patient's tumor, thereby better determining the benign or malignant nature of the patient's tumor and the biological behavior, and providing a scientific basis for prognosis evaluation and formulation of treatment plans. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0043] Figure 1 is a flow chart of a method for visually identifying spatiotemporal heterogeneity of hemodynamics inside a breast tumor in an embodiment of the present invention;

[0044] Figure 2 19 types of TIC definitions and color coding diagrams in an embodiment of the present invention;

[0045] Figure 3 It is a comparison of 19 types of TIC feature sets of benign and malignant tumors in the embodiments of the present invention;

[0046] Figure 4 This is a diagram verifying the diagnostic value of benign and malignant tumor hemodynamic heterogeneity in 19 types of TIC in the embodiment of the present invention.

[0047] Figure 5 Schematic diagram of the functional modules of the system for visually identifying the spatiotemporal heterogeneity of hemodynamics inside breast tumors in an embodiment of the present invention. DETAILED DESCRIPTION

[0048] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention. In addition, the technical solutions between the various embodiments can be combined with each other, but they must be based on the ability of ordinary technicians in this field to implement them. When the combination of technical solutions is contradictory or cannot be implemented, it should be deemed that such a combination of technical solutions does not exist and is not within the scope of protection required by the present invention.

[0049] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art to which this application belongs. The terms used herein are only for the purpose of describing the embodiments of this application and are not intended to limit this application.

[0050] At present, the existing dynamic enhanced magnetic resonance imaging in the diagnosis, molecular typing and prognosis of breast cancer has the problem that the qualitative analysis cannot analyze the entire tumor, is easily affected by the randomness of the selection of the region of interest, the average effect of the signal, the inconsistency between observers, and cannot reflect the spatial heterogeneity of the hemodynamics inside the tumor. In order to solve the above technical problems, the present invention proposes a method and system for visually identifying the spatiotemporal heterogeneity of the hemodynamics inside the breast tumor.

[0051] Example 1

[0052] like Figure 1 and Figure 2 As shown, a method for visually identifying the spatiotemporal heterogeneity of hemodynamics inside a breast tumor comprises the following steps:

[0053] S1, obtain DCE-MRI images of breast tumor patients and delineate the region of interest on the dynamic enhancement phase 3 images.

[0054] In this embodiment, the region of interest was delineated by two experienced attending physicians in the Department of Radiology using ITK-SNAP software to manually delineate the breast tumor ROI layer by layer on the third phase images of dynamic enhanced MRI; based on the delineated tumor target area, Python software was used to normalize the pixel values ​​of the pre-enhancement image and the post-enhancement images.

[0055] S2, using Python software, each segmented lesion was classified and encoded voxel by voxel based on 19 types of TIC curves, generating a new 3D encoding atlas of TIC subtype spatial mapping.

[0056] 19 types of TIC classification and proportion traditional features were extracted, and the time signal curves of the lesions were automatically classified based on the 19 types of TIC curves, and the category and corresponding proportion of the TIC curve of the 3D target area of ​​each lesion were automatically calculated. The percentage of voxels with the same TIC curve type to all voxels in the lesion represents a feature set.

[0057] S3, Radiomic features based on the dynamic enhancement phase 3 target map and the newly generated 3D coding atlas of 19 TIC subtypes were extracted using Pyradiomics package 11 on an in-house python platform. A total of 1130 radiomic features were obtained, including raw features (n = 107), logarithmic features (n = 279), and wavelet features (n = 744). Raw features were further divided into shape features (n = 14), first-order features (n = 18), and texture features (n = 75). Logarithmic features were obtained by calculating sigma values ​​(1.0 mm, 3.0 mm, 5 mm) from first-order features and texture features ((18 + 75) × 3 = 279). Wavelet features were obtained from first-order features and texture features by combining high-pass filters and low-pass filters in each of the three dimensions (93 × 8 = 744).

[0058] S41, feature screening, using the hold-out scheme, the data set was divided into a training set and a test set in a ratio of 7:3. First, the independent sample student-t test was used to screen out the features with statistical differences between benign and malignant (P<0.05). Then, the least absolute shrinkage and selection operator (LASSO) regression algorithm was used to eliminate irrelevant features to prevent overfitting and enhance interpretability. The Lasso algorithm determines the feature with the most information based on the lambda (λ) value (tuning parameter) with the smallest deviation. The Lasso coefficient is used to evaluate the impact of each feature on the classification performance. The Lasso coefficient is an independent filtering model to avoid interference with the classification learning algorithm.

[0059] For the 19-type TIC radiomics features, 767 features were selected by t-test. Then, 15 features (12 texture features and 3 first-order features) were selected by Lasso regression (|coefficient|>0.02). For the radiomics features of dynamic enhancement phase 3, 823 and 17 radiomics features (5 first-order features, 2 shape features, and 10 texture features) were selected by t-test and Lasso regression analysis (|coefficient|>0.02), respectively. In terms of feature importance ranking, the top five features with the highest absolute coefficients were all texture features that were logarithmically transformed (n=3) and wavelet transformed (n=2) for the type-19 radiomics feature set. In contrast, the top five features with the highest absolute coefficients included 4 first-order features (wavelet transformed [n=3] and logarithmically transformed [n=1]) and 1 logarithmically transformed texture feature of the dynamic enhancement phase 3 radiomics feature set.

[0060] like Figure 3 As shown in Figure 2, when the hemodynamic heterogeneity of benign and malignant tumors is compared, benign lesions have only a few major TIC types, while malignant lesions show a greater variety of TICs and a more even distribution in the spectrum of breast lesions. This suggests that the hemodynamic heterogeneity of malignant tumors is greater than that of benign tumors.

[0061] S42, based on the selected features, a dynamic enhanced phase 3 radiomics model, a 19-category TIC curve traditional feature model, a 19-category TIC curve radiomics model, and a 19-category TIC curve fusion model (19-category TIC curve traditional features + screened 19-category TIC curve radiomics features) were constructed. The diagnostic value of the enhanced phase 3 radiomics model, the 19-category TIC feature set model, the 19-category TIC radiomics model, and the fusion model of the two for the differentiation of benign and malignant tumors was evaluated. The machine learning models based on different feature sets were trained using support vector machines (SVM), and a 10-fold cross-validation method was used to ensure that the model performance could be compared between different models.

[0062] Through the analysis of the clinical value of 19 types of TIC imaging, such as Figure 4 As shown in the figure, in terms of differentiating benign and malignant lesions, both 10-fold cross-validation and independent tests showed that the 19-category TIC radiomics model and the joint model had high diagnostic efficacy. This to a certain extent verifies the spatial distribution heterogeneity of TIC profiles quantified by radiomics, and provides a new method for simultaneously quantifying the temporal and spatial hemodynamic heterogeneity within breast lesions, which is of great value.

[0063] Example 2

[0064] The detailed steps of this method are described in detail below in conjunction with specific embodiments:

[0065] 1. First, dynamic contrast-enhanced magnetic resonance images of 428 breast cancer patients who were treated at the Shenzhen Hospital of the Chinese Academy of Medical Sciences were retrospectively collected, with a total of 639 lesions. The tumor area was delineated using ITK-SNAP software, and image preprocessing was performed using Python software.

[0066] 2. Using the Python platform, we performed voxel-by-voxel TIC classification and coding on the target lesions, generated a TIC space mapping color coding atlas, calculated the TIC type and proportion of each target lesion, and extracted radiomics features based on the TIC color coding atlas. In addition, we extracted radiomics features based on the dynamically enhanced phase 3 images, and included a total of 1,130 different radiomics features.

[0067] 3. The independent sample student-t test was used to screen out the features with statistical differences between benign and malignant tumors (P<0.05). Then, the least absolute shrinkage and selection operator (LASSO) regression algorithm was used to eliminate irrelevant features to prevent overfitting and enhance interpretability, and finally 15 radiomic features were included. For the radiomic features in the third stage, 823 and 17 radiomic features (5 first-order features, 2 shape features, and 10 texture features) were selected respectively by t-test and Lasso regression (|coefficient|>0.02).

[0068] 4. Based on the selected features, the dynamic enhanced phase 3 radiomics model, the 19-category TIC curve traditional feature model, the 19-category TIC curve radiomics model, and the 19-category TIC curve fusion model (19-category TIC curve traditional features + screened 19-category TIC curve radiomics features) were constructed. The diagnostic value of the model for distinguishing benign and malignant tumors was evaluated, and support vector machine (SVM) was used to train machine learning models based on different feature sets.

[0069] 5. Among the 19 types of TIC signatures, the proportions of TIC subtypes 1, 2, and 19 in benign lesions were significantly higher than those in malignant lesions, while the proportions of TIC subtypes 9, 10, 11, 12, 15, 16, 17, and 18 were significantly lower than those in malignant lesions. Benign lesions have only a few TIC types, while malignant lesions have more TIC types and are more evenly distributed, confirming that the heterogeneity of malignant tumors is greater than that of benign lesions.

[0070] 6. In terms of distinguishing benign and malignant lesions, both 10-fold cross-validation and independent tests showed that the 19-category TIC joint model was significantly superior to the dynamic enhancement phase 3 imaging model (AUC = 0.906 vs. 0.823, P < 0.001; AUC = 0.867 vs. 0.762, P = 0.026). However, in 10-fold cross-validation and independent tests, there was no significant difference between the dynamic enhanced phase 3 radiomics model and the 19-category TIC curve traditional feature model (AUC=0.823 vs. 0.809, P=0.577, AUC=0.76211), there was no significant difference between the dynamic enhanced phase 3 radiomics model and the 19-category TIC curve radiomics model (AUC=0.823 vs. 0.854, P=0.182, AUC=0.762 vs. 0.816, P=0.200), and there was no significant difference in the diagnostic efficacy between the 19-category TIC curve traditional feature model and the 19-category TIC curve radiomics (AUC=0.854 vs. 0.809, P=0.073; AUC=0.816 vs. 0.844, P=0.454).

[0071] Example 3

[0072] like Figure 5 As shown, a system for visually identifying the spatiotemporal heterogeneity of hemodynamics inside a breast tumor, used in the method for visually identifying the spatiotemporal heterogeneity of hemodynamics inside a breast tumor, comprises:

[0073] An image acquisition module, used for acquiring DCE-MRI images;

[0074] A delineation module is used to perform delineation processing on the image acquired by the image acquisition module;

[0075] The classification and coding module uses 19 types of TIC curves to automatically classify and encode each voxel in the tumor image and generate a TIC color coding map;

[0076] The statistical extraction module calculates the type and proportion of TIC for each lesion based on the TIC color coding map, counts the percentage of voxels with the same TIC in all voxels to represent a feature set, and extracts the imaging omics features of the TIC color coding map that reflect the internal heterogeneity of the tumor;

[0077] Screening module, which screens the radiomics features of TIC color-coded images;

[0078] Model building module, based on the screened features, 19 types of TIC curve traditional feature models, 19 types of TIC curve imaging omics models, and 19 types of TIC curve fusion models were constructed;

[0079] The visualization module is used to display images and model results of internal heterogeneity of breast tumors.

[0080] Compared with the prior art, the technical solution of the present invention has the following beneficial effects:

[0081] The present invention provides a method and system for visually identifying the spatiotemporal heterogeneity of hemodynamics within a breast tumor, including the extraction of imaging genomics features related to tumor hemodynamic heterogeneity, interpretation and verification of imaging genomics-based tumor dynamic heterogeneity phenotypes, and analysis and establishment of the clinical value of imaging genomics heterogeneity phenotypes, thereby achieving non-invasive, convenient, and comprehensive visual identification of the temporal and spatial heterogeneity of hemodynamics within a tumor. The present invention can be used to determine the differences in the temporal and spatial heterogeneity of hemodynamics within a patient's tumor, thereby better determining the benign or malignant nature of the patient's tumor and the biological behavior, and providing a scientific basis for prognosis evaluation and formulation of treatment plans.

[0082] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for visually identifying the spatiotemporal heterogeneity of hemodynamics within a breast tumor, characterized by: include: S1. Obtain DCE-MRI images of benign and malignant breast tumors and delineate the region of interest on the dynamic enhancement phase 3 images; S2. Taking voxels as units, 19 types of TIC curves are used to automatically classify and encode each voxel in the tumor, generate a 3D color coding map containing TIC information, and obtain a TIC color coding map; S3. Based on the TIC color coding map, the type and proportion of TIC in each lesion were calculated. The percentage of voxels with the same TIC in all voxels represented a feature set, and the imaging features of the TIC color coding map reflecting the internal heterogeneity of the tumor were extracted; S4. Screen the radiomics features of TIC color coding images, and construct 19 types of TIC curve traditional feature models, 19 types of TIC curve radiomics models, and 19 types of TIC curve fusion models based on the screened features.

2. The method for visually identifying the spatiotemporal heterogeneity of hemodynamics within a breast tumor according to claim 1, characterized in that: In step S2, The 19 types of TIC curves are as follows: the TIC spectrum is divided into 19 different subtypes according to the inflow rate, outflow rate and outflow stability to cover all 19 types of TIC morphological types.

3. The method for visually identifying the spatiotemporal heterogeneity of hemodynamics within a breast tumor according to claim 2, characterized in that: The inflow rate is divided into four categories: no intensification, slow, medium and fast; among which: I0 and I p represent the signal intensity value before enhancement and the enhancement peak value within the three phases, T P It is the time required for the signal intensity to reach the peak within the three phases; in terms of classification, when the inflow rate is <0.1, it is no enhancement; when the inflow rate = 0.1-0.5, it is slow; when the inflow rate = 0.5-1.0, it is medium speed; when the inflow rate >1.0, it is fast.

4. The method for visually identifying the spatiotemporal heterogeneity of hemodynamics within a breast tumor according to claim 2, characterized in that: The outflow rate is divided into three categories: continuous, plateau and declining; among them, I last and I p They represent the signal intensity value of the last period of enhancement and the peak value of enhancement within the three periods respectively. In terms of classification, when the outflow rate>0.05, it is continuous; when the outflow rate=-0.05~0.05, it is a platform; when the outflow rate<-0.05, it is a decrease.

5. The method for visually identifying the spatiotemporal heterogeneity of hemodynamics within a breast tumor according to claim 2, characterized in that: The outflow stability is divided into two categories: stable and unstable. Among them, the outflow stability uses the residual sum of squares to calculate the outflow fluctuation degree. p is T P The corresponding period number, n is the number of the last period, I i is the signal strength of the i-th phase, f(I i ) is I i The linear prediction signal strength value; in terms of category classification, when the outflow stability is <0.1, it is stable; when the outflow stability is ≥0.1, it is unstable.

6. The method for visually identifying the spatiotemporal heterogeneity of hemodynamics within a breast tumor according to claim 1, characterized in that: In step S2, the generation of a 3D color coding map containing TIC information includes the following steps: normalizing the images before and after enhancement on a voxel-by-voxel basis, generating a TIC subtype spatial mapping 3D coding map for each 3D lesion target area in units of voxels, and obtaining a TIC color coding map.

7. The method for visually identifying the spatiotemporal heterogeneity of hemodynamics within a breast tumor according to claim 1, characterized in that: In step S3, it also includes: S301, extracted the radiomics features of the 3D coding atlas of spatial mapping of 19 TIC subtypes and the radiomics features of the dynamic enhancement phase 3, and obtained 1130 radiomics features, including 107 original features, 279 logarithmic features and 744 wavelet features.

8. The method for visually identifying the spatiotemporal heterogeneity of hemodynamics within a breast tumor according to claim 1, characterized in that: In step S4, The screening TIC color-coded image radiomics features include: S401, using the flow-out method, the data set was divided into a training set and a test set in a ratio of 7:3; independent sample student-t test was used to screen out the features with statistical differences between benign and malignant tumors; S402, using the least absolute shrinkage and selection operator regression algorithm to eliminate irrelevant features to prevent overfitting and enhance interpretability; the Lasso algorithm determines the feature with the most information based on the tuning parameters with the smallest deviation; and the Lasso coefficient is used to evaluate the impact of each feature on the classification performance.

9. The method for visually identifying the spatiotemporal heterogeneity of hemodynamics within a breast tumor according to claim 1, characterized in that: In step S4, it also includes: S411, using support vector machines to train machine learning models based on different feature sets; S412, a 10-fold cross-validation method was used to ensure that model performance could be compared between different models; S413, receiver operating characteristic curve analysis was used to evaluate model performance, and the DeLong test was used to compare the areas under the curve.

10. A system for visually identifying the spatiotemporal heterogeneity of hemodynamics within breast tumors, characterized by: The method for visually identifying the spatiotemporal heterogeneity of hemodynamics within a breast tumor as claimed in any one of claims 1 to 9 comprises: An image acquisition module, used for acquiring DCE-MRI images; A delineation module is used to perform delineation processing on the image acquired by the image acquisition module; The classification and coding module uses 19 types of TIC curves to automatically classify and encode each voxel in the tumor image and generate a TIC color coding map; The statistical extraction module calculates the type and proportion of TIC for each lesion based on the TIC color coding map, counts the percentage of voxels with the same TIC in all voxels to represent a feature set, and extracts the imaging omics features of the TIC color coding map that reflect the spatiotemporal heterogeneity of hemodynamics within the tumor; Screening module, which screens the radiomics features of TIC color-coded images; Model building module, based on the screened features, 19 types of TIC curve traditional feature models, 19 types of TIC curve imaging omics models, and 19 types of TIC curve fusion models were constructed; The visualization module is used to display images and model results of the spatiotemporal heterogeneity of hemodynamics within breast tumors.

Images