Method and system for predicting tumor EPR effect based on MR radiomics

Abstract

The invention discloses a method and system for predicting a tumor EPR effect based on MR radiomics, and the method comprises the steps: collecting MR image data of a tumor patient and EPR effect evaluation gold standard data, carrying out the image denoising, normalization and registration preprocessing, segmenting a tumor region through a U-Net model, and extracting basic and advanced radiomics features, thereby achieving the prediction of the tumor EPR effect. A recursive feature elimination (RFE) method and other screening features are utilized to construct a machine learning prediction model of a support vector machine (SVM) and the like, and model parameters are optimized through cross validation. According to the method, the tumor EPR effect can be effectively predicted in a non-invasive mode, a reliable basis is provided for personalized treatment, and it is verified that the prediction model accuracy reaches up to 85.2%, the sensitivity reaches up to 87.1%, and the specificity reaches up to 84.0%.

Description

Technical Field

[0001] The present invention relates to the technical field of computer-aided diagnosis, and in particular to a method and system for predicting the EPR effect of tumors based on MR imaging omics. Background Art

[0002] The high permeability and long retention effect (EPR effect) of tumors refers to the physiological phenomenon that due to the abnormal structural and functional characteristics of tumor tissue blood vessels, macromolecular substances (such as nanomedicines) can more easily penetrate into tumor tissues and stay there for a long time. This phenomenon is particularly common in solid tumors, such as breast cancer, lung cancer, colorectal cancer, liver cancer, and pancreatic cancer. The EPR effect is of great significance in the diagnosis of tumors, the formulation of treatment plans, and the development of new drugs, providing key information support for the realization of personalized treatment of tumors.

[0003] At present, the evaluation of the EPR effect of tumors mainly relies on invasive tissue biopsy methods. Although tissue biopsy can provide more accurate information on the EPR effect, as an invasive examination method, it will bring certain pain and risks to patients, and its test results only represent the situation of local tissues, and it is difficult to fully reflect the state of the overall EPR effect of the tumor. In addition, the imaging principles of traditional imaging techniques, such as CT and MRI, are mainly based on the physical properties of tissues, such as density and relaxation time. However, the EPR effect involves the complex physiological and pathological changes of tumor blood vessels. The information provided by these traditional imaging methods lacks a direct and specific correlation with the EPR effect, so it is difficult to accurately and effectively evaluate the EPR effect.

[0004] With the continuous development of medical imaging technology, radiomics has emerged as an emerging technology that can extract rich quantitative features from medical imaging data and deeply explore and analyze potential imaging information. In other related research fields of tumors, such as tumor classification and prognosis assessment, radiomics has demonstrated its unique advantages and value. However, to date, there is no effective method and system based on MR radiomics that can accurately and reliably predict the EPR effect of tumors, which has become a technical problem that needs to be solved urgently in this field. Summary of the invention

[0005] The purpose of this section is to summarize some aspects of the embodiments of the present invention and briefly introduce some preferred embodiments. Some simplifications or omissions may be made in this section and the specification abstract and the invention title of this application to avoid blurring the purpose of this section, the specification abstract and the invention title, and such simplifications or omissions cannot be used to limit the scope of the present invention.

[0006] Therefore, the object of the present invention is to provide a method and system for predicting the enhanced permeability and retention (EPR) effect of tumors based on MR radiomics, so as to solve the deficiencies in the existing methods for evaluating the EPR effect of tumors, and to accurately predict the EPR effect of tumors in a non-invasive manner, providing a more reliable basis for the personalized treatment of tumors.

[0007] To solve the above technical problems, according to one aspect of the present invention, the following technical solutions are provided:

[0008] A method for predicting the EPR effect of tumors based on MR radiomics, the steps are as follows:

[0009] S1. Data collection: Collect MR image data of tumor patients, and at the same time collect the gold standard data for evaluating the EPR effect of the corresponding tumors, and the patients from whom the data is collected meet the pre-set inclusion criteria;

[0010] S2. Image preprocessing: Perform denoising, normalization, and registration operations on the collected MR image data in sequence;

[0011] S3. Tumor region segmentation: Use the U-Net model based on deep learning, train the corresponding model with the labeled tumor image data, and after adjusting the model parameters, input the MR image to be segmented into the trained model to obtain the segmentation result;

[0012] S4. Feature extraction: Use the Pyradiomics toolkit to extract radiomics features from the segmented tumor region, and the radiomics features include basic features and advanced features;

[0013] S5. Feature screening: Adopt a feature selection algorithm to screen out the most representative feature subset from the extracted radiomics features;

[0014] S6. Model construction: Use the screened feature subset to construct a prediction model for the EPR effect of tumors by using a machine learning algorithm;

[0015] S7. Model training and verification: Divide the collected data into a training set and a verification set according to a ratio of 7:3, and use the training set to train and verify the prediction model to obtain a trained prediction model;

[0016] S8. Prediction application: For new tumor patients, obtain their MR image data, and after processing according to the above steps S1-S5 in sequence, input the obtained features into the trained prediction model, and output the prediction result of the EPR effect of the tumor, and the prediction result is divided into two types: high EPR effect and low EPR effect.

[0017] As a preferred embodiment of the method for predicting the EPR effect of tumors based on MR imaging omics according to the present invention, in step S1, the MR imaging data sequence includes one or more of T1-weighted images, T2-weighted images, and diffusion-weighted imaging sequences;

[0018] The gold standard data for evaluation is to quantitatively analyze the concentration of the active ingredient of the nanodrug in the tumor through tissue biopsy;

[0019] The inclusion criteria include, but are not limited to, tumor type, stage, age range, and whether the patient has received other chemotherapy or radiotherapy.

[0020] As a preferred embodiment of the method for predicting the EPR effect of tumors based on MR imaging omics according to the present invention, in step S2, the specific method steps of denoising, normalization, and registration operations are as follows:

[0021] Use the Gaussian filtering algorithm for denoising to remove image noise interference;

[0022] Use the linear transformation method to normalize the image gray value to the interval [0,1];

[0023] Use the rigid registration algorithm to register MR images of different sequences to ensure that the positions of the tumors are consistent on different images.

[0024] As a preferred embodiment of the method for predicting the EPR effect of tumors based on MR imaging omics according to the present invention, in step S4, the basic features include intensity features, shape features, 2D shape features, gray level co-occurrence matrix features, gray level run length matrix features, gray level zone size matrix features, and adjacent gray matrices;

[0025] The advanced features include any one or more of the intensity features, gray level co-occurrence matrix features, gray level run length matrix features, and gray level zone size matrix features extracted again after wavelet transform.

[0026] As a preferred embodiment of the method for predicting the EPR effect of tumors based on MR imaging omics according to the present invention, in step S5, when the feature selection algorithm is specifically operated, if the recursive feature elimination (RFE) combined with the support vector machine (SVM) is used, first train an SVM model using all features, calculate the importance score of each feature, and then gradually delete the feature with the lowest importance score and retrain the SVM model until the model performance no longer improves.

[0027] As a preferred embodiment of the method for predicting the EPR effect of tumors based on MR imaging omics according to the present invention, in step S5, the feature selection algorithm is one or more of the recursive feature elimination method and the least absolute shrinkage and selection operator.

[0028] As a preferred embodiment of the method for predicting the EPR effect of tumors based on MR radiomics according to the present invention, in step S6, the machine learning algorithm is one or more of support vector machine, random forest, and neural network.

[0029] As a preferred embodiment of the method for predicting the EPR effect of tumors based on MR radiomics according to the present invention, in step S7, during the training process, the model parameters are adjusted by the cross-validation method to make the model reach the best performance, and the trained model is verified using the validation set to evaluate the prediction accuracy, sensitivity, and specificity indexes of the model.

[0030] A system for realizing the prediction of the EPR effect of tumors based on MR radiomics includes:

[0031] A data acquisition module, which is used to collect MR image data of tumor patients and the corresponding gold standard data for evaluating the EPR effect of tumors, obtain MR image data by connecting to the PACS, and collect the gold standard data from the patient medical record system;

[0032] An image preprocessing module, which is written in Python language and uses the OpenCV and Nibabel libraries to preprocess the collected MR image data, including denoising, normalization, and registration operations;

[0033] A tumor region segmentation module, which is implemented based on the deep learning frameworks TensorFlow or PyTorch, and uses a pre-trained U-Net model to segment the tumor region from the preprocessed MR images;

[0034] A feature extraction module, which uses the Pyradiomics toolkit to extract radiomics features from the segmented tumor region, and the radiomics features include basic features and advanced features;

[0035] A feature screening module, which uses the recursive feature elimination method and the least absolute shrinkage and selection operator feature selection algorithm to screen the extracted radiomics features based on the Scikit-learn library to obtain the most representative feature subset;

[0036] A model construction module, which uses the screened feature subset and adopts one or more machine learning algorithms such as support vector machine, random forest, and neural network to construct a tumor EPR effect prediction model, and is implemented based on the Scikit-learn library;

[0037] A model training and verification module, which divides the collected data into a training set and a validation set, trains and verifies the prediction model, evaluates the model performance, and uses the cross-validation method to adjust the model parameters during the training process to obtain the best performance;

[0038] The prediction application module processes the MR image data of a new patient through each module in sequence, inputs the obtained features into the trained prediction model, and outputs the prediction result of the tumor EPR effect. The prediction result is displayed to the doctor through a graphical user interface.

[0039] Compared with the prior art, the beneficial effects of the present invention are as follows: The present invention extracts quantitative features from MR images, excavates potential information related to the tumor EPR effect, and realizes accurate prediction. It has been verified that the accuracy rate of the prediction model on the validation set is 85.2%, the sensitivity is 87.1%, and the specificity is 84.0%, providing a basis for personalized tumor treatment. This method is non-invasive, avoiding the pain and risks of tissue biopsy, and can obtain the overall information of the tumor, overcoming the limitations of traditional evaluation methods. The prediction model is trained and verified with a large amount of data, having high accuracy, strong sensitivity, good specificity, and stable prediction. At the same time, the device provided by the present invention has a clear structure, clear module division of labor, realizes an automated prediction process, and improves work efficiency and prediction accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the present invention will be described in detail below with reference to the drawings and detailed embodiments. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts. Among them:

[0041] Figure 1 It is a model training flowchart of a method for accurately predicting the high-permeability and long-retention effect of tumors based on MR imagingomics provided by an embodiment of the present invention;

[0042] Figure 2 It is a confusion matrix of the prediction model in the test set provided by an embodiment of the present invention;

[0043] Figure 3 It is an ROC curve of the prediction model in the test set provided by an embodiment of the present invention;

[0044] Figure 4 It is a system structure diagram for realizing the prediction of the tumor EPR effect based on MR imagingomics provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0045] In order to make the above objects, features, and advantages of the present invention more obvious and understandable, the specific embodiments of the present invention will be described in detail below with reference to the drawings.

[0046] The present invention provides a method and system for predicting the enhanced permeability and retention (EPR) effect of tumors based on MR radiomics, so as to solve the deficiencies in the existing methods for evaluating the EPR effect of tumors, and achieve accurate prediction of the EPR effect of tumors in a non-invasive manner, providing a more reliable basis for the personalized treatment of tumors.

[0047] To achieve the object of the present invention, breast cancer is selected as the tumor type for training in this embodiment, and the training process is as Figure 1 shown, and the implemented steps include the following links:

[0048] S1. Data collection:

[0049] Collect the MR image data of 520 breast cancer patients who received albumin-bound paclitaxel as a neoadjuvant chemotherapy regimen, including T1-weighted images, T2-weighted images, and diffusion-weighted imaging (DWI) sequences;

[0050] Among them, the inclusion criteria are: aged 18 or above; the patient has complete MRI image data before puncture or surgery, including T1-weighted images, T2-weighted images, and diffusion-weighted imaging (DWI) sequences; the patient must be pathologically diagnosed with breast cancer by puncture or surgery. The patient has not received any other chemotherapy or radiotherapy before joining the study; the patient receives albumin-bound paclitaxel as part of the neoadjuvant chemotherapy regimen.

[0051] The exclusion criteria are: EPR effect evaluation: lack of necessary MRI image sequences (T1-weighted image, T2-weighted image, DWI); the paclitaxel drug concentration in the postoperative tumor specimen cannot be quantitatively analyzed by high-performance liquid chromatography (HPLC); there are underlying diseases such as hypertension and diabetes that may interfere with the delivery of albumin-bound paclitaxel; pregnant or lactating women.

[0052] Meanwhile, collect the postoperative tumor specimens of these patients, and use high-performance liquid chromatography (HPLC) to quantitatively analyze the paclitaxel drug concentration as the gold standard data for EPR effect evaluation.

[0053] The MR imaging parameters are as follows: All patients underwent MRI examinations in the prone position on a 1.5T scanner, and the following imaging protocol was used: (I) Axial turbo inversion recovery magnitude [repetition time (TR) / echo time (TE), 5,000 / 61 ms; field of view (FOV), 340 mm × 340 mm; matrix, 576 × 403; flip angle, 80°; slice thickness, 4 mm]; (II) Axial diffusion-weighted imaging (b value, 50 / 800 s / mm 2; TR / TE, 5,400 / 86 ms; FOV, 360 mm × 180 mm; matrix, 192 × 82; flip angle, 180°; slice thickness, 4 mm); (III) Dynamic contrast-enhanced sequence (TR / TE, 4.23 / 1.57 ms; FOV, 340 mm × 340 mm; matrix, 448 × 296; slice thickness, 1 mm; flip angle, 10°; pixel resolution, 1.1 × 0.8 × 0.9 mm 3 ; temporal resolution, 1 minute); and (IV) Sagittal fat-suppressed T2-weighted imaging (TR / TE, 3,000 / 72 ms; FOV, 340 mm × 340 mm; matrix, 269 × 384; flip angle, 80°; slice thickness, 4 mm). Gadopentetate dimeglumine was injected via an automatic injector at a dose of 0.1 mmol / kg and a rate of 3 mL / s, followed by an injection of 20 mL of normal saline.

[0054] S2, Image preprocessing:

[0055] Denoising: The Gaussian filtering algorithm was used to denoise the MR images, with the standard deviation of the Gaussian kernel set to 1.0 to remove the noise interference in the images and improve the image quality.

[0056] Normalization: The gray values of the images were normalized to the range [0, 1] using the formula $I_{norm}=\frac{I - I_{min}}{I_{max}-I_{min}}$, where $I$ is the gray value of the original image, and $I_{min}$ and $I_{max}$ are the minimum and maximum gray values in the image respectively, to make the image data of different patients comparable.

[0057] Registration: The rigid registration algorithm was used to register the MR images of different sequences. The T1-weighted image was used as the reference image, and the T2-weighted image and the DWI sequence images were registered to the T1-weighted image to ensure that the position of the tumor was consistent in different images.

[0058] S3, Tumor region segmentation:

[0059] The U-Net model based on deep learning was used to segment the tumor region in the preprocessed MR images. First, 150 cases of labeled breast cancer tumor image data were collected as the training set, and the U-Net model was trained with the number of epochs set to 50, the learning rate set to 0.001, and the Adam optimizer used. After training, the MR images to be segmented were input into the trained U-Net model to obtain the segmentation results of the tumor region.

[0060] S4, Feature extraction:

[0061] Extract radiomics features from the segmented tumor regions using the Pyradiomics toolkit, and extract them according to the preset parameters to ensure the consistency of feature extraction.

[0062] S5. Feature screening:

[0063] Use the recursive feature elimination method (RFE) combined with support vector machine (SVM) to screen the extracted radiomics features. During the screening process, set the kernel function of SVM to a linear kernel and the penalty parameter C to 1.0. After multiple iterations of screening, finally obtain the most representative feature subset.

[0064] S6. Model construction:

[0065] Use the random forest (RF) algorithm to construct a tumor EPR effect prediction model, and set the number of decision trees to 100 and the maximum depth to 10.

[0066] S7. Model training and validation:

[0067] Divide the collected data into a training set and a validation set according to a ratio of 7:3. Train the random forest model on the training set. During the training process, adjust the model parameters through the 5-fold cross-validation method. After training, use the validation set to validate the trained model, and evaluate the prediction accuracy, sensitivity, specificity and other indicators of the model. As Figure 2 Shown is the confusion matrix of the prediction model in the test set. After multiple experiments, the accuracy of the obtained prediction model on the validation set reaches 85.2%, the sensitivity is 87.1%, and the specificity is 84.0%. As Figure 3 Shown is the ROC curve of the prediction model in the test set.

[0068] S8. Prediction application:

[0069] For new tumor patients, obtain their MR image data, and perform preprocessing, tumor region segmentation, feature extraction and screening according to the above steps. Input the obtained feature subset into the trained random forest model, and output the prediction results of the tumor EPR effect. The prediction results are divided into two types: high EPR effect and low EPR effect.

[0070] To implement the above method, the present invention also provides a system for predicting the tumor EPR effect based on MR radiomics. As Figure 4 Shown, this system includes the following modules:

[0071] Data acquisition module: Establish a connection with the hospital's PACS system to obtain the MR image data of tumor patients in real time. At the same time, extract the corresponding gold standard data for tumor EPR effect evaluation from the hospital's electronic medical record system to ensure the accuracy and integrity of the data.

[0072] Image preprocessing module: This module is implemented in the form of Python scripts. It uses the Gaussian filtering function in the OpenCV library to denoise images and uses the Nibabel library to normalize and register images. During actual operation, it automatically executes the corresponding preprocessing process according to the input MR image data.

[0073] Tumor region segmentation module: Build a U-Net model based on the TensorFlow deep learning framework and save the trained model parameters. In actual applications, load the pre-trained model and segment the tumor region of the input preprocessed MR image, and output the segmented tumor region image.

[0074] Feature extraction module: Call the interface function of the Pyradiomics toolkit to extract various radiomics features from the segmented tumor region image and store the extracted features in the form of a data table for subsequent feature screening and model construction.

[0075] Feature screening module: Implement feature selection algorithms such as recursive feature elimination (RFE) and least absolute shrinkage and selection operator (LASSO) based on the Scikit-learn library. During actual operation, screen the extracted radiomics features according to the algorithm selected by the user and the set parameters, and output the most representative feature subset.

[0076] Model construction module: Use machine learning algorithms such as random forest, support vector machine, and neural network in the Scikit-learn library to construct a tumor EPR effect prediction model. Users can select different algorithms or algorithm combinations according to actual needs and set corresponding model parameters, such as the number of decision trees in the random forest, the kernel function and penalty parameter of the support vector machine, etc.

[0077] Model training and validation module: This module implements the functions of data division, model training and validation. During the training process, automatically adjust the model parameters according to the cross-validation method and parameters set by the user to obtain the best performance. After validation, output various evaluation indicators of the model, such as accuracy, sensitivity, specificity, etc.

[0078] Prediction application module: Process the MR image data of a new patient through the previous modules in sequence. After obtaining the feature subset, input it into the trained prediction model. The prediction results are displayed to the doctor through a graphical user interface (GUI), and the prediction results of the tumor EPR effect are intuitively displayed on the interface, which is convenient for doctors to make diagnoses and treatment decisions.

[0079] It should be understood that the above-mentioned technical means can be implemented separately in the form of hardware or software, or through the integration of both. Therefore, the method and apparatus involved in the present invention, as well as several of its constituent elements or aspects, can be built into a physical medium, such as a floppy disk, CD-ROM, hard disk, or program code (i.e., a series of instructions) carried by any other machine-readable storage medium. When this program is loaded onto a device such as a computer and executed, the device is transformed into a tool for implementing the present invention.

[0080] Although the present invention has been described above with reference to the embodiments, various improvements can be made thereto and components thereof can be replaced with equivalents without departing from the scope of the present invention. In particular, as long as there is no structural conflict, the features in the embodiments disclosed in the present invention can be combined with each other in any manner, and the exhaustive description of these combinations is not given in this specification only for the sake of saving space and resources. Therefore, the present invention is not limited to the specific embodiments disclosed herein, but includes all technical solutions falling within the scope of the claims.

Claims

1. A method for predicting tumor EPR effect based on MR imaging, characterized in that: Here are the steps: S1. Data collection: Collect MR imaging data of tumor patients and the corresponding gold standard data for tumor EPR effect evaluation, and the patients whose data are collected meet the pre-set inclusion criteria; S2, image preprocessing: performing denoising, normalization and registration operations on the acquired MR image data in sequence; S3. Tumor region segmentation: A U-Net model based on deep learning is used to train the corresponding model using labeled tumor image data. After adjusting the model parameters, the MR image to be segmented is input into the trained model to obtain the segmentation result. S4. Feature extraction: The Pyradiomics toolkit is used to extract radiomics features from the segmented tumor area, wherein the radiomics features include basic features and advanced features; S5. Feature screening: Using feature selection algorithm, the most representative feature subset is selected from the extracted radiomics features; S6. Model construction: Using the screened feature subset, a machine learning algorithm was used to construct a tumor EPR effect prediction model; S7, model training and verification: the collected data is divided into a training set and a verification set in a ratio of 7:3, and the training set is used to train and verify the prediction model to obtain a trained prediction model; S8. Prediction application: For new tumor patients, obtain their MR imaging data, process them in sequence according to the above steps S1-S5, input the obtained features into the trained prediction model, and output the prediction results of the tumor EPR effect. The prediction results are divided into two types: high EPR effect and low EPR effect.

2. The method for predicting tumor EPR effect based on MR imaging according to claim 1, characterized in that: In step S1, the MR image data sequence includes one or more of T1-weighted image, T2-weighted image and diffusion-weighted imaging sequence; The gold standard data for the evaluation is the quantitative concentration of active ingredients of nanomedicine in tumors through tissue biopsy; The inclusion criteria included, but were not limited to, tumor type, stage, age range, and whether other chemotherapy or radiotherapy had been received.

3. The method for predicting tumor EPR effect based on MR imaging according to claim 1, characterized in that: In step S2, the specific method steps of denoising, normalization and registration operations are as follows: Gaussian filtering algorithm is used for denoising to remove image noise interference; The linear transformation method is used to normalize the image grayscale value to the [0,1] interval; A rigid registration algorithm is used to register MR images of different sequences to ensure that the position of the tumor is consistent in different images.

4. The method for predicting tumor EPR effect based on MR imaging according to claim 1, characterized in that: In step S4, the basic features include intensity features, shape features, 2D shape features, grayscale co-occurrence matrix features, grayscale travel matrix features, grayscale region size matrix features, and adjacent grayscale matrix; The high-level features include any one or more of intensity features extracted again after wavelet transformation, grayscale co-occurrence matrix features, grayscale run matrix features, and grayscale region size matrix.

5. The method for predicting tumor EPR effect based on MR imaging according to claim 1, characterized in that: In step S5, when the feature selection algorithm is specifically operated, if RFE is combined with a support vector machine, all features are first used to train an SVM model, the importance score of each feature is calculated, and then the features with the lowest importance scores are gradually deleted and the SVM model is retrained until the model performance no longer improves.

6. The method for predicting tumor EPR effect based on MR imaging according to claim 1, characterized in that: In step S5, the feature selection algorithm is one or more of recursive feature elimination, least absolute shrinkage and selection operator.

7. The method for predicting tumor EPR effect based on MR imaging according to claim 1, characterized in that: In step S6, the machine learning algorithm is one or more of a support vector machine, a random forest, and a neural network.

8. The method for predicting tumor EPR effect based on MR imaging according to claim 1, characterized in that: In step S7, the model parameters are adjusted by the cross-validation method during the training process to achieve the best performance of the model. The trained model is validated using the validation set to evaluate the prediction accuracy, sensitivity, and specificity of the model.

9. A system for implementing the method for predicting tumor EPR effect based on MR imaging omics as described in any one of claims 1 to 8, characterized in that: include: The data acquisition module is used to collect MR imaging data of tumor patients and the corresponding gold standard data for tumor EPR effect evaluation. The MR imaging data is obtained by connecting with PACS, and the gold standard data for evaluation is collected from the patient medical record system. The image preprocessing module is written in Python and uses OpenCV and Nibabel libraries to preprocess the acquired MR image data, including denoising, normalization, and registration operations; The tumor region segmentation module is implemented based on the deep learning framework TensorFlow or PyTorch, and uses the pre-trained U-Net model to perform tumor region segmentation on the pre-processed MR images; A feature extraction module, which uses the Pyradiomics toolkit to extract radiomics features from the segmented tumor area, wherein the radiomics features include basic features and advanced features; The feature screening module uses recursive feature elimination, least absolute shrinkage and selection operator feature selection algorithms to screen the extracted radiomics features based on the Scikit-learn library to obtain the most representative feature subset; The model building module uses the filtered feature subset and one or more machine learning algorithms such as support vector machine, random forest, and neural network to build a tumor EPR effect prediction model based on the Scikit-learn library; The model training and validation module divides the collected data into training sets and validation sets, trains and validates the prediction model, evaluates the model performance, and uses the cross-validation method to adjust the model parameters during the training process to obtain the best performance; The prediction application module processes the MR imaging data of a new patient through each module in turn, inputs the obtained features into the trained prediction model, and outputs the prediction results of the tumor EPR effect, which are displayed to the doctor through a graphical user interface.

Images