Breast cancer prediction model construction system based on MRI image and storage medium

Abstract

According to the breast cancer prediction model construction system based on the MRI image and the storage medium provided by the invention, the model for predicting the neoadjuvant chemotherapy pathological alleviation condition and prognosis of the breast cancer patient is constructed by combining the dynamic change of the blood vessel characteristics and the clinical pathological characteristics, so that the system has relatively high accuracy and clinical practicability, and is suitable for popularization and application. And making of an individualized adjuvant therapy scheme is facilitated. The method can early predict breast pathology remission and long-term prognosis, and can early and accurately predict neoadjuvant chemotherapy curative effect and long-term prognosis of breast cancer patients. The model predicts whether a patient can achieve complete relieving of breast pathology or not by analyzing changes of breast tumor blood vessel characteristics in the chemotherapy process and correlation with clinical factors, and further predicts the recurrence-free lifetime of the patient, so that a reference basis is provided for individualized treatment.

Description

Technical Field

[0001] The present invention relates to the field of medical prediction, and particularly relates to a system for constructing a breast cancer prediction model based on MRI images and a storage medium. Background Art

[0002] Breast cancer is one of the most common malignant tumors among women globally. For patients with locally advanced breast cancer, surgery after neoadjuvant chemotherapy (NAC) is the standard treatment option. Neoadjuvant chemotherapy can reduce the tumor volume before surgery, lower the surgical difficulty, and provide a valuable opportunity to evaluate the effectiveness of chemotherapy. Currently, magnetic resonance imaging (MRI) of the breast is widely used in the diagnosis and efficacy evaluation of breast cancer patients. In particular, through MRI imaging technology, the size of the tumor, lymph node status, and the response of the tumor to chemotherapy can be evaluated.

[0003] Although existing MRI technologies have been widely applied clinically, there are still certain limitations in predicting the efficacy of neoadjuvant chemotherapy, such as: (1) Traditional imaging assessment techniques often only consider the volume change of the tumor, while ignoring other biological characteristics related to tumor invasiveness.

[0004] (2) Most current models are based on the tumor characteristics of specific subtypes and lack comprehensive coverage of breast cancers of different molecular subtypes.

[0005] (3) For the early prediction of the efficacy of neoadjuvant chemotherapy, existing means are not sensitive enough, the accuracy rate is not high, and it is difficult to provide individualized treatment recommendations for patients in a timely manner.

[0006] (4) The feature extraction and analysis techniques in clinical applications are relatively complex, increasing the difficulty of clinical promotion. Summary of the Invention

[0007] The purpose of the present invention is to provide a system for constructing a breast cancer prediction model based on MRI images and a storage medium.

[0008] To solve the above problems, the present invention provides a system for constructing a breast cancer prediction model based on MRI images, including: An acquisition module, configured to acquire dynamic change data of vascular characteristics of the breast tumor of a subject and clinicopathological data; A feature selection module, configured to perform feature selection on the dynamic change data of vascular characteristics of the breast tumor of the subject and the clinicopathological data, so as to respectively screen out a first variable for training a prediction model for breast pathological remission and a second variable for training a prognosis model; A training module, configured to train a breast pathological remission prediction model based on a first variable of a subject to obtain a trained breast pathological remission prediction model; and train a prognosis model based on a second variable of the subject to obtain a trained prognosis model; respectively verify the trained breast pathological remission prediction model and the prognosis model to obtain a verified breast pathological remission prediction model and a prognosis model; An identification module, configured to obtain a first variable of a person to be detected, and input the first variable of the person to be detected into the verified breast pathological remission prediction model to obtain a prediction result of remission or non-remission of the breast tumor of the person to be detected; or, obtain a second variable of the person to be detected, and input the second variable of the person to be detected into the verified prognosis model to obtain a prediction result of high or low recurrence probability at each time point of the breast tumor of the person to be detected.

[0009] Further, in the above method, from the perspective of pathological remission, the subjects are divided into subjects with remission or non-remission of breast tumors; from the perspective of prognosis, the subjects are divided into subjects with high or low recurrence probability of breast tumors at each time point; The collection module collects clinicopathological data, including: age, height, weight, menstrual status, clinical stage, and pathological stage.

[0010] Further, in the above method, the collection module is configured to: Place the subject in the prone position, place the subject's breasts in a preset bilateral four-channel phased array breast coil, and obtain a pre-enhancement scan image sequence, including: pre-enhancement cross-sectional T1-weighted, T2-weighted imaging, fat-suppressed sagittal T2-weighted imaging, and diffusion-weighted imaging; After injecting the subject with gadopentetate dimeglumine contrast agent at a dose of 0.1 mmol / kg body weight, obtain a dynamic contrast-enhanced MR imaging sequence, including: post-enhancement cross-sectional T1-weighted, T2-weighted imaging, fat-suppressed sagittal T2-weighted imaging, and diffusion-weighted imaging; Subtract the T1-weighted imaging in the enhanced MR imaging sequence from the T1-weighted imaging in the pre-enhancement scan image sequence to obtain an image subtraction sequence; based on the image subtraction sequence, obtain a maximum intensity projection image MIP; Based on the maximum intensity projection image MIP, measure the number of blood vessels passing through the breast tumor; wherein, for each subject, an MRI scan is performed before neoadjuvant chemotherapy, after the first and second cycles of neoadjuvant chemotherapy, and after the third and fourth cycles of neoadjuvant chemotherapy respectively to obtain the corresponding maximum intensity projection image MIP, and the number of blood vessels VTL passing through the breast tumor in the corresponding maximum intensity projection image MIP obtained by each MRI scan is measured, and the number of blood vessels VTL passing through the breast tumor before neoadjuvant chemotherapy is recorded respectively1 The number of vessels VTL passing through the breast tumor after the first and second cycles of neoadjuvant chemotherapy 2 The number of vessels VTL passing through the breast tumor after the third and fourth cycles of neoadjuvant chemotherapy 3 ; Calculate ΔVTL 12 and ΔVTL 13 where ΔVTL 12 respectively represents the percentage change in the number of vessels VTL passing through the breast tumor after the first and second cycles of neoadjuvant chemotherapy 2 relative to the number of vessels VTL passing through the breast tumor before neoadjuvant chemotherapy 1 ; ΔVTL 13 represents the percentage change in the number of vessels VTL passing through the breast tumor after the third and fourth cycles of neoadjuvant chemotherapy 3 relative to the number of vessels VTL passing through the breast tumor before neoadjuvant chemotherapy 1 ; Using the surv_cutpoint function in the survminer package of R language, calculate the optimal cut-off values of the vascular characteristics corresponding to ΔVTL 12 and ΔVTL 13 respectively; Compare the calculated ΔVTL 12 and ΔVTL 13 with the corresponding optimal cut-off values respectively to determine whether it is significant. If it is significant, update the value of the corresponding ΔVTL 12 or ΔVTL 13 to 1. If it is not significant, update the value of the corresponding ΔVTL 12 or ΔVTL 13 to 0.

[0011] Furthermore, in the above method, the feature selection module is used for: Randomly divide all included subjects into non-overlapping training sets and validation sets at a ratio of 7:3; In the training set, use the LASSO feature selection algorithm and ten-fold cross-validation on the R (4.4.0) software to perform feature selection on the dynamic change data of vascular characteristics and clinicopathological factors, so as to screen out the first variable for predicting the breast pathological remission status and the second variable for predicting the prognosis model respectively. Among them, the clinicopathological factors come from the clinicopathological data, and the clinicopathological factors include age, hormone receptor status, HER2 status, T stage, Ki-67 index, lymph node status, and BMI variables; The training module is used for: Based on the validation set, the trained breast pathological remission prediction model and prognosis model are respectively validated to obtain the validated breast pathological remission prediction model and prognosis model.

[0012] Further, in the above method, the feature selection module is used for: The implementation of the LASSO feature selection algorithm depends on the glmnet package. First, all potential predictors, namely the dynamic change data of vascular features and clinicopathological factors, are transformed into matrix form through the model.matrix() function, and missing values are removed; then, the cv.glmnet() function uses ten-fold cross-validation to generate a sequence of different hyperparameters lambda (usually 100 values) for controlling the penalty strength of L1 regularization on the model coefficients; during the ten-fold cross-validation process, the best lambda.min value is determined by calculating the prediction error at each lambda, that is, the lambda when the misclassification rate is the lowest, namely lambda.min; subsequently, based on lambda.min, a LASSO regression model is constructed through the glmnet() function; based on the LASSO regression model, variable screening is performed to respectively screen out the first variables for training the breast pathological remission prediction model and the second variables for training the prognosis model.

[0013] Further, in the above method, the feature selection module is used for: In the variable screening stage, for the breast pathological remission prediction model, the LASSO regression model screens the potential predictors according to the magnitude of the coefficient values corresponding to the potential predictors, and sets the threshold thresh = 0.1, that is, retains the potential predictors with the absolute value of the coefficient greater than the threshold 0.1 as effective predictors, namely the first variables, and finally screens out 7 variables including ΔVTL 12 、age, hormone receptor status, HER2 status, clinical T stage, Ki-67 index, and BMI as the first variables for constructing the breast pathological remission prediction model.

[0014] Further, in the above method, the feature selection module is used for: In the variable screening stage, for the prognosis model, in addition to using the best penalty coefficient lambda.min selected through cross-validation for feature selection, lambda = 0.01 is additionally selected for comparative analysis, and the screening criterion is set as: when the absolute value of the regression coefficient of the potential predictor is greater than 0.5 based on the LASSO regression model, it is included in the subsequent analysis as a significant predictor, and finally ΔVTL 13、Four variables, namely age, clinical T stage, and BMI, are used as the second variables to construct a prognostic model.

[0015] Further, in the above method, the training module is used for: Using the Logistic regression analysis algorithm on R (4.4.0) software, based on the first variables corresponding to the subjects with remission or non-remission of breast tumors in the training set, construct a prediction model for breast pathological remission, and determine whether the model is trained completed through a preset convergence criterion. Among them, the implementation of the Logistic regression analysis algorithm depends on the lrm() function in the rms package to perform multivariate Logistic regression analysis. The maximum number of iterations maxit during the training of the breast pathological remission prediction model is set to 1000; during the training process of breast pathological remission prediction, the convergence of the model is monitored by setting the parameter tol = 1e-7. During each iteration, the change amount of the log-likelihood function or the change amount of the parameter estimation is automatically calculated. When the change amount is less than 1e-7, the model is considered to have converged, and the training process automatically terminates; after the model converges, a nomogram is drawn based on the nomogram() function in the rms package, and the nomogram shows the contribution of each prediction factor, that is, the first variable, to the prediction result.

[0016] Further, in the above method, the training module is used for: The training module uses the Cox regression analysis algorithm on R (4.4.0) software. The implementation of the Cox regression analysis algorithm depends on the cph() function in the rms package, which supports the fitting of survival data and performs parameter estimation through the partial likelihood function; during the model training process, the analysis is performed with recurrence-free survival as the end event; during model training, the maximum number of iterations is set to the default value of 1000; the convergence criterion is set through the tol = 1e-9 parameter; during each iteration, the Cox regression algorithm calculates the change amount of the objective function or the change amount of the parameter estimation. When the change amount is less than 1e-9, the model is considered to have converged, and the training process automatically terminates; after the model converges, a nomogram is drawn based on the Cox regression analysis results of the training set using the nomogram() function, showing the recurrence probability at different time points and providing a visual display.

[0017] According to another aspect of the present invention, there is also provided a computer-readable storage medium, on which computer-executable instructions are stored. When the computer-executable instructions are executed by a processor, the processor is caused to: execute the steps in the above-mentioned breast cancer prediction model construction system based on MRI images.

[0018] Compared with the prior art, the model proposed by the present invention combines the dynamic change data of the vascular characteristics of breast tumors extracted from MRI images (such as ΔVTL 12 and ΔVTL 13 ), and combines these characteristics with clinicopathological factors (such as age, hormone receptor status, HER2 status, T stage, Ki-67 index, and BMI) to construct a personalized prediction model. This prediction model can not only predict the breast pathological remission (bpCR) (remission means there is no breast tumor at all; or non-remission means there is still a breast tumor), but also effectively predict the long-term prognosis of patients (relapse-free survival, RFS), and shows high prediction accuracy in clinical trials. In addition, this prognostic model can divide patients into a high-risk group (high recurrence probability) and a low-risk group (low recurrence probability) according to the nomogram risk score, providing a valuable basis for prognostic stratification for clinicians, thereby helping to formulate more individualized treatment plans.

[0019] The present invention constructs a model for predicting the pathological remission and prognosis of breast cancer patients undergoing neoadjuvant chemotherapy by combining the dynamic changes of vascular characteristics and clinicopathological characteristics, which has high accuracy and clinical practicability, and helps to formulate individualized adjuvant treatment plans. It can early predict the breast pathological remission (bpCR) and long-term prognosis (RFS), providing a method for earlier and more accurate prediction of the efficacy and long-term prognosis of breast cancer patients undergoing neoadjuvant chemotherapy. This model predicts whether a patient can achieve breast pathological complete response (bpCR) by analyzing the changes in the vascular characteristics of breast tumors during chemotherapy and the association with clinical factors, and further predicts the relapse-free survival period (RFS) of the patient, thereby providing a reference basis for individualized treatment.

[0020] The two prediction models of the present invention can early and accurately predict the pathological response and long-term survival of breast cancer patients after neoadjuvant chemotherapy by combining the dynamic changes of vascular characteristics and clinicopathological characteristics during neoadjuvant chemotherapy of breast cancer extracted from MRI images, and more accurately stratify the prognosis of these patients, assisting clinicians to formulate individualized adjuvant treatment plans for patients and optimizing the treatment effect. Brief Description of the Drawings

[0021] Figure 1 It is a schematic diagram of screening markers included in the breast pathological response prediction model by LASSO regression in the training set; Figure 2 It is a nomogram of the breast pathological remission prediction model in the training set; Figure 3ROC curve diagram comparing the breast pathological remission prediction model in the training set and test set with the model without vascular features; Figure 4 Decision curve analysis diagram comparing the breast pathological remission prediction model in the training set and test set with the model without vascular features; Figure 5 Schematic diagram of screening and incorporating markers for the prognostic model through LASSO regression in the training set; Figure 6 Nomogram of the prognostic model in the training set; Figure 7 ROC curve diagram comparing the prognostic model in the training set and test set with the prediction of breast and lymph node pathological remission using tpCR; Figure 8 Kaplan–Meier survival curve diagram of RFS with risk scores divided into different risk groups calculated according to the nomogram of the prognostic model. Detailed implementation mode

[0022] The present invention will be further described in detail below with reference to the accompanying drawings.

[0023] In a typical configuration of the present application, the terminal, the device of the service network, and the trusted party all include one or more processors (CPUs), input / output interfaces, network interfaces, and memories.

[0024] The memory may include non-permanent memory in the computer-readable medium, in the form of random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. The memory is an example of the computer-readable medium.

[0025] A computer-readable medium includes permanent and non-permanent, removable and non-removable media that can implement information storage by any method or technology. The information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassette tapes, magnetic disk storage or other magnetic storage devices, or any other non-transitory medium that can be used to store information accessible by a computing device. As defined herein, a computer-readable medium does not include non-transitory computer-readable media (transitory media), such as modulated data signals and carrier waves.

[0026] The present invention provides a breast cancer prediction model construction system based on MRI images, and the system includes: An acquisition module, configured to acquire dynamic change data of vascular characteristics and clinicopathological data of breast tumors of a subject; wherein, from the perspective of pathological remission, the subject is divided into subjects with remission or non-remission of breast tumors; from the perspective of prognosis, the subject is divided into subjects with a high or low probability of breast tumor recurrence at each time point (such as a preset number of years). A feature selection module, configured to perform feature selection on the dynamic change data of vascular characteristics and clinicopathological data of breast tumors of the subject, so as to respectively screen out a first variable for training a breast pathological remission prediction model and a second variable for training a prognosis model. A training module, configured to train a breast pathological remission prediction model based on the first variable of the subject to obtain a trained breast pathological remission prediction model; and train a prognosis model based on the second variable of the subject to obtain a trained prognosis model; respectively verify the trained breast pathological remission prediction model and prognosis model to obtain a verified breast pathological remission prediction model and prognosis model. An identification module, configured to obtain the first variable of the person to be detected, and input the first variable of the person to be detected into the verified breast pathological remission prediction model to obtain a prediction result of remission or non-remission of the breast tumor of the person to be detected; or, obtain the second variable of the person to be detected, and input the second variable of the person to be detected into the verified prognosis model to obtain a prediction result of a high or low probability of recurrence of the breast tumor of the person to be detected at each time point (such as a preset number of years).

[0027] In an embodiment of the breast cancer prediction model construction system based on MRI images of the present invention, the clinical and pathological data collected by the collection module includes: age, height, weight, menstrual status, clinical stage, and pathological stage.

[0028] Specifically, the present invention can include 182 patients with locally advanced breast cancer who were histologically confirmed to have invasive breast cancer, received neoadjuvant chemotherapy, and successfully underwent radical surgery at this breast diagnosis and treatment center as subjects.

[0029] The inclusion criteria include: histologically confirmed operable unilateral breast cancer, untreated with any cancer treatment, and eligible for neoadjuvant chemotherapy.

[0030] The exclusion criteria include: pregnant or lactating patients, advanced breast cancer with distant metastases, bilateral breast cancer, a history of other malignancies, and HER2 positive but not receiving anti-HER2 targeted therapy.

[0031] The present invention can prospectively collect their clinical and pathological data, including: age, height, weight, menstrual status, clinical stage, pathological stage, etc., and retrospectively collect three breast MRI images before neoadjuvant chemotherapy, after the first and second cycles of neoadjuvant chemotherapy, and after the third and fourth cycles of neoadjuvant chemotherapy for vascular feature extraction. Among them, after the third and fourth cycles of neoadjuvant chemotherapy is the end of neoadjuvant chemotherapy.

[0032] In an embodiment of the breast cancer prediction model construction system based on MRI images of the present invention, the collection module is used to collect the dynamic change data of the vascular features of the breast tumors of the subjects, including: S11, place the subject in the prone position, place the subject's breasts in a preset bilateral four-channel phased array breast coil, and obtain a pre-enhancement scan image sequence, including: pre-enhancement cross-sectional T1-weighted, T2-weighted imaging, fat-suppressed sagittal T2-weighted imaging, and diffusion-weighted imaging; S12, after injecting the subject with gadopentetate dimeglumine contrast agent at a dose of 0.1 mmol / kg body weight, obtain a dynamic contrast-enhanced (DCE) MR imaging (Image) sequence, including: post-enhancement cross-sectional T1-weighted, T2-weighted imaging, fat-suppressed sagittal T2-weighted imaging, and diffusion-weighted imaging; Here, the contrast-enhanced (DCE) MR imaging is an enhanced image obtained by using a high-resolution isotropic volume examination sequence after injecting gadopentetate dimeglumine contrast agent at a dose of 0.1 mmol / kg body weight; S13, subtract the T1-weighted imaging in the enhanced MR imaging sequence from the T1-weighted imaging in the pre-enhancement scan image sequence to obtain an image subtraction sequence; based on the image subtraction sequence, obtain a maximum intensity projection image MIP; S14. Based on the maximum intensity projection (MIP) image, measure the number of vessels passing through the breast tumor (lesion area) (Vessel-through-lesion, VTL). For each subject, an MRI scan is performed before neoadjuvant chemotherapy, after the first and second cycles of neoadjuvant chemotherapy, and after the third and fourth cycles of neoadjuvant chemotherapy to obtain the corresponding MIP images. Measure the number of vessels VTL passing through the breast tumor in the corresponding MIP images obtained from each MRI scan, and record them as the number of vessels VTL passing through the breast tumor before neoadjuvant chemotherapy, 1 the number of vessels VTL passing through the breast tumor after the first and second cycles of neoadjuvant chemotherapy, 2 and the number of vessels VTL passing through the breast tumor after the third and fourth cycles of neoadjuvant chemotherapy. 3 ; Here, before neoadjuvant chemotherapy, the first MRI scan is performed to obtain the corresponding MIP image; after the first and second cycles of neoadjuvant chemotherapy, the second MRI scan is performed to obtain the corresponding MIP image; after the third and fourth cycles of neoadjuvant chemotherapy, the third MRI scan is performed to obtain the corresponding MIP image. Measure the number of vessels VTL passing through the breast tumor in the MIP image of the first MRI scan and record it as the number of vessels VTL passing through the breast tumor before neoadjuvant chemotherapy. 1 ; Measure the number of vessels VTL passing through the breast tumor in the MIP image of the second MRI scan and record it as the number of vessels VTL passing through the breast tumor after the first and second cycles of neoadjuvant chemotherapy. 2 ; Measure the number of vessels VTL passing through the breast tumor in the MIP image of the third MRI scan and record it as the number of vessels VTL passing through the breast tumor after the third and fourth cycles of neoadjuvant chemotherapy. 3 ; S15. Calculate ΔVTL 12 and ΔVTL, 13 where ΔVTL 12 represents the percentage change in the number of vessels VTL passing through the breast tumor after the first and second cycles of neoadjuvant chemotherapy 2 relative to the number of vessels VTL passing through the breast tumor before neoadjuvant chemotherapy; 1 ΔVTL 13 represents the percentage change in the number of vessels VTL passing through the breast tumor after the third and fourth cycles of neoadjuvant chemotherapy 3 relative to the number of vessels VTL passing through the breast tumor before neoadjuvant chemotherapy. 1Percentage change Preferably, the formulas are as follows, where ΔVTL 12 and ΔVTL 13 A positive value indicates a decrease in the number of blood vessels passing through the lesion after chemotherapy, and the larger the value, the more significant the decrease:

[0033] S16. Using the surv_cutpoint function in the survminer package of the R language, calculate the optimal cut-off values of the vascular characteristics corresponding to ΔVTL 12 and ΔVTL 13 respectively; compare the calculated ΔVTL 12 and ΔVTL 13 with the corresponding optimal cut-off values respectively to determine whether it is significant. If it is significant, update the corresponding ΔVTL 12 or ΔVTL 13 to 1. If it is not significant, update the corresponding ΔVTL 12 or ΔVTL 13 to 0.

[0034] In an embodiment of the breast cancer prediction model construction system based on MRI images of the present invention, a feature selection module is used to perform feature selection on the dynamic change data of the vascular characteristics of the breast tumors of the subjects and the clinicopathological data, so as to screen out the first variables for training the breast pathological remission situation prediction model and the second variables for training the prognosis model respectively, including: S20. Randomly divide all the included subjects into non-overlapping training sets and validation sets at a ratio of 7:3, where the validation set is used for the final test of the model; S21. In the training set, use the LASSO feature selection algorithm and ten-fold cross-validation on the R (4.4.0) software to perform feature selection on the dynamic change data of the vascular characteristics and the clinicopathological factors, so as to screen out the first variables for training the breast pathological remission situation prediction model and the second variables for training the prognosis model respectively. Among them, the clinicopathological factors come from the clinicopathological data, and the clinicopathological factors include age, hormone receptor status, HER2 status, T stage, Ki-67 index, lymph node status, and BMI variables.

[0035] In an embodiment of the breast cancer prediction model construction system based on MRI images of the present invention, S21. In the training set, use the LASSO feature selection algorithm and ten-fold cross-validation on the R (4.4.0) software to perform feature selection on the dynamic change data of the vascular characteristics and the clinicopathological factors, so as to screen out the first variables for training the breast pathological remission situation prediction model and the first variables for training the prognosis model respectively, including: The implementation of the LASSO feature selection algorithm in S211 depends on the glmnet package. First, all potential predictors, namely the dynamically changing data of vascular features and clinicopathological factors, are transformed into matrix form through the model.matrix() function, and missing values are removed. Then, the cv.glmnet() function uses ten-fold cross-validation to generate a sequence of different hyperparameters lambda (usually 100 values) to control the penalty strength of L1 regularization on the model coefficients. During the ten-fold cross-validation process, the best lambda.min value is determined by calculating the prediction error at each lambda, that is, the lambda when the misclassification rate is the lowest, namely lambda.min. Subsequently, based on lambda.min, a LASSO regression model is constructed through the glmnet() function. Based on the LASSO regression model, variable screening is performed to screen out the first variable for breast pathological remission prediction model training and the second variable for prognostic model training respectively.

[0036] In one embodiment of the breast cancer prediction model construction system based on MRI images of the present invention, in S211, based on the LASSO regression model, variable screening is performed to screen out the first variable for breast pathological remission prediction model training, including: In S2111, in the variable screening stage, for the breast pathological remission prediction model, the LASSO regression model screens potential predictors according to the magnitude of the coefficient values corresponding to the potential predictors. Set the threshold thresh = 0.1, that is, retain the potential predictors with the absolute value of the coefficient greater than the threshold 0.1 as effective predictors, namely the first variable. Finally, the screened variables include ΔVTL 12 Seven variables including age, hormone receptor status, HER2 status, clinical T stage, Ki-67 index, and BMI are used as the first variables to construct the breast pathological remission prediction model.

[0037] In one embodiment of the breast cancer prediction model construction system based on MRI images of the present invention, in S211, based on the LASSO regression model, variable screening is performed to screen the second variable for prognostic model training, including: In S2112, in the variable screening stage, for the prognostic model, in addition to using the best penalty coefficient lambda.min selected by cross-validation for feature selection, lambda = 0.01 is additionally selected for comparative analysis. The screening criterion is set as follows: when the absolute value of the regression coefficient of the potential predictor is greater than 0.5 based on the LASSO regression model, it is included in the subsequent analysis as a significant predictor. Finally, the screened variables include ΔVTL 13、Four variables, namely age, clinical T stage, and BMI, are used as the second variables to construct the prognostic model.

[0038] In an embodiment of the breast cancer prediction model construction system based on MRI images of the present invention, the training module is used to train a breast pathological remission situation prediction model based on the first variables of the subjects to obtain a trained breast pathological remission situation prediction model, including: S31. The training module constructs and trains a breast pathological remission situation prediction model based on the Logistic regression model and the first variables corresponding to the subjects with remission or non-remission of breast tumors in the training set to obtain a trained remission situation prediction model; Preferably, the training module uses the Logistic regression analysis algorithm on the R (4.4.0) software, constructs a breast pathological remission situation prediction model based on the first variables corresponding to the subjects with remission or non-remission of breast tumors in the training set, and judges whether the model is trained by a preset convergence criterion. Among them, the implementation of the Logistic regression analysis algorithm depends on the lrm() function in the rms package to perform multivariate Logistic regression analysis. The maximum number of iterations maxit for training the breast pathological remission situation prediction model is set to 1000 to ensure sufficient iteration of the algorithm; during the training process of the breast pathological remission situation prediction, the convergence of the model is monitored by setting the parameter tol = 1e-7. During each iteration, the change amount of the log-likelihood function or the change amount of the parameter estimation is automatically calculated. When the change amount is less than 1e-7, it is considered that the model has converged and the training process automatically terminates; after the model converges, the training module draws a nomogram based on the nomogram() function in the rms package, and the nomogram shows the contribution of each prediction factor, that is, the first variable, to the prediction result.

[0039] Here, in tol = 1e-7, tol is the tolerance, that is, the convergence threshold, which is used to control the convergence criterion during the model training process; 1e-7 is the scientific notation representation, equivalent to 1×10 -7 , that is, 0.0000001.

[0040] In an embodiment of the breast cancer prediction model construction system based on MRI images of the present invention, the training module is used to train a prognostic model based on the second variables of the subjects to obtain a trained prognostic model, including: S32. Construct and train a prognostic model based on the Cox model and the subjects with a high or low probability of breast tumor recurrence at each preset time limit to obtain a trained prognostic model; Preferably, the training module uses the Cox regression analysis algorithm on R (4.4.0) software. The implementation of the Cox regression analysis algorithm depends on the cph() function in the rms package. This function supports the fitting of survival data and performs parameter estimation through the partial likelihood function. During model training, the recurrence-free survival (RFS) is used as the end event for analysis. When training the model, to ensure that the algorithm can fully iterate and reach a stable state, the maximum number of iterations (maxit) is set to the default value of 1000. At the same time, to detect the convergence of the model, the algorithm sets the convergence criterion through the tol = 1e-9 parameter. During each iteration, the Cox regression algorithm calculates the change in the objective function (i.e., the log-likelihood function) or the change in parameter estimation. When the change is less than 1e-9, the model is considered to have converged, and the training process automatically terminates. After the model converges, the nomogram() function is used to draw a nomogram based on the Cox regression analysis results of the training set, showing the recurrence probability (survival probability) at different time points (such as 1 year, 3 years, and 5 years), and providing a clear visual display to assist clinical decision-making.

[0041] Here, in tol = 1e-9, tol is the tolerance, that is, the convergence threshold, which is used to control the convergence criterion during model training; 1e-9 is the scientific notation representation, equivalent to 1×10 -9 , that is, 0.000000001.

[0042] In an embodiment of the breast cancer prediction model construction system based on MRI images of the present invention, the training module verifies the trained breast pathological remission prediction model and prognosis model respectively to obtain the verified breast pathological remission prediction model and prognosis model, including: S33. The training module verifies the trained breast pathological remission prediction model and prognosis model respectively based on the validation set to obtain the verified breast pathological remission prediction model and prognosis model.

[0043] Here, in the training set, the influence degree of each prediction factor in the two models on the breast pathological response and prognosis can be shown through the nomogram. In the training set and the validation set, the prediction efficacy of the two models can be comprehensively verified through the ROC curve and its area under the curve AUC, calibration curve, and decision curve analysis (DCA). The prognostic stratification ability of the prognostic model can be verified through Kaplan-Meier analysis, and the survival differences between groups can be evaluated through the log-rank test. Statistical analysis is performed using R language (version 4.3.1), and P < 0.05 indicates statistical significance.

[0044] Known situation: Among 182 patients, a total of 72 (39.56%) achieved breast pathological complete remission (bpCR), while 110 (60.44%) did not achieve bpCR. 50 (39.37%) patients in the training set achieved bpCR, and 22 (40%) patients in the validation set achieved bpCR. In the entire study set of 182 patients, the 1-year, 3-year, and 5-year recurrence-free survival rates (RFS) were 97.2% (95% CI 94.8% - 99.6%), 86.8% (95% CI 82.0% - 92.0%), and 83.6% (95% CI 78.3% - 89.4%), respectively. The overall survival rates (OS) were 100.0% (95% CI 100.0% - 100.0%), 94.0% (95% CI 90.6% - 97.5%), and 91.7% (95% CI 87.8% - 95.8%), respectively.

[0045] Figure 1 To screen and include markers for the breast pathological response prediction model through LASSO regression in the training set, Figure 1 The abscissa of A in is the hyperparameter λ, which is the logarithmic form Log(λ) of lambda; the ordinate is the binomial deviance. Figure 1 The abscissa of B in is the hyperparameter λ, and the ordinate is the coefficient value (Coefficients) of each variable in the model.

[0046] Figure 2 For the nomogram of the breast pathological remission situation prediction model in the training set, where, Age: Age, divided into two groups of ≥35 years old and <35 years old with 35 years old as the boundary. In breast cancer research, age is an important clinical factor. The physiological status, tumor biological behavior, etc. of patients in different age groups may vary, which will affect the occurrence, development, and prognosis of the disease.

[0047] HorR status: Hormone Receptor status, divided into positive (Positive) and negative (Negative). For breast cancer patients with positive hormone receptors, endocrine therapy may be effective, and their treatment strategies and prognosis are different from those of negative patients.

[0048] HER2 status: Human epidermal growth factor receptor 2 (HER2) status, divided into positive and negative. HER2-positive breast cancer is more invasive, but anti-HER2 targeted therapy can significantly improve the prognosis of patients.

[0049] Clinical T stage: The clinical T stage, where T1 - T2 are relatively early stages and T3 - T4 are relatively late stages, reflects the size of the primary tumor and the degree of local invasion, and is related to disease progression and prognosis.

[0050] Ki67 index: The cell proliferation index, where positive represents active cell proliferation and negative represents relatively inactive proliferation. It is an important indicator for evaluating the proliferative activity and prognosis of tumor cells.

[0051] ΔVTL 12 : The percentage change in the number of blood vessels before neoadjuvant chemotherapy and after the first and second cycles of neoadjuvant chemotherapy, divided into high and low, reflecting the changes in blood vessel characteristics after chemotherapy and potentially related to the tumor's response to chemotherapy.

[0052] Points: Corresponding to different groups of each variable, corresponding scores can be assigned for calculating the total score.

[0053] Total Points: The total score axis, where the scores of each variable are added to obtain the total score, based on which the patient's condition can be evaluated.

[0054] bpCR: Represents Breast Pathological Complete Response. Its axis shows the bpCR probability corresponding to different total scores, used to predict the likelihood of a patient achieving breast pathological complete response.

[0055] Figure 3 In it, A is the ROC curve of the prediction model for breast pathological remission in the training set compared with the model without blood vessel characteristics; Figure 3 In it, B is the ROC curve of the prediction model for breast pathological remission in the validation set compared with the model without blood vessel characteristics; among them, Figure 3 The abscissa in it is 1 - Specificity, and the ordinate is Sensitivity.

[0056] Figure 4 In it, A is the decision curve analysis of the prediction model for breast pathological remission in the training set compared with the model without blood vessel characteristics; Figure 4In B, it is the decision curve analysis of the breast pathological remission prediction model in the validation set compared with the model without vascular features. Among them, the abscissa is High Risk Threshold: the high-risk threshold, which represents the probability threshold for judging a patient as high risk. It also corresponds to the Cost:Benefit Ratio below, reflecting the trade-off between the cost and benefit when judging a patient as positive (high risk); the ordinate is Standardized Net Benefit: the standardized net benefit.

[0057] Figure 5 It is the marker screened by LASSO regression for inclusion in the prognostic model in the training set. Among them, Figure 5 In A, the abscissa is Log(λ), where λ is the hyperparameter in LASSO regression; the ordinate is the Partial Likelihood Deviance. Figure 5 In B, the abscissa is Log(λ), and the ordinate is Coefficients (the coefficient values of each variable in the model, i.e., the regression coefficients), representing the coefficient values corresponding to each variable in the model.

[0058] Figure 6 It is the nomogram of the prognostic model in the training set. Among them, Age: The age is grouped with 35 years old as the boundary.

[0059] Clinical T stage: The clinical T stage is divided into T1 - T2 (relatively early stage) and T3 - T4 (relatively late stage).

[0060] BMI status: It is grouped according to whether BMI is ≥25. There is an association between BMI and the prognosis of breast cancer patients, which may involve factors such as hormone levels and metabolic status that affect tumor development.

[0061] ΔVTL 13 : The percentage change in the number of blood vessels before neoadjuvant chemotherapy and after the third and fourth cycles of neoadjuvant chemotherapy.

[0062] Points: Scores can be assigned corresponding to different groups of each variable, which are used to calculate the total score.

[0063] Total Points: The total score axis, which summarizes the scores of each variable.

[0064] Linear Predictor: The linear predictor axis, which is calculated based on the total score and is an intermediate variable of the prediction model.

[0065] 1-year RFS Probability, 3-year RFS Probability, 5-year RFS Probability: These are the probability axes for 1-year, 3-year, and 5-year recurrence-free survival (RFS), respectively.

[0066] Figure 7 These are the ROC curves comparing the prognostic model in the training set and validation set with the prediction using pathological complete response (tpCR) of the breast and lymph nodes. Figure 7 In [A], it is the ROC curve for predicting RFS using the model containing ΔVTL in the training set. 13 In [A], it is the ROC curve for predicting RFS using the model containing ΔVTL. Figure 7 In [B], it is the ROC curve for predicting RFS using the model containing ΔVTL in the validation set. 13 In [B], it is the ROC curve for predicting RFS using the model containing ΔVTL. Figure 7 In [C], it is the ROC curve for predicting RFS using only tpCR in the training set. Figure 7 In [D], it is the ROC curve for predicting RFS using only tpCR in the validation set. Figure 7 In [the figure], the horizontal axis is the False positive rate, which is the proportion of samples that are actually negative but are misjudged as positive; the vertical axis is the True positive rate, which is the proportion of samples that are actually positive and are correctly judged as positive.

[0067] Figure 8 The risk scores calculated according to the nomogram of the prognostic model are divided into different risk groups. Figure 8 In [A], it is the Kaplan–Meier survival curve of RFS plotted in the training set. Figure 8 In [B], it is the Kaplan–Meier survival curve of RFS plotted in the validation set. Figure 8 In [C], it is the Kaplan–Meier survival curve of RFS plotted in the whole set. Figure 8 Horizontal axis in [the figure]: Follow up time (m), which is the follow-up time (in months); vertical axis: Survival probability, which is the survival probability; the blue curve represents the High - risk group, and the red curve represents the Low - risk group. The p - value marked in the figure is the result of the log - rank test. Number at risk: The following table shows the remaining number of patients in each of the two strata, the High - risk group and the Low - risk group, at different follow-up time points, reflecting the change in the number of patients at risk over time.

[0068] Verification result: The breast pathological response prediction model was constructed based on the variables selected by LASSO regression, including ΔVTL 12 , age, hormone receptor status, HER2 status, clinical T stage, Ki67 index, and body mass index (as shown in Figure 1 ), presented as a nomogram (as shown in Figure 2 ). The predictive performance of the models with and without ΔVTL 12 was compared using the ROC curve (as shown in 12 ). It was found that in the training set, the AUC of the model combining ΔVTL Figure 3 with clinicopathological factors was 0.819, while that of the model based only on clinicopathological factors was 0.768 (as shown in A of 12 Figure 3 ); in the validation set, the AUCs of the two models were 0.892 and 0.857 respectively (as shown in B of Figure 3 Figure 3 ), indicating that the model containing ΔVTL 12 had better predictive ability. According to the DCA analysis, the model containing ΔVTL 12 had better clinical utility in predicting breast pathological remission (as shown in 12 Figure 4 ).

[0069] Figure 5 Figure 6 For the prognostic model, through LASSO regression analysis as shown in Figure 5 , ΔVTL 13 , age, clinical T stage, and BMI were selected as important factors for RFS prediction. The predictive model established by combining these four factors, as shown in the nomogram of Figure 6 , had AUCs of 0.745, 0.772, and 0.703 for 1-year, 3-year, and 5-year RFS in the training set, and AUCs of 0.918, 0.767, and 0.717 in the validation set, respectively, all superior to the model using overall pCR (tpCR) alone for prediction (as shown in Figure 7 ). It was also found that the prognostic models containing ΔVTL 13 showed higher clinical utility in predicting RFS. In addition, the risk scores obtained from the prognostic model nomogram were used to divide patients into high-risk and low-risk groups. The Kaplan-Meier survival curve showed that the RFS of the high-risk group was significantly lower than that of the low-risk group (P<0.0001), further verifying the prognostic stratification ability of the model (as shown in Figure 8 ).

[0070] Traditionally, MRI evaluation usually relies on changes in tumor volume and lymph node metastasis. However, with the development of imaging techniques, more parameters have been incorporated into the evaluation of the efficacy of breast cancer. Studies have shown that the dynamic changes in tumor vascular characteristics during neoadjuvant chemotherapy can provide important information for predicting efficacy. In particular, contrast-enhanced MRI can be used to further analyze the changes in blood vessels and their reflection of tumor biological behavior.

[0071] Traditional imaging evaluation techniques often only consider changes in tumor volume and ignore other biological characteristics related to tumor invasiveness, such as the dynamic changes in angiogenesis used in the present invention. Most current models are based on the tumor characteristics of specific subtypes and lack comprehensive coverage of breast cancers of different molecular subtypes. The model of the present invention can be applied to all subtypes of breast cancer. The present invention has a high prediction accuracy and can provide individualized treatment recommendations for patients in a timely manner. The feature extraction and analysis techniques in the clinical application of the present invention are simple and easy to promote clinically. The present invention combines imaging and pathological features to predict the treatment response and long-term prognosis of patients.

[0072] Breast pathological remission prediction model: This model aims to predict whether breast cancer patients can achieve breast pathological complete remission (bpCR) after receiving neoadjuvant chemotherapy. The model variables include: ΔVTL 12 (percentage change in VTL relative to baseline after two cycles of neoadjuvant chemotherapy), age, hormone receptor status, HER2 status, clinical T stage, Ki-67 index, BMI. ΔVTL 12 Combined with the above clinicopathological factors to construct an individualized prediction model. Based on the results of the retrospective analysis of the prospective cohort in this center, the AUC of this model in predicting the breast pathological remission of breast cancer patients receiving neoadjuvant chemotherapy was 0.819 in the training set and 0.892 in the validation set, both significantly better than the model using only clinicopathological factors. The calibration curve and decision curve analysis (DCA) verified the accuracy and clinical practicability of the model.

[0073] Prognosis model: This model aims to predict the recurrence-free survival (RFS) of breast cancer patients after receiving neoadjuvant chemotherapy. The model variables include: ΔVTL 13 (percentage change in VTL relative to baseline after the end of chemotherapy), age, clinical T stage, BMI. ΔVTL 13Combined with the above clinical and pathological variables, a survival prediction model was constructed. Based on the results of the retrospective analysis of the prospective cohort in this center, in the training set, the AUCs of 1-year, 3-year, and 5-year RFS of this model were 0.745, 0.772, and 0.703 respectively for breast cancer patients receiving neoadjuvant chemotherapy; in the validation set, the AUCs were 0.918, 0.767, and 0.717 respectively. This model can use the nomogram risk score to divide patients into high-risk and low-risk groups. The Kaplan-Meier survival curve verification showed that the RFS of the high-risk group was significantly lower than that of the low-risk group (P<0.0001), further verifying the prognostic stratification ability of the model.

[0074] The model proposed by the present invention combines the dynamic change data of the vascular characteristics of breast tumors extracted from MRI images (such as ΔVTL 12 and ΔVTL 13 ), and combines these characteristics with clinical and pathological factors (such as age, hormone receptor status, HER2 status, T stage, Ki-67 index, and BMI) to construct a personalized prediction model. This prediction model can not only predict the breast pathological remission (bpCR) (remission means there is no breast tumor at all; or non-remission means there is still a breast tumor), but also effectively predict the long-term prognosis of patients (relapse-free survival, RFS), and shows high prediction accuracy in clinical trials. In addition, this prognostic model can divide patients into high-risk groups (high recurrence probability) and low-risk groups (low recurrence probability) according to the nomogram risk score, providing a valuable basis for prognostic stratification for clinicians, thus helping to formulate more individualized treatment plans.

[0075] The present invention constructs a model for predicting the pathological remission and prognosis of neoadjuvant chemotherapy in breast cancer patients by combining the dynamic changes of vascular characteristics and clinical and pathological characteristics, which has high accuracy and clinical practicability, and helps to formulate individualized adjuvant treatment plans. It can early predict the breast pathological remission (bpCR) and long-term prognosis (RFS), providing a method for predicting the efficacy and long-term prognosis of neoadjuvant chemotherapy in breast cancer patients earlier and more accurately. This model predicts whether a patient can achieve breast pathological complete response (bpCR) and further predicts the relapse-free survival period (RFS) of the patient by analyzing the changes in the vascular characteristics of breast tumors during chemotherapy and their association with clinical factors, so as to provide a reference basis for individualized treatment.

[0076] By combining the dynamically changing vascular characteristics and clinicopathological features extracted from MRI images during neoadjuvant chemotherapy for breast cancer, the two prediction models of the present invention can accurately predict the pathological response and long-term survival of breast cancer patients after neoadjuvant chemotherapy at an early stage, and more accurately stratify the prognosis of these patients, assisting clinicians in formulating individualized adjuvant treatment plans for patients and optimizing the treatment effect.

[0077] According to another aspect of the present invention, there is also provided a computer-readable storage medium having computer-executable instructions stored thereon, wherein when the computer-executable instructions are executed by a processor, the processor is caused to: execute the steps in the breast cancer prediction model construction system based on MRI images described in any one of the above.

[0078] For the detailed content of each device embodiment of the present invention, reference may be specifically made to the corresponding parts of each method embodiment, which will not be elaborated herein.

[0079] Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalent technologies, the present application is also intended to include these changes and modifications.

[0080] It should be noted that the present invention can be implemented in software and / or a combination of software and hardware. For example, it can be implemented using an application-specific integrated circuit (ASIC), a general-purpose computer, or any other similar hardware device. In one embodiment, the software program of the present invention can be executed by a processor to implement the steps or functions described above. Similarly, the software program of the present invention (including related data structures) can be stored in a computer-readable recording medium, such as a RAM memory, a magnetic or optical drive, or a floppy disk and the like. Additionally, some steps or functions of the present invention can be implemented using hardware, for example, as a circuit that cooperates with the processor to execute each step or function.

[0081] In addition, a part of the present invention can be applied as a computer program product. For example, computer program instructions, when executed by a computer, can call or provide the methods and / or technical solutions according to the present invention through the operation of the computer. The program instructions for calling the methods of the present invention may be stored in a fixed or removable recording medium, and / or transmitted through a data stream in a broadcast or other signal-bearing medium, and / or stored in the working memory of a computer device operating according to the program instructions. Here, an embodiment according to the present invention includes a device that includes a memory for storing computer program instructions and a processor for executing the program instructions, wherein when the computer program instructions are executed by the processor, the device is triggered to operate based on the methods and / or technical solutions according to multiple embodiments of the present invention described above.

[0082] For those skilled in the art, it is obvious that the present invention is not limited to the details of the above-described exemplary embodiments, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention. Therefore, in any aspect, the embodiments should be regarded as exemplary and non-limiting. The scope of the present invention is defined by the appended claims rather than the above description. Therefore, all changes falling within the meaning and scope of the equivalent elements of the claims are intended to be embraced within the present invention. Any reference signs in the claims should not be construed as limiting the claims concerned. In addition, it is obvious that the word "comprising" does not exclude other elements or steps, and the singular does not exclude the plural. The multiple elements or devices recited in the apparatus claims can also be implemented by one element or device through software or hardware. First, second, etc. are used to denote names and do not denote any particular order.

Claims

1. A breast cancer prediction model construction system based on MRI images, characterized in that: include: An acquisition module is used to collect data on dynamic changes in vascular characteristics and clinical pathological information of breast tumors of subjects; A feature selection module, used to perform feature selection in the dynamic change data of vascular characteristics of breast tumors of subjects and clinical pathological data, so as to respectively screen out a first variable for training a breast pathology remission prediction model and a second variable for training a prognosis model; A training module, used for training a breast pathology remission prediction model based on a first variable of the subject to obtain a trained breast pathology remission prediction model; and training a prognosis model based on a second variable of the subject to obtain a trained prognosis model; and verifying the trained breast pathology remission prediction model and prognosis model respectively to obtain a verified breast pathology remission prediction model and prognosis model; The identification module is used to obtain a first variable of the person to be tested, and input the first variable of the person to be tested into a verified breast pathology remission prediction model to obtain a prediction result of remission or non-remission of the breast tumor of the person to be tested; or, obtain a second variable of the person to be tested, and input the second variable of the person to be tested into a verified prognosis model to obtain a prediction result of a high probability of recurrence or a low probability of recurrence of the breast tumor of the person to be tested at each time point.

2. The breast cancer prediction model construction system based on MRI images according to claim 1, characterized in that: From the perspective of pathological remission, the subjects are divided into subjects with or without remission of breast tumor; from the perspective of prognosis, the subjects are divided into subjects with high probability of breast tumor recurrence or low probability of breast tumor recurrence at each time point; The clinical pathological data collected by the collection module include: age, height, weight, menstrual status, clinical stage and pathological stage.

3. The breast cancer prediction model construction system based on MRI images according to claim 1, characterized in that: The acquisition module is used for: The subjects were placed in a prone position, and both breasts of the subjects were placed in a preset bilateral four-channel phased array breast coil to obtain a pre-enhancement scanning image sequence, including: pre-enhancement transverse T1-weighted, T2-weighted imaging, fat-suppressed sagittal T2-weighted imaging, and diffusion-weighted imaging; After the subjects were injected with gadopentetate dimeglumine contrast agent, dynamic contrast-enhanced MR imaging sequences were obtained, including enhanced transverse T1-weighted, T2-weighted imaging, fat-suppressed sagittal T2-weighted imaging, and diffusion-weighted imaging; The T1-weighted imaging in the enhanced MR imaging sequence is subtracted from the T1-weighted imaging in the scanning image sequence before enhancement to obtain an image subtraction sequence; based on the image subtraction sequence, a maximum intensity projection image MIP is obtained; Based on the maximum intensity projection image MIP, the number of blood vessels passing through the breast tumor is measured; wherein, for each subject, an MRI scan is performed before neoadjuvant chemotherapy, after the first and second cycles of neoadjuvant chemotherapy, and after the third and fourth cycles of neoadjuvant chemotherapy to obtain the corresponding maximum intensity projection image MIP, and the number of blood vessels VTL passing through the breast tumor in the corresponding maximum intensity projection image MIP obtained by each MRI scan is measured, and recorded as the number of blood vessels passing through the breast tumor before neoadjuvant chemotherapy VTL1, the number of blood vessels passing through the breast tumor after the first and second cycles of neoadjuvant chemotherapy VTL2, and the number of blood vessels passing through the breast tumor after the third and fourth cycles of neoadjuvant chemotherapy VTL3; Calculation of ΔVTL 12 and ΔVTL 13 , where ΔVTL 12 ΔVTL represents the percentage change of the number of blood vessels passing through the breast tumor after the first and second cycles of neoadjuvant chemotherapy (VTL2) relative to the number of blood vessels passing through the breast tumor before neoadjuvant chemotherapy (VTL1); 13 It represents the percentage change of the number of blood vessels passing through the breast tumor VTL3 after the third and fourth cycles of neoadjuvant chemotherapy relative to the number of blood vessels passing through the breast tumor VTL1 before neoadjuvant chemotherapy; The surv_cutpoint function in the survminer package of R language was used to calculate ΔVTL 12 and ΔVTL 13 The best threshold value of the corresponding vascular characteristics; the calculated ΔVTL 12 and ΔVTL 13 Compare with the corresponding optimal threshold to determine whether it is significant. If significant, update the corresponding ΔVTL 12 or ΔVTL 13 The value is updated to 1. If it is not significant, update the corresponding ΔVTL 12 or ΔVTL 13 The value of is 0.

4. The breast cancer prediction model construction system based on MRI images according to claim 3, characterized in that: The feature selection module is used to: All included subjects were randomly divided into non-overlapping training and validation sets in a ratio of 7:3; In the training set, the LASSO feature selection algorithm and ten-fold cross validation are used to perform feature selection on the dynamic change data of vascular characteristics and clinical pathological factors, so as to respectively screen out a first variable for training a breast pathology remission prediction model and a second variable for training a prognosis model, wherein the clinical pathological factors are from the clinical pathological data, and the clinical pathological factors include age, hormone receptor status, HER2 status, T stage, Ki-67 index, lymph node status and BMI variables; The training module is used to: Based on the validation set, the trained breast pathology remission prediction model and prognosis model are respectively validated to obtain the validated breast pathology remission prediction model and prognosis model.

5. The breast cancer prediction model construction system based on MRI images according to claim 4, characterized in that: The feature selection module is used to: The implementation of the LASSO feature selection algorithm relies on the glmnet package. First, all potential predictors, i.e., the dynamic change data of vascular characteristics and clinical pathological factors, are converted into matrix form through the model.matrix() function, and missing values ​​are removed. Then, the cv.glmnet() function uses ten-fold cross validation to generate sequences corresponding to different hyperparameters lambda, which are used to control the penalty intensity of L1 regularization on model coefficients. In the ten-fold cross validation process, the optimal lambda.min value is determined by calculating the prediction error under each lambda, i.e., the lambda with the lowest misclassification rate, i.e., lambda.min. Subsequently, the LASSO regression model was constructed based on lambda.min using the glmnet() function. Based on the LASSO regression model, variable screening was performed to select the first variable for training the breast pathology remission prediction model and the second variable for training the prognostic model.

6. The breast cancer prediction model construction system based on MRI images according to claim 5, characterized in that: The feature selection module is used to: In the variable screening stage, for the breast pathology remission prediction model, the LASSO regression model screened the potential predictors according to the size of the coefficient values ​​corresponding to the potential predictors, and set the threshold thresh = 0.1, that is, the potential predictors with absolute values ​​of coefficient values ​​greater than the threshold 0.1 were retained as effective predictors, i.e., the first variable, and finally the ΔVTL 12 Seven variables including age, hormone receptor status, HER2 status, clinical T stage, Ki-67 index and BMI were used as the first variable to construct a breast pathology remission prediction model.

7. The breast cancer prediction model construction system based on MRI images according to claim 5, characterized in that: The feature selection module is used to: In the variable screening stage, for the prognostic model, in addition to the optimal penalty coefficient lambda.min selected by cross-validation for feature selection, lambda = 0.01 was additionally selected for comparative analysis. The screening criteria were set as follows: when the absolute value of the regression coefficient of the potential predictor based on the LASSO regression model was greater than 0.5, it was included in the subsequent analysis as a significant predictor, and ΔVTL was finally screened out. 13、 A total of four variables, including age, clinical T stage and BMI, were used as the second variable to construct the prognostic model.

8. The breast cancer prediction model construction system based on MRI images according to claim 1, characterized in that: The training module is used to: A breast pathology remission prediction model is constructed based on the first variable corresponding to the subjects with or without remission of breast tumors in the training set by using a Logistic regression analysis algorithm, and a preset convergence standard is used to determine whether the model training is complete, wherein the implementation of the Logistic regression analysis algorithm relies on the lrm() function in the rms package to perform multi-factor Logistic regression analysis, and the maximum number of iterations maxit is set to 1000 during the training of the breast pathology remission prediction model; during the training process of breast pathology remission prediction, the convergence of the model is monitored by setting the parameter tol=1e-7, and during each iteration, the change in the log-likelihood function or the change in the parameter estimate is automatically calculated, and when the change is less than 1e-7, the model is considered to have converged and the training process is automatically terminated; after the model converges, a nomogram is drawn based on the nomogram() function in the rms package, and the nomogram shows the contribution of each predictor, i.e., the first variable, to the prediction result.

9. The breast cancer prediction model construction system based on MRI images according to claim 1, characterized in that: The training module is used to: The training module uses the Cox regression analysis algorithm, and the implementation of the Cox regression analysis algorithm relies on the cph() function in the rms package, which supports the fitting of survival data and performs parameter estimation through the partial likelihood function; during the model training process, recurrence-free survival is used as the outcome event for analysis; during model training, the maximum number of iterations is set to the default value of 1000; the convergence criterion is set by the tol=1e-9 parameter; During each iteration, the Cox regression algorithm calculates the change in the objective function or the change in the parameter estimate. When the change is less than 1e-9, the model is considered to have converged and the training process is automatically terminated. After the model converges, the nomogram() function is used to draw a nomogram based on the Cox regression analysis results of the training set, showing the recurrence probability at different time points and providing a visual display.

10. A computer-readable storage medium having computer-executable instructions stored thereon, wherein: When the computer executable instruction is executed by a processor, the processor is caused to: execute the steps in the system for building a breast cancer prediction model based on MRI images as described in any one of claims 1 to 9.

Images