Imaging feature extraction method and device, electronic equipment and storage medium

Abstract

Embodiments of the present application provide an imageomics feature extraction method and device, electronic equipment and storage medium, relating to the technical field of image processing. The original image data in the sample set is acquired, the original image data is preprocessed to obtain optimized image data; the optimized image data is subjected to feature extraction to obtain a first feature extraction result; a preset stable feature is matched with the first feature extraction result to obtain a second feature extraction result; a plurality of initial feature extraction models are trained based on the second feature extraction result to obtain a plurality of optimized feature extraction models; the optimized feature extraction model with the highest performance evaluation score is determined as a target feature extraction model, and a prediction result of the current image data is determined according to the target feature extraction model. The present application can effectively solve the problems of feature redundancy and lack of explanation in the existing imageomics feature extraction and classification problems.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of image processing, in particular to an imageomics feature extraction method and device, electronic equipment and storage medium. BACKGROUND

[0002] Imageomics feature extraction is a process of extracting high-throughput quantitative information from medical image data. The existing imageomics feature extraction method has the following defects: first, a large number of features may be extracted in the feature extraction process, some of which may have a high degree of correlation or redundancy, and some features may be unstable and non-reproducible, which will lead to model overfitting and reduced generalization ability; second, many imageomics features lack clear biological interpretation, which limits clinical application. SUMMARY

[0003] In order to solve the above technical problems, the present application provides an imageomics feature extraction method, device, electronic equipment and storage medium.

[0004] In a first aspect, the present application provides an imageomics feature extraction method, which comprises:

[0005] Obtaining original image data in a sample set, pre-processing the original image data to obtain optimized image data;

[0006] Performing feature extraction on the optimized image data to obtain a first feature extraction result;

[0007] Matching a preset stable feature with the first feature extraction result to obtain a second feature extraction result, the preset stable feature including a first feature name and a first feature biological interpretation;

[0008] Training a plurality of initial feature extraction models based on the second feature extraction result to obtain a plurality of optimized feature extraction models;

[0009] Evaluating the performance of each of the optimized feature extraction models, and determining the optimized feature extraction model with the highest performance evaluation score as a target feature extraction model;

[0010] Inputting real-time image data into the target feature extraction model to obtain a real-time prediction result.

[0011] In an embodiment, the pre-processing of the original image data comprises:

[0012] In an embodiment, the original image data is subjected to quality improvement processing to obtain improved image data;

[0013] intensity standardization and intensity normalization are performed on the improved image data to obtain standard image data;

[0014] cutting is performed on the standard image data to obtain regional image data;

[0015] resampling is performed on the regional image data to obtain the optimized image data.

[0016] The preprocessing of the original image data further comprises:

[0017] if the collection time points or modalities of the original image data are different, the first original image data in the sample set is determined as a registration template;

[0018] the collection time points or modalities of the remaining original image data in the sample set are adjusted according to the registration template.

[0019] In an embodiment, the first feature extraction result comprises a plurality of preliminary extracted features, and the matching of the preset stable features with the first feature extraction result to obtain a second feature extraction result comprises:

[0020] a second feature name of each of the preliminary extracted features is obtained;

[0021] if the second feature name of the preliminary extracted feature is the same as the first feature name, the preliminary extracted feature is determined as a target feature;

[0022] a set of the target features is determined as the second feature extraction result.

[0023] In an embodiment, the initial feature extraction model is a combination of a feature selection algorithm and a classifier;

[0024] the feature selection algorithm comprises at least one of minimum redundancy maximum relevance, recursive feature elimination, LASSO algorithm, Ridge regression, principal component analysis, linear discriminant analysis, mutual information, t-distributed stochastic neighbor embedding, elastic net and ReliefF algorithm;

[0025] the classifier comprises at least one of a vector machine, logistic regression, random forest, gradient boosting decision tree, K-nearest neighbor, multilayer perception, naive Bayes, AdaBoost model, XGBoost model and LightGBM model.

[0026] In an embodiment, before the training of a plurality of initial feature extraction models based on the second feature extraction result, the method comprises:

[0027] the second feature extraction result is segmented based on a ten-fold cross-validation method to obtain a training set and a test set.

[0028] The training set is used to train the initial feature extraction model, and the test set is used to evaluate the performance of each optimized feature extraction model.

[0029] In an embodiment, the real-time prediction result includes a third feature name and a third feature biological explanation.

[0030] In a second aspect, an embodiment of the present application provides an imaging omics feature extraction device, which comprises:

[0031] A preprocessing module is configured to obtain original image data in a sample set, pre-process the original image data, and obtain optimized image data.

[0032] An extraction module is configured to extract features from the optimized image data and obtain a first feature extraction result.

[0033] A matching module is configured to match preset stable features with the first feature extraction result and obtain a second feature extraction result, wherein the preset stable features include a first feature name and a first feature biological explanation.

[0034] A training module is configured to train a plurality of initial feature extraction models based on the second feature extraction result and obtain a plurality of optimized feature extraction models, wherein each initial feature extraction model includes a feature selection algorithm and a classifier.

[0035] An evaluation module is configured to evaluate the performance of each optimized feature extraction model and determine the optimized feature extraction model with the highest performance evaluation score as a target feature extraction model.

[0036] In a third aspect, an embodiment of the present application provides an electronic device, which comprises a memory and a processor, wherein the memory is configured to store a computer program, and the computer program is configured to execute the imaging omics feature extraction method provided in the first aspect when the processor runs.

[0037] In a fourth aspect, an embodiment of the present application provides a computer readable storage medium, which stores a computer program, and the computer program is configured to execute the imaging omics feature extraction method provided in the first aspect when the processor runs.

[0038] The imaging omics feature extraction method provided in the present application reduces feature redundancy and correlation in the feature extraction process, matches preset stable features and their biological explanations based on historical research with the extracted features, and then trains the model, so that the features with stability and clinical significance can be finally output. The reliability, interpretability and practicability of the features are improved. BRIEF DESCRIPTION OF DRAWINGS

[0039] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the embodiments. It should be understood that the following drawings only show some of the embodiments of the present application, and therefore should not be considered as limiting the scope. For those skilled in the art, other related drawings can also be obtained without creative labor.

[0040] Figure 1 A flowchart of the image feature extraction method provided by the embodiments of the present application is shown;

[0041] Figure 2 A structural diagram of the image feature extraction device provided by the embodiments of the present application is shown. DETAILED DESCRIPTION

[0042] In order to make the objects, technical solutions and advantages of the embodiments of the present application clearer, the following will combine the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are some of the embodiments of the present application, but not all the embodiments. The components of the embodiments of the present application described and shown in the drawings can be arranged and designed in various different configurations.

[0043] Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the claimed present application, but only represents selected embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of the present application.

[0044] It should be noted that: similar reference numbers and letters represent similar items in the following drawings, and therefore, once an item is defined in one drawing, it does not need to be further defined and explained in subsequent drawings.

[0045] In the description of the present application, it should be noted that if the terms "upper", "lower", "inner", "outer" and the like indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, or the orientation or positional relationship when the product of the present application is usually placed, and are only for the convenience of describing the present application and simplifying the description, and therefore cannot be understood as indicating or implying that the indicated device or element must have a particular orientation, be constructed and operated in a particular orientation, and therefore cannot be understood as limiting the present application.

[0046] In addition, if the terms "first", "second" and the like appear, they are only used to distinguish the description, and cannot be understood as indicating or implying relative importance.

[0047] It should be noted that the features in the embodiments of the application can be combined with each other without conflict.

[0048] Embodiment 1

[0049] The embodiment of the present disclosure provides an imageomics feature extraction method.

[0050] Imageomics feature extraction is a process of extracting high-throughput quantitative information from medical image data. These features can help researchers and clinicians better understand the characteristics of diseases, thereby providing strong support for diagnosis, prognosis and treatment. However, the existing image extraction method has certain defects, specifically, mainly in two aspects:

[0051] Feature redundancy and instability: a large number of features may be extracted in the feature extraction process, some of which may have high correlation or redundancy, and many features may be unstable and non-reproducible. This may lead to model overfitting and reduced generalization ability.

[0052] Feature interpretability: many imageomics features may lack clear biological explanation, leading to limited clinical application. Therefore, extracting features with biological significance and clinical relevance is a key challenge.

[0053] Based on this, the present application proposes an imageomics feature extraction method, which aims to reduce feature redundancy and correlation in the feature extraction process, and based on the stable features reported in the literature and their biological basis as preset options, these features with stability and clinical significance can be selectively output. This method can improve the reliability, interpretability and practicality of the features.

[0054] Specifically, referring to Figure 1 , the imageomics feature extraction method comprises:

[0055] Step S110, obtaining original image data in a sample set, pre-processing the original image data to obtain optimized image data;

[0056] The original image data can be imported through a data input module. The original image data supports common medical image formats (DICOM, NIFTI), and can be lung CT images, or can be applied to other types of tumors, such as breast cancer, liver cancer, etc.

[0057] In an embodiment, the pre-processing of the original image data comprises:

[0058] Quality improvement processing is performed on the original image data to obtain improved image data;

[0059] The improved image data is intensity standardized and intensity normalized to obtain standard image data; the standard image data is cut to obtain regional image data; and the regional image data is resampled to obtain the optimized image data.

[0060] In an embodiment, the preprocessing of the original image data further includes:

[0061] If the collection time points or modalities of the original image data are different, the first original image data in the sample set is determined as a registration template; and the collection time points or modalities of the remaining original image data in the sample set are adjusted according to the registration template.

[0062] The image preprocessing includes:

[0063] Quality improvement: denoising, smoothing and other operations are performed on the imported image data to improve the image quality and provide a good foundation for subsequent feature extraction.

[0064] Standardization: intensity standardization is performed on the image to unify the image data obtained under different devices and different parameters to the same intensity range, and eliminate the influence of device and parameter differences. Z-score standardization (Standard Scaling) is used, that is, the intensity value of each pixel is subtracted by the average intensity value of the entire image, and then divided by the standard deviation. This method makes the average value of the data become 0 and the standard deviation become 1. The formula of Z-score standardization is: X_standardized=(X-X_mean) / X_std, wherein X is the original intensity value, X_mean is the average value of the original data, and X_std is the standard deviation of the original data.

[0065] Intensity normalization: the intensity value of the image is normalized to a standard range to reduce the image intensity difference caused by different scanning conditions. The formula is as follows: X_normalized=(X-X_min) / (X_max-X_min). Wherein, X is the original intensity value, X_min and X_max are the minimum value and maximum value of the original data respectively.

[0066] Cutting: taking lung CT as an example, the image is cut according to the lung region, and the regions irrelevant to the lung, such as background and chest wall, are removed.

[0067] Resampling: the image is resampled to 1×1×1mm 3 voxels, so that different images have the same spatial resolution.

[0068] Registration: If needed, image registration can be performed to align the high-resolution CT scan images of each patient's first examination as a template with images at different time points or different modalities for further feature extraction and analysis.

[0069] Step S120, feature extraction is performed on the optimized image data to obtain a first feature extraction result;

[0070] The process of feature extraction usually includes the following types of features:

[0071] First-order statistics: These features describe the basic statistical properties of pixel values, such as mean, standard deviation, skewness, and kurtosis, etc.

[0072] Shape features: These features describe the geometric shape of biological structures, such as area, perimeter, volume, major axis, and minor axis, etc.

[0073] Gray Level Co-occurrence Matrix (GLCM) features: GLCM is a method to describe the texture characteristics of an image, which reflects the texture information of an image by calculating the gray level co-occurrence probability between pixels. Common GLCM features include contrast, correlation, energy, and homogeneity, etc.

[0074] Gray Level Run Length Matrix (GLRLM) features: GLRLM describes the length of consecutive pixels with the same gray level in an image. Common GLRLM features include short run emphasis, long run emphasis, gray level nonuniformity, and run length nonuniformity, etc.

[0075] Gray Level Size Zone Matrix (GLSZM) features: GLSZM describes the size of pixel regions with the same gray level. Common GLSZM features include small zone emphasis, large zone emphasis, gray level nonuniformity, and zone size nonuniformity, etc.

[0076] Higher-order features: These features extract more complex image features by transforming the original image (such as wavelet transform, Fourier transform, etc.).

[0077] After feature extraction, feature selection and dimensionality reduction are usually performed to screen the most discriminative and clinically meaningful features. These features can be used to establish a prediction model related to the disease, thereby guiding clinical decision-making.

[0078] automatically extracting a plurality of features from the processed image, such as first-order statistics, shape features, a gray level co-occurrence matrix (GLCM), a gray level run length matrix (GLRLM), a gray level size zone matrix (GLSZM), high-order features (by transforming the original image, such as wavelet transformation), and the like, and taking all of the features as the first feature extraction result.

[0079] In step S130, the preset stable features are matched with the first feature extraction result to obtain a second feature extraction result, and the preset stable features include a first feature name and a first feature biological interpretation.

[0080] During the extraction process, the preset stable features in the preset stable feature library can be automatically matched, and all features and stable features can be output simultaneously.

[0081] The preset stable features are features that have been proven to have clear biological interpretation and clinical relevance, high reliability, and come from past empirical research or literature reports; the first feature name is used as a matching basis; and the first feature biological interpretation comes from literature reports and has clear biological and clinical significance.

[0082] In an embodiment, the first feature extraction result includes a plurality of preliminary extraction features, and the matching of the preset stable features with the first feature extraction result to obtain the second feature extraction result includes:

[0083] A second feature name of each of the preliminary extraction features is obtained, and if the second feature name of the preliminary extraction feature is the same as the first feature name, the preliminary extraction feature is determined as a target feature, and a set of the target features is determined as the second feature extraction result.

[0084] A string-based matching method can be used to match according to the feature name to ensure that the feature name in the second feature extraction result, i.e., the second feature name, is consistent with the first feature name. Generally, they need to be completely consistent to be matched.

[0085] In step S140, a plurality of initial feature extraction models are trained based on the second feature extraction result to obtain a plurality of optimized feature extraction models.

[0086] In an embodiment, the initial feature extraction model is a combination of a feature selection algorithm and a classifier.

[0087] The feature selection algorithm includes at least one of minimum redundancy maximum relevance, recursive feature elimination, LASSO algorithm, Ridge regression, principal component analysis, linear discriminant analysis, mutual information, t-distributed stochastic neighbor embedding, elastic net, and ReliefF algorithm.

[0088] The classifier includes at least one of a vector machine, logistic regression, random forest, gradient boosting decision tree, K-nearest neighbor, multilayer perception, naive Bayes, AdaBoost model, XGBoost model, and LightGBM model.

[0089] Specifically, the feature selection algorithm can be integrated in a feature selection submodule, including 10 commonly used feature selection methods of minimum redundancy maximum relevance (mRMR), recursive feature elimination (RFE), LASSO (Least Absolute Shrinkage and Selection Operator), Ridge regression, principal component analysis (PCA), linear discriminant analysis (LDA), mutual information (Mutual Information), t-distributed stochastic neighbor embedding (t-SNE), elastic net (Elastic Net), and ReliefF algorithm.

[0090] Machine learning classifiers are integrated in the model training submodule, including 10 machine learning classifiers of support vector machine (SVM), logistic regression (LR), random forest (RF), gradient boosting decision tree (GBDT), K-nearest neighbor (KNN), multilayer perception (MLP), naive Bayes (Naive Bayes), AdaBoost, eXtreme Gradient Boosting (XGBoost), and LightGBM (Light Gradient Boosting Machine) for model construction.

[0091] In an embodiment, before training the plurality of initial feature extraction models based on the second feature extraction result, the method includes: segmenting the second feature extraction result based on a ten-fold cross-validation method to obtain a training set and a test set; wherein the training set is used to train the initial feature extraction model, and the test set is used to evaluate the performance of each optimized feature extraction model.

[0092] Using a 10-fold cross-validation method, the dataset is divided into training, validation, and test sets. The software automatically evaluates the model's performance on the validation set. The original dataset is divided into ten equal parts, with nine parts used as the training set and the remaining part as the test set. This process is repeated ten times, ensuring that each subset is used to validate the model once. Finally, the average of these ten validation results is used as the model's performance metric.

[0093] The advantage of 10-fold cross-validation is that it improves the robustness and accuracy of the model because all samples in the dataset are used for both training and testing, and each subset is validated once. Furthermore, it better avoids overfitting and underfitting, thus improving the model's generalization ability.

[0094] Step S150: Evaluate the performance of each optimized feature extraction model, and determine the optimized feature extraction model with the highest performance evaluation score as the target feature extraction model.

[0095] Commonly used evaluation metrics include accuracy, recall, precision, F1 score, and AUC-ROC curve. By comparing the performance of different feature selection methods and classifiers, the optimal combination can be found.

[0096] It is important to note that the final target feature extraction model with the best performance may only use a small portion of the features from the second feature extraction results, which are automatically selected during training.

[0097] Step S160: Input the real-time image data into the target feature extraction model to obtain the real-time prediction result.

[0098] In one embodiment, the real-time prediction result includes a third feature name and a third feature biological interpretation. That is, the output result can output the name of the prediction result and the corresponding biological interpretation based on the user's selection, thereby selectively outputting these stable and clinically significant features; improving the reliability, interpretability, and usability of the features.

[0099] The radiomics feature extraction method provided in this application can be applied to machine learning classification software. In the results display submodule, the user is shown the results of feature extraction and matching, feature selection, model training, and evaluation. Users can view the performance metrics of various feature selection methods and classifiers and select the optimal combination. Furthermore, users can choose to output results that match preset stable features.

[0100] Machine learning classification software can also include parameter setting and tuning modules, allowing users to configure parameters for various modules, such as feature extraction methods and matching algorithms. Furthermore, it can provide automatic tuning capabilities, such as using grid search and Bayesian optimization to find the optimal parameter combinations.

[0101] In addition, a user interface and interaction module can be included to provide a friendly graphical user interface or web interface, allowing users to easily operate the software and perform data input, parameter setting, result display, and other functions. In addition, a command line interface can be provided to facilitate advanced users and batch processing tasks.

[0102] Stable features can be collected and organized according to different diseases or research objectives, making the software more targeted and practical. In addition, the stable feature library can be updated regularly to allow users to access the latest research results and feature information.

[0103] The imageomics feature extraction method provided by the present application has the following advantages:

[0104] Reliability: The preset stable features are based on literature reports and empirical research, so they have high reliability. In some cases, these features may have more clinical value than automatically extracted features; Improved interpretability: These stable features usually have clear biological explanations and clinical relevance, so they can help users better understand the results; Flexibility: Stable features are provided as preset options, and users can choose whether to output these features according to their needs. This provides flexibility for users, allowing them to select the most suitable features according to the actual application scenario; Save computing resources: In some cases, users may only be interested in these preset stable features and do not need to calculate other features. This can save computing resources and time.

[0105] In summary, the imageomics feature extraction method provided by the present embodiment can effectively solve the problems of feature redundancy and low correlation in existing imageomics feature extraction and classification problems; and improves image quality, solves the problem of difficult-to-explain features, and improves the reliability, interpretability and practicality of features, providing strong support for research and clinical applications.

[0106] Embodiment 2

[0107] In addition, the present disclosure provides an imageomics feature extraction device.

[0108] Specifically, as shown in Figure 2 The imageomics feature extraction device 200 includes:

[0109] The preprocessing module 210 is configured to obtain original image data in a sample set, and pre-process the original image data to obtain optimized image data;

[0110] The extraction module 220 is configured to perform feature extraction on the optimized image data to obtain a first feature extraction result;

[0111] The matching module 230 is configured to match the preset stable features with the first feature extraction result to obtain a second feature extraction result, wherein the preset stable features include a first feature name and a first feature biological interpretation.

[0112] The training module 240 is configured to train a plurality of initial feature extraction models based on the second feature extraction result to obtain a plurality of optimized feature extraction models.

[0113] The evaluation module 250 is configured to perform performance evaluation on each of the optimized feature extraction models, and determine the optimized feature extraction model with the highest performance evaluation score as a target feature extraction model.

[0114] The prediction module 260 is configured to input real-time image data into the target feature extraction model to obtain a real-time prediction result.

[0115] The imageomics feature extraction apparatus 200 provided in the embodiment can implement the imageomics feature extraction method provided in the embodiment 1, and thus the details are not repeated here.

[0116] The imageomics feature extraction apparatus provided in the embodiment can effectively solve the problems of feature redundancy and low correlation in the existing imageomics feature extraction and classification problems, and can improve the image quality, solve the problems of difficult-to-explain features, and the like. The reliability, interpretability and practicability of the features are improved, and strong support can be provided for research and clinical application.

[0117] Embodiment 3

[0118] In addition, the embodiment of the disclosure provides an electronic device including a memory and a processor, the memory stores a computer program, and the computer program executes the imageomics feature extraction method provided in the embodiment 1 when the processor runs.

[0119] The electronic device provided in the embodiment can implement the imageomics feature extraction method provided in the embodiment 1, and thus the details are not repeated here.

[0120] The electronic device provided in the embodiment can effectively solve the problems of feature redundancy and low correlation in the existing imageomics feature extraction and classification problems, and can improve the image quality, solve the problems of difficult-to-explain features, and the like. The reliability, interpretability and practicability of the features are improved, and strong support can be provided for research and clinical application.

[0121] Embodiment 4

[0122] The application further provides a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to implement the imageomics feature extraction method provided in Embodiment 1.

[0123] In the embodiment, the computer readable storage medium can be a Read-Only Memory (ROM), a Random Access Memory (RAM), a magnetic disk or an optical disk, etc.

[0124] The computer readable storage medium provided in the embodiment can implement the imageomics feature extraction method provided in Embodiment 1. To avoid repetition, details are not described herein.

[0125] It should be noted that, in this document, the term “comprising” or “including” or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or terminal including a series of elements not only includes those elements, but also includes other elements not explicitly listed, or further includes elements inherent to such a process, method, article or terminal. Without more limitations, the element defined by the statement “including a…” does not exclude the presence of another identical element in the process, method, article or terminal including the element.

[0126] From the above description of the embodiments, those skilled in the art can clearly understand that the above-mentioned embodiment method can be realized by means of software and necessary general hardware platform, of course, it can also be realized by hardware, but in many cases, the former is a better embodiment. Based on such understanding, the technical solutions of the present application can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes a plurality of instructions for making a terminal (which can be a mobile phone, computer, server, air conditioner, or network device, etc.) execute the methods described in various embodiments of the present application.

[0127] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any skilled person in the art can easily think of changes or replacements within the technical scope disclosed in the present application, which should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A method for extracting radiomics features, characterized in that, The method includes: Obtain the original image data from the sample set, and preprocess the original image data to obtain optimized image data; Feature extraction is performed on the optimized image data to obtain the first feature extraction result; The preset stable features are matched with the first feature extraction result to obtain the second feature extraction result. The preset stable features include the first feature name and the first feature biological interpretation. Based on the second feature extraction result, multiple initial feature extraction models are trained to obtain multiple optimized feature extraction models; The performance of each optimized feature extraction model is evaluated, and the optimized feature extraction model with the highest performance evaluation score is determined as the target feature extraction model. Real-time image data is input into the target feature extraction model to obtain real-time prediction results; the real-time prediction results include the name of the third feature and the biological interpretation of the third feature; The first feature extraction result includes multiple preliminary extracted features. The step of matching the preset stable features with the first feature extraction result to obtain the second feature extraction result includes: Obtain the second feature name of each of the initially extracted features; If the name of the second feature extracted in the preliminary extraction is the same as the name of the first feature, then the preliminary extracted feature is determined as the target feature; The set of target features is determined as the second feature extraction result.

2. The method according to claim 1, characterized in that, The preprocessing of the original image data includes: The original image data is subjected to quality improvement processing to obtain improved image data; The improved image data is subjected to intensity standardization and intensity normalization to obtain standard image data; The standard image data is segmented to obtain regional image data; The image data of the region is resampled to obtain the optimized image data.

3. The method according to claim 2, characterized in that, The preprocessing of the original image data further includes: If the acquisition time points or modalities of the original image data are different, then the first original image data in the sample set is determined as the registration template; Adjust the acquisition time point or modality of the remaining raw image data in the sample set according to the registration template.

4. The method according to claim 1, characterized in that, The initial feature extraction model is a combination of a feature selection algorithm and a classifier; The feature selection algorithm includes at least one of the following: minimum redundancy maximum correlation, recursive feature elimination, LASSO algorithm, Ridge regression, principal component analysis, linear discriminant analysis, mutual information, t-distributed random neighborhood embedding, elastic network, and ReliefF algorithm; The classifier includes at least one of the following: vector machine, logistic regression, random forest, gradient boosting decision tree, K-nearest neighbor, multilayer perceptron, Naive Bayes, AdaBoost model, XGBoost model, and LightGBM model.

5. The method according to claim 1, characterized in that, Before training multiple initial feature extraction models based on the second feature extraction result, the process includes: The second feature extraction result is segmented based on the ten-fold cross-validation method to obtain a training set and a test set; The training set is used to train the initial feature extraction model, and the test set is used to evaluate the performance of each optimized feature extraction model.

6. A radiomics feature extraction device, characterized in that, The device includes: The preprocessing module is used to acquire the original image data in the sample set, preprocess the original image data, and obtain optimized image data. The extraction module is used to extract features from the optimized image data to obtain a first feature extraction result; A matching module is used to match a preset stable feature with the first feature extraction result to obtain a second feature extraction result. The preset stable feature includes a first feature name and a first feature biological interpretation. The training module is used to train multiple initial feature extraction models based on the second feature extraction result to obtain multiple optimized feature extraction models, wherein each initial feature extraction model includes a feature selection algorithm and a classifier; The evaluation module is used to evaluate the performance of each of the optimized feature extraction models and determine the optimized feature extraction model with the highest performance evaluation score as the target feature extraction model. The prediction module is used to input real-time image data into the target feature extraction model to obtain real-time prediction results; the real-time prediction results include the name of the third feature and the biological interpretation of the third feature; The first feature extraction result includes multiple preliminary extracted features. The step of matching the preset stable features with the first feature extraction result to obtain the second feature extraction result includes: Obtain the second feature name of each of the initially extracted features; If the name of the second feature extracted in the preliminary extraction is the same as the name of the first feature, then the preliminary extracted feature is determined as the target feature; The set of target features is determined as the second feature extraction result.

7. An electronic device, characterized in that, The device includes a memory and a processor, wherein the memory stores a computer program that executes the radiomics feature extraction method according to any one of claims 1 to 5 when the processor is running.

8. A computer-readable storage medium, characterized in that, It stores a computer program that, when run on a processor, executes the radiomics feature extraction method according to any one of claims 1 to 5.

Images