Method and system for identifying plasmocyte mastitis in inflammatory mass stage

Abstract

The invention relates to a method and a system for identifying plasmocytic mastitis in a mass stage, and the method comprises the steps: S1, obtaining a gray-scale ultrasound image and an ultrasound contrast image of a mammary gland; s2, image omics characteristics of the gray-scale ultrasonic image and the ultrasonic contrast image related to the mammary gland are extracted; and S3, predicting the radiomics characteristics through a preset model so as to determine whether the mammary gland has plasmocyte mastitis in the inflammatory mass stage or not. The method can be used for effectively identifying the plasmocyte mastitis in the plasmocyte mastitis stage, and is relatively good in practicability.

Description

Technical Field

[0001] The invention relates to the technical field of breast diagnosis, in particular to a method and system for identifying plasma cell mastitis in the inflammatory stage. Background Art

[0002] Existing technology can construct a differential diagnosis model for plasma cell mastitis in the inflammatory stage based on radiomics features. Automatic or semi-automatic software is used to non-invasively obtain image features that cannot be recognized by the naked eye from medical images such as CT, MRI, and ultrasound images and convert them into quantitative features. Through deep data mining, machine learning algorithms are used to build models with diagnostic, prognostic or predictive value to provide information for personalized diagnosis, sub-stratification, prediction and disease management. The main process is as follows: First, extract medical omics features: medical image acquisition; outline / segment the part of interest (lesion) in the image (ROI); use automatic or semi-automatic software to extract the radiomics features of the segmented part. Secondly, use a variety of methods to reduce the dimension and screen the extracted radiomics features. Finally, a single machine learning algorithm such as logistic regression is used to build a model; the model results can be further combined with clinical information to build a joint model for the diagnosis of plasma cell mastitis.

[0003] In terms of radiomics feature extraction, previous studies have mostly used CT and MRI images, and the data is relatively simple. In terms of model construction, existing technologies mostly use single machine learning algorithms such as logistic regression and support vector machines to build diagnostic models. The models may have problems such as overfitting and poor generalization ability, and cannot effectively diagnose mastitis. Summary of the invention

[0004] To this end, the technical problem to be solved by the present invention is to overcome the problem that the method for detecting inflammatory plasma cell mastitis in the prior art uses a single machine learning algorithm, which may lead to overfitting of the model, poor generalization ability, etc., and cannot effectively diagnose mastitis.

[0005] In order to solve the above technical problems, the present invention provides a method for identifying plasma cell mastitis in the inflammatory stage, comprising:

[0006] Step S1: acquiring a grayscale ultrasound image and an ultrasound contrast image of a breast;

[0007] Step S2: extracting imaging features of the grayscale ultrasound image and ultrasound contrast-enhanced image of the breast;

[0008] Step S3: predicting the radiomics features by a preset model to determine whether the breast has plasma cell mastitis in the inflammatory stage.

[0009] In one embodiment of the present invention, the preset model in step S3 includes a primary learner and a secondary learner, the primary learner includes at least two basic learners, and the secondary learner is a meta learner, wherein,

[0010] The prediction results obtained by independently training each of the basic learners are combined into a new data set, and the new data set is used as an input of the meta-learner to train the meta-learner to obtain a preset model.

[0011] In one embodiment of the present invention, when the new data set is used as the input of the meta-learner, the method further includes:

[0012] First, the K-fold cross-validation method is used to evenly divide the imaging features into K data subsets. For each data subset, the The data is used as the training set, and the rest The data is used as the validation set;

[0013] Secondly, for the training set in each data subset, after the basic learner is trained with the training set each time, the basic learner is used to predict the corresponding validation set in the data subset to obtain the corresponding prediction results. All the prediction results corresponding to the K data subsets are used as a new data set to train the meta-learner to obtain the preset model.

[0014] In one embodiment of the present invention, different weights are set for the prediction results obtained for each data subset, and the prediction results are multiplied by the corresponding weights to obtain a new data set for input into the meta-learner. The formula is:

[0015]

[0016] in, is the prediction result calculated by cross-validation for each base learner; α i are the weights of each basic learner, all weights are non-negative and the sum is 1; is a new dataset used to input the meta-learner.

[0017] The weight α of each basic learner i Make adjustments, including:

[0018] In one embodiment of the present invention, when the weight α of the base learner i When α is between 0 and 0.3, the first weight formula is used to increase the weight α of the basic learner. i Size, where the first weight formula is:

[0019] α i ←-α i ln(α i), where ← is the assignment symbol;

[0020] When the weight α of the base learner i When α is between 0.7 and 1, the second weight formula is used to reduce the weight α of the base learner. i Size, where the second weight formula is:

[0021] α i ←k·α i , where k is a constant coefficient and k∈{0.8,0.85,0.9,0.95} or k=α i +0.1;

[0022] When the weight α of the base learner i When the weight α of the base learner is between 0.4 and 0.6, i constant;

[0023] If the weight α of the adjusted base learner i If the sum is not 1, then the weight α of the base learner is increased or decreased between 0.4 and 0.6. i To ensure the weight α of the basic learner i The sum is 1.

[0024] In one embodiment of the present invention, the radiomics features are input into each basic learner to obtain the corresponding prediction results, and then compared with the formula The final prediction result of the meta-learner is obtained by combining the weights calculated in

[0025]

[0026] in, is the predicted value obtained by each base learner using radiomics features; is the final prediction value of the meta-learner.

[0027] In one embodiment of the present invention, before step S3, the method further includes: performing feature dimensionality reduction on the radiomics features by LASSO feature screening or principal component analysis.

[0028] In one embodiment of the present invention, the meta-learner is any one of a support vector machine, a logistic regression, a random forest, and an artificial neural network; the basic learner is any one of a K-nearest neighbor algorithm, a support vector machine, a logistic regression, and a naive Bayes.

[0029] In order to solve the above technical problems, the present invention provides a system for identifying plasma cell mastitis in the inflammatory stage, comprising:

[0030] Acquisition module: acquiring grayscale ultrasound images and ultrasound contrast images of breast;

[0031] Extraction module: extracting the imaging features of the grayscale ultrasound image and ultrasound contrast-enhanced image of the breast;

[0032] Prediction module: used to predict the imaging genomics features through a preset model to determine whether the breast has plasma cell mastitis in the inflammatory stage.

[0033] To solve the above technical problems, the present invention provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the above-mentioned method for identifying inflammatory stage plasma cell mastitis when executing the computer program.

[0034] In order to solve the above technical problems, the present invention provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps of the above-mentioned method for identifying inflammatory plasma cell mastitis are implemented.

[0035] The above technical solution of the present invention has the following advantages compared with the prior art:

[0036] The method for identifying plasma cell mastitis in the inflammatory stage of the present invention constructs an efficient model, which combines the prediction results of each basic learner through a hierarchical process, so that the preset model can learn how to best combine the prediction results of different models to improve the prediction performance of the model;

[0037] The model for identifying plasma cell mastitis in the inflammatory stage constructed by the present invention has strong generalization ability and high robustness, and has been proved to have good detection effect through experiments, and can be widely used in practice. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] In order to make the contents of the present invention more clearly understood, the present invention is further described in detail below based on specific embodiments of the present invention in conjunction with the accompanying drawings.

[0039] Figure 1 is a flow chart of the method of the present invention;

[0040] Figure 2 is a schematic diagram of building a preset model using 5-fold cross validation as an example in an embodiment of the present invention;

[0041] Figure 3 Schematic diagram of grayscale ultrasound and ultrasound contrast imaging of a patient with plasma cell mastitis in the inflammatory stage according to an embodiment of the present invention;

[0042] Figure 4 is a calibration diagram of the final model in the training set and the validation set in an embodiment of the present invention;

[0043] Figure 5is a decision curve diagram of the final model in the training set and the validation set in the embodiment of the present invention;

[0044] Figure 6 It is the nomogram of the final model in the embodiment of the present invention. DETAILED DESCRIPTION

[0045] The present invention is further described below in conjunction with the accompanying drawings and specific embodiments so that those skilled in the art can better understand the present invention and implement it, but the embodiments are not intended to limit the present invention.

[0046] Embodiment 1

[0047] Reference Figure 1 As shown, the present invention relates to a method for identifying plasma cell mastitis in the inflammatory stage, comprising:

[0048] Step S1: acquiring a grayscale ultrasound image and an ultrasound contrast image of a breast;

[0049] Step S2: extracting imaging features of the grayscale ultrasound image and ultrasound contrast-enhanced image of the breast;

[0050] Step S3: predicting the radiomics features by a preset model to determine whether the breast has plasma cell mastitis in the inflammatory stage.

[0051] The following is a detailed description of this embodiment.

[0052] First, ultrasound imaging features are extracted from grayscale ultrasound images and ultrasound contrast images. The features are used for model construction after dimensionality reduction (for example, LASSO feature screening or principal component analysis can be used) and screening.

[0053] The reason why grayscale ultrasound images and ultrasound contrast images are used in this embodiment is that: when examining breast-related diseases, ultrasound images are the main examination method, which has the characteristics of no radiation, good real-time performance, low cost, non-invasiveness, and strong repeatability, and can obtain corresponding high-quality data without increasing the burden on patients.

[0054] Secondly, the preset model constructed in this embodiment includes a two-layer structure: (1) two or more basic learners (first-level learners); (2) a meta-learner (second-level learner). The prediction results obtained by the independent training data of each basic learner in the first layer constitute a new data set; the new data set is used as the input of the second layer to train the meta-learner to obtain the preset model.

[0055] Since directly using the prediction results generated by the base learner as the input of the meta-learner may lead to model overfitting, this embodiment uses the K-fold cross-validation method to first divide the radiomics features into K data subsets on average. For each data subset, and each time using The data is used as the training set, and the rest The data is used as the validation set. Secondly, for the training set in each data subset, each time a basic learner is trained, the validation set is predicted by the basic learner to obtain the corresponding prediction results. After repeating K times, all the prediction results are used as a new data set to train the meta-learner to obtain the preset model. This embodiment takes 5-fold cross validation as an example. The specific process is as follows Figure 2 As shown, the data set (radiology features) is divided into five parts, 80% of which is used as a training set. Models are built according to different basic learners, and the remaining 20% ​​of the validation set data is predicted to obtain corresponding prediction results. There are five ways to divide the training set in the 5-fold cross-validation, and a total of 5 groups of prediction results are obtained for the same basic learner algorithm (it should be noted that, for example, the 5 groups of prediction results of basic learner 1 are arranged in a long length as the entire prediction results of basic learner 1, and finally input into the meta-learner). The prediction results of different basic learners are combined as the input data set of the meta-learner, and the meta-learner is trained to build a preset model.

[0056] In this embodiment, multiple algorithms with small correlation between learners and similar performance can be selected as basic learners. The meta-learner can select any algorithm, such as support vector machine (SVM); logistic regression (LR); random forest (RF); artificial neural network (Artificial Neural Unit), etc. Among them, the method of using K-fold cross-validation to construct the best weighted combination of prediction results from each basic learner is also called "Super Learner". Taking the "Super Learner" as an example, the main steps of the preset model are as follows:

[0057] (1) The prediction results obtained by cross-validation of each base learner are combined together to form a new dataset.

[0058] (2) The new data set is used as the input of the meta-learner. The meta-learner calculates the weight of each basic learner according to a "standard" in order to achieve the best performance of the model. The "standard" is used to define and quantify the performance of the algorithm in predicting or explaining data, and can be defined by oneself. For example: in regression analysis, the "standard" can be set to the minimum mean square error between the model's predicted results and the actual outcomes; in binary classification analysis, the "standard" can be set to the maximum area under the receiver operating characteristic (ROC) curve, that is, the minimum value of 1-AUC (Areaunder curve).

[0059] Different weights are set for the prediction results obtained from each data subset, and the prediction results are multiplied by the corresponding weights to obtain a new data set for input into the meta-learner. The formula is:

[0060]

[0061] in, is the prediction result calculated by cross-validation for each base learner; α i The weights of each basic learner when meeting the "standard" are all non-negative numbers and the sum is 1; It is the weighted prediction value of each base learner calculated according to the prediction result dataset, that is, the new dataset used to input the meta learner.

[0062] It is worth mentioning that in this embodiment, the weights α of each basic learner are i Make adjustments, including:

[0063] (1) When the weight α of the basic learner i When α is between 0 and 0.3, the first weight formula is used to increase the weight α of the basic learner. i Size, where the first weight formula is:

[0064] α i ←-α i ln(α i )

[0065] Among them, ← is the assignment symbol;

[0066] (2) When the weight α of the basic learner i When α is between 0.7 and 1, the second weight formula is used to reduce the weight α of the base learner. i Size, where the second weight formula is:

[0067] α i ←k·α i

[0068] Where k is a constant coefficient and k∈{0.8,0.85,0.9,0.95} or k=α i +0.1;

[0069] (3) When the weight α of the basic learner i When the weight α of the base learner is between 0.4 and 0.6, i constant.

[0070] The reason why the higher weights are lowered and the lower weights are increased in this embodiment is that it can more effectively prevent the model from overfitting in the future, making the model of this embodiment more generalizable.

[0071] It should be noted that if the weight α of the adjusted basic learner i If the sum is not 1, then the weight α of the base learner is increased or decreased between 0.4 and 0.6. i To ensure the weight α of the basic learner i The sum is 1, specifically:

[0072] (a) If the weight α of the base learner i The sum satisfies 1<∑α i ≤1.4 or 0.7<∑α i When <1, |∑α i -1|divided by the number of weights n of the base learner between 0.4 and 0.6 0.4-0.6 get If the weight α in the base learner i The sum satisfies 1<∑α i ≤1.4, each weight α in the base learner between 0.4 and 0.6 i minus If the weight α in the base learner i The sum satisfies 0.7<∑α i <1, each weight α in the base learner between 0.4 and 0.6 i Plus The weight α of the basic learner after adjustment i The sum is equal to 1.

[0073] (b) If the weight α of the base learner i The sum does not satisfy 1<∑α i ≤1.4 or does not satisfy 0.7<∑α i When <1, the weight α of each basic learner is not adjusted. i Adjustments are made because the adjusted values ​​are too high or too low, which is not conducive to the generalization of the model in the later stage.

[0074] (3) All original data (i.e., radiomics features) are used for each basic learner to obtain the corresponding prediction results, which are combined with the weights calculated in formula (1) to obtain the final prediction results of the meta-learner:

[0075]

[0076] in, is the predicted value obtained by each basic learner using the entire original data set (i.e., radiomics features); α i is the weight of each basic learner calculated in step (1); It is the final prediction value calculated by combining the prediction results obtained by each base learner based on the entire original data set with the weights (i.e. the final prediction result of the meta-learner).

[0077] In summary, the external testing method of the preset model is as follows: (1) Use the same basic learner to build a model based on all the test set data to obtain the corresponding prediction results; (2) The prediction results are combined with the calculated weights to obtain the final prediction value.

[0078] The stacking methods in the present invention are K-Nearest Neighbor (KNN), support vector machine (SVM), logistic regression (LR) and naive Bayes ( Bayes, NB) as the basic learner. Each basic learner is parameterized, and the specific tuning methods are as follows: (1) KNN: find the K value with the highest accuracy through 10-fold cross-validation; (2) SVM: grid search the optimal parameters gamma and cost through 10-fold cross-validation; (3) LR: select the optimal variables to build the model through stepwise regression algorithm; (4) NB: set Laplace to 1. The number of cross-validation (K) is set to 2 according to the sample size.

[0079] The present invention is further illustrated by a specific case below:

[0080] The method for constructing a differential diagnosis model for inflammatory plasma cell mastitis provided by the present invention is used to distinguish patients with inflammatory plasma cell mastitis and patients with invasive ductal carcinoma of the breast, and comprises the following steps:

[0081] (1) Extraction of ultrasound imaging features based on ultrasound images:

[0082] First, we collected two ultrasound images and clinical information of 341 patients with plasma cell mastitis or invasive ductal carcinoma of the breast who underwent grayscale ultrasound and ultrasound contrast examination.TM Class C ultrasound instruments. According to the different ultrasound instruments used when the patients were examined, the training set (Mindray Resona7, LOGIQ E9) and the validation set (MyLab TM ClassC). The training set is used for feature dimension reduction and screening as well as model construction, and the validation set is used for model performance verification.

[0083] Secondly, an ultrasound physician with many years of ultrasound experience used ITK-SNAP 3.8.0 software to delineate and segment the region of interest (ROI) in the image (see Figure 3 , Figure 3 A and C in the figure represent grayscale ultrasound and ultrasound contrast examination images, respectively. Figure 3 B and D in the figure represent the segmentation of the region of interest (ROI) of the lesion, respectively. The ultrasound image features in the two images are extracted, reduced in dimension and screened. The dimension reduction and screening process includes: removing the feature column containing "0"; selecting the features with a P value less than 0.05 by the selectKbest method; and finally further screening using the Lasso algorithm to obtain the final feature data set for model construction. In the present invention, 788 features were extracted from the two ultrasound images respectively. After dimension reduction and screening, 14 features were finally obtained from the grayscale ultrasound image; ultrasound angiography obtained 11 features and 32 joint features combining the two images, a total of 3 feature data sets.

[0084] (2) Stacking method to build the model:

[0085] The model was constructed using three feature datasets. Two, three or four of the four algorithms, KNN, SVM, LR and NB, were used as basic learners, and the prediction results of the basic learners were stacked to obtain the final results. A total of 33 models were constructed. The comparison results showed that the model constructed with KNN and SVM as basic learners based on the joint dataset performed best. The AUC (95% CI), accuracy, sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV) and F1 values ​​of the model in the training set and validation set were 0.974 (0.956, 0.993), 0.932, 0.942, 0.922, 0.924, 0.941, 0.933; 0.863 (0.802, 0.924), 0.778, 0.714, 0.846, 0.833, 0.733 and 0.769, respectively. This model is superior to the single model constructed using the KNN algorithm or the SVM algorithm alone. The AUC (95% CI), accuracy, sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV) and F1 value of the model constructed using the KNN algorithm alone in the training set and validation set were 0.882 (0.837, 0.927), 0.816, 0.893, 0.738, 0.773, 0.874, 0.829; 0.882 (0.828, 0.937), 0.741, 0.600, 0.892, 0.857, 0.674 and 0.706, respectively. The AUC (95% CI), accuracy, sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV) and F1 value of the model constructed by using the SVM algorithm alone in the training set and validation set were: 0.985 (0.970, 1.000), 0.971, 0.971, 0.971, 0.971, 0.971, 0.971; 0.806 (0.732, 0.880), 0.733, 0.714, 0.754, 0.758, 0.710 and 0.735. The results showed that the model constructed by stacking KNN and SVM algorithms using the Stacking method improved the accuracy of a single model; and in the validation set, the values ​​of sensitivity and specificity of the model were relatively balanced. This model is called the best model (SL model).

[0086] (3) Construction of the final model:

[0087] The prediction results of the best model were combined with clinical data to construct the final model. The final model was evaluated in terms of discrimination, calibration, and clinical utility. Univariate and multivariate logistic screening variables were used to construct the final model. The final model included: patient age, maximum lesion diameter, and whether the neutrophil count was greater than 6.3×10 9 / L and the classification results of the best model. The AUC (95% CI), accuracy, sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV) and F1 value of the final model in the training set and validation set were: 0.978 (0.961, 0.996), 0.942, 0.961, 0.922, 0.925, 0.960, 0.942; 0.938 (0.902, 0.974), 0.830, 0.814, 0.846, 0.851, 0.809 and 0.832. The model has good calibration ability and clinical utility. The calibration graph of the model is shown in the figure below. Figure 4 shown. Figure 5 The decision curve diagram of the model in the training set and the validation set is shown in Figure 2. The results show that the model has high clinical benefits in both the training set and the validation set. The nomogram of the final model is shown in Figure 2. Figure 6 shown.

[0088] Embodiment 2

[0089] This embodiment provides a system for identifying plasma cell mastitis in the inflammatory stage, comprising:

[0090] Acquisition module: acquiring grayscale ultrasound images and ultrasound contrast images of breast;

[0091] Extraction module: extracting the imaging features of the grayscale ultrasound image and ultrasound contrast-enhanced image of the breast;

[0092] Prediction module: used to predict the imaging genomics features through a preset model to determine whether the breast has plasma cell mastitis in the inflammatory stage.

[0093] Embodiment 3

[0094] This embodiment provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of the method for identifying plasma cell mastitis in the inflammatory stage described in Embodiment 1 are implemented.

[0095] Embodiment 4

[0096] This embodiment provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the steps of the method for identifying plasma cell mastitis in the inflammatory stage described in the first embodiment are implemented.

[0097] Those skilled in the art will appreciate that the embodiments of the present application can be provided as methods, systems, or computer program products. Therefore, the present application can adopt the form of complete hardware embodiments, complete software embodiments, or embodiments in combination with software and hardware. Moreover, the present application can adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code. The scheme in the embodiments of the present application can be implemented in various computer languages, for example, object-oriented programming language Java and literal scripting language JavaScript, etc.

[0098] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0099] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0100] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0101] Although the preferred embodiments of the present application have been described, those skilled in the art may make other changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications falling within the scope of the present application.

[0102] Obviously, the above embodiments are merely examples for clear explanation and are not intended to limit the implementation methods. For those skilled in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to list all the implementation methods here. The obvious changes or modifications derived from these are still within the protection scope of the invention.

Claims

1. A method for identifying plasma cell mastitis in the inflammatory stage, characterized in that: include: Step S1: acquiring a grayscale ultrasound image and an ultrasound contrast image of a breast; Step S2: extracting imaging features of the grayscale ultrasound image and ultrasound contrast-enhanced image of the breast; Step S3: predicting the radiomics features by a preset model to determine whether the breast has plasma cell mastitis in the inflammatory stage.

2. The method for identifying plasma cell mastitis in the inflammatory stage according to claim 1, wherein: The preset model in step S3 includes a primary learner and a secondary learner, wherein the primary learner includes at least two basic learners, and the secondary learner is a meta learner, wherein: The prediction results obtained by independently training each of the basic learners are combined into a new data set, and the new data set is used as an input of the meta-learner to train the meta-learner to obtain a preset model.

3. The method for identifying plasma cell mastitis in the inflammatory stage according to claim 2, characterized in that: When the new dataset is used as the input of the meta-learner, it also includes: First, the K-fold cross-validation method is used to evenly divide the imaging features into K data subsets. For each data subset, the The data is used as the training set, and the rest The data is used as the validation set; Secondly, for the training set in each data subset, after the basic learner is trained with the training set each time, the basic learner is used to predict the corresponding validation set in the data subset to obtain the corresponding prediction results. All the prediction results corresponding to the K data subsets are used as a new data set to train the meta-learner to obtain the preset model.

4. The method for identifying plasma cell mastitis in the inflammatory stage according to claim 3, characterized in that: Different weights are set for the prediction results obtained from each data subset, and the prediction results are multiplied by the corresponding weights to obtain a new data set for input into the meta-learner. The formula is: in, is the prediction result calculated by cross-validation for each base learner; α i are the weights of each basic learner, all weights are non-negative and the sum is 1; is a new dataset used to input the meta-learner.

5. The method for identifying plasma cell mastitis in the inflammatory stage according to claim 4, characterized in that: The weight α of each basic learner i Make adjustments, including: When the weight α of the base learner i When α is between 0 and 0.3, the first weight formula is used to increase the weight α of the basic learner. i Size, where the first weight formula is: α i ←-α i ln(α i ), where ← is the assignment symbol; When the weight α of the base learner i When α is between 0.7 and 1, the second weight formula is used to reduce the weight α of the base learner. i Size, where the second weight formula is: α i ←k·α i , where k is a constant coefficient and k∈{0.8,0.85,0.9,0.95} or k=α i +0.1; When the weight α of the base learner i When the weight α of the base learner is between 0.4 and 0.6, i constant; If the weight α of the adjusted base learner i If the sum is not 1, then the weight α of the base learner is increased or decreased between 0.4 and 0.

6. i To ensure the weight α of the basic learner i The sum is 1.

6. The method for identifying plasma cell mastitis in the inflammatory stage according to claim 4, characterized in that: The radiomics features are input into each basic learner to obtain the corresponding prediction results, and then compared with the formula The final prediction result of the meta-learner is obtained by combining the weights calculated in in, is the predicted value obtained by each base learner using radiomics features; is the final prediction value of the meta-learner.

7. The method for identifying plasma cell mastitis in the inflammatory stage according to claim 2, characterized in that: The meta-learner is any one of support vector machine, logistic regression, random forest, and artificial neural network; the basic learner is any one of K nearest neighbor algorithm, support vector machine, logistic regression, and naive Bayes.

8. A system for identifying plasma cell mastitis in the inflammatory stage, characterized in that: include: Acquisition module: acquiring grayscale ultrasound images and ultrasound contrast images of breast; Extraction module: extracting the imaging features of the grayscale ultrasound image and ultrasound contrast-enhanced image of the breast; Prediction module: used to predict the imaging genomics features through a preset model to determine whether the breast has plasma cell mastitis in the inflammatory stage.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the steps of the method for identifying inflammatory plasma cell mastitis according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method for identifying plasma cell mastitis in the inflammatory stage according to any one of claims 1 to 7 are implemented.

Images