Breast cancer radiotherapy sensitivity prediction method and system based on deep learning

Abstract

The present invention relates to the field of bioinformatics computing technology, addressing the technical problem of effectively integrating histopathology image data with miRNA data. In particular, it relates to a deep learning-based method for predicting breast cancer radiotherapy sensitivity, comprising the following steps: obtaining histopathology image data and miRNA data from the TCGA database for female breast cancer patients undergoing radiotherapy; and a system for applying the above-mentioned method for predicting breast cancer radiotherapy sensitivity, comprising: a dataset acquisition module for obtaining histopathology image data and miRNA data from the TCGA database for female breast cancer patients undergoing radiotherapy. The present invention effectively predicts the radiotherapy sensitivity of breast cancer patients, enabling a relatively accurate determination of whether a patient requires radiotherapy, thereby enabling precise treatment of the patient and avoiding wasted time and harm caused by radiotherapy.

Description

Technical Field

[0001] The present invention relates to the field of bioinformatics computing technology, and in particular to a method and system for predicting breast cancer radiotherapy sensitivity based on deep learning. Background Art

[0002] Breast cancer is the most common cancer in women worldwide and a leading cause of death in women. Currently, radiotherapy is an important treatment option for breast cancer patients. However, due to the heterogeneity and individual differences in breast cancer, not all patients achieve optimal results after radiotherapy; some even experience worsening of their condition. Therefore, prior to radiotherapy, it is necessary to determine whether the patient is radiosensitive, meaning that radiotherapy can effectively prolong their survival.

[0003] Predicting radiotherapy sensitivity in breast cancer patients can provide doctors with scientific advice when developing personalized treatment plans, avoiding ineffective treatments. In recent years, researchers have proposed a variety of breast cancer radiotherapy sensitivity prediction models. Most of these models are based on genomic data and ignore the histopathology images of cancer patients. Histopathology images play a vital role in the diagnosis and treatment of breast cancer. Furthermore, studies have shown that histopathology images contain rich information that can effectively predict the survival of breast cancer patients. Summary of the Invention

[0004] In response to the shortcomings of the existing technology, the present invention provides a deep learning-based breast cancer radiotherapy sensitivity prediction method and system, which solves the technical problem of effectively integrating tissue pathology image data with miRNA data, and achieves the purpose of effectively predicting the radiotherapy sensitivity of breast cancer patients.

[0005] To solve the above technical problems, the present invention provides the following technical solution: a method for predicting breast cancer radiotherapy sensitivity based on deep learning, characterized by comprising the following steps:

[0006] Obtain histopathological image data and miRNA data of female breast cancer patients who received radiotherapy from the TCGA database;

[0007] The acquired tissue pathology image data and miRNA data were screened, samples with a survival time of less than 30 days were removed, and the data were divided into positive and negative sample data sets based on the three-year survival time;

[0008] Extract histopathological image features and miRNA data features from positive and negative sample datasets;

[0009] A radiotherapy sensitivity prediction model was constructed based on histopathological image features and miRNA data features;

[0010] Extracting histopathological image features and miRNA data features from patient samples and inputting them into the radiotherapy sensitivity prediction model to predict and determine whether the patient has radiotherapy sensitivity;

[0011] If the probability value output by the radiotherapy sensitivity prediction model is greater than or equal to the preset threshold, it will be predicted as a positive sample;

[0012] If the probability value output by the radiotherapy sensitivity prediction model is less than the preset threshold, it is predicted as a negative sample.

[0013] Furthermore, the division of positive and negative sample data sets includes:

[0014] Female breast cancer patients who received radiotherapy were divided into long-term survivors and short-term survivors based on three-year survival time;

[0015] Patients with long-term survival of more than 3 years after radiotherapy are marked as 1, which is a positive sample data set;

[0016] Patients with short survival time less than 3 years after radiotherapy are marked as 0, which is a negative sample data set.

[0017] Furthermore, after extracting the tissue pathology image features and the miRNA data features, the method further includes: using a StandardScaler method to perform standardization processing on the tissue pathology image features and the miRNA data features to transform them into a normal distribution.

[0018] Furthermore, extracting histopathological image features from the positive and negative sample datasets includes the following process:

[0019] The histopathological images in the positive and negative sample data sets are segmented, and the histopathological image of each patient is segmented into several histopathological image sub-blocks;

[0020] Discard the histopathology image sub-blocks where the tissue area accounts for less than 70% of the histopathology image;

[0021] Use the LBP function in the Skimage library to extract several 10-dimensional feature vectors from several tissue pathology image sub-blocks;

[0022] Combine several 10-dimensional feature vectors into a feature matrix and remove the median of each column;

[0023] A 10-dimensional LBP texture feature vector of tissue pathology image features is extracted from the feature matrix.

[0024] Furthermore, the size of the tissue pathology image sub-block is 1024×1024.

[0025] Furthermore, differential expression analysis was used as a method to extract miRNA data features from positive and negative sample datasets, and the process included:

[0026] The miRNAs in the positive and negative sample datasets were analyzed using the R package DESeq2 (|log2 fold change|>1 and P<0.05), and 13 differentially expressed miRNA data features were finally obtained.

[0027] Furthermore, the process of constructing a radiotherapy sensitivity prediction model also includes:

[0028] The histopathological image features and miRNA data features are input into two identical bidirectional long short-term memory networks respectively, and the histopathological image features and miRNA data features encoded by the bidirectional long short-term memory network are spliced;

[0029] Batch normalization method is used to normalize the spliced ​​and combined tissue pathology image features and miRNA data features;

[0030] The attention mechanism is used to assign weights to the normalized histopathological image features and miRNA data features;

[0031] The weighted tissue pathology image features and miRNA data features were input into a three-layer neural network to obtain a radiotherapy sensitivity prediction model.

[0032] Furthermore, in the step of constructing a radiotherapy sensitivity prediction model, the following steps are also included:

[0033] The five-fold cross-validation method is used to randomly divide the positive and negative sample data sets into training sets and validation sets;

[0034] The training set is input into the radiotherapy sensitivity prediction model for training;

[0035] The validation set was input into the radiotherapy sensitivity prediction model for parameter adjustment, and finally the optimal radiotherapy sensitivity prediction model was obtained.

[0036] The present invention also provides a system for use in the above-mentioned method for predicting breast cancer radiotherapy sensitivity, comprising:

[0037] A data acquisition module, wherein the data set acquisition module is used to obtain histopathological image data and miRNA data of female breast cancer patients who received radiotherapy from the TCGA database;

[0038] A data screening module is used to screen the acquired tissue pathology image data and miRNA data, remove samples with a survival time of less than 3 days, and divide them into positive and negative sample data sets according to the three-year survival time;

[0039] A data feature extraction module is used to extract histopathological image features and miRNA data features from positive and negative sample data sets;

[0040] A prediction model construction module, wherein the prediction model construction module is used to construct a radiotherapy sensitivity prediction model based on tissue pathology image features and miRNA data features;

[0041] The sensitivity prediction and judgment module is used to extract tissue pathology image features and miRNA data features from patient samples and input them into the radiotherapy sensitivity prediction model to predict and judge whether the patient has radiotherapy sensitivity.

[0042] Furthermore, it also includes:

[0043] The normal distribution conversion module is used to use the StandardScaler method to perform standardization processing on the tissue pathology image features and miRNA data features and convert them into normal distribution.

[0044] Through the above technical solution, the present invention provides a method and system for predicting breast cancer radiotherapy sensitivity based on deep learning, which has at least the following beneficial effects:

[0045] 1. The present invention combines tissue pathology image data and miRNA data to construct a breast cancer radiotherapy sensitivity prediction model, which can more accurately determine whether a patient needs radiotherapy, thereby enabling precise treatment of the patient and avoiding the waste of time and harm caused by radiotherapy to the patient.

[0046] 2. This invention deeply integrates tissue pathology image data and miRNA data, and constructs a deep learning model to predict the radiotherapy sensitivity of breast cancer patients, contributing to the personalized treatment of breast cancer patients. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0048] Figure 1 This is a flow chart of the method for predicting breast cancer radiotherapy sensitivity in Example 1 of the present invention;

[0049] Figure 2This is a flowchart of extracting tissue pathology image features from positive and negative sample data sets in Example 1 of the present invention;

[0050] Figure 3 This is a flowchart of constructing a radiotherapy sensitivity prediction model in Example 1 of the present invention;

[0051] Figure 4 This is a functional block diagram of a breast cancer radiotherapy sensitivity prediction system according to a first embodiment of the present invention;

[0052] Figure 5 This is a flow chart of the method for predicting breast cancer radiotherapy sensitivity in Examples 2 and 3 of the present invention;

[0053] Figure 6 This is a functional block diagram of a breast cancer radiotherapy sensitivity prediction system in Example 2 of the present invention;

[0054] Figure 7 This is a flowchart of constructing a radiotherapy sensitivity prediction model in Example 3 of the present invention;

[0055] Figure 8 This is a principle block diagram of the method for predicting breast cancer radiotherapy sensitivity of the present invention.

[0056] In the figure: 100, data acquisition module; 200, data screening module; 300, data feature extraction module; 301, normal distribution conversion module; 400, prediction model construction module; 500, sensitivity prediction and judgment module. DETAILED DESCRIPTION

[0057] To make the above-mentioned objectives, features, and advantages of the present invention more clearly understood, the present invention is further described below in detail with reference to the accompanying drawings and specific embodiments. This will enable a full understanding of how this application uses technical means to solve technical problems and achieve technical effects, and to implement the invention accordingly.

[0058] Those skilled in the art will appreciate that all or part of the steps in the above-mentioned embodiment methods can be accomplished by instructing the relevant hardware through a program. Therefore, the present application may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Furthermore, the present application may take 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.) containing computer-usable program code.

[0059] Example 1

[0060] Please refer to Figures 1-4 , shows a method for predicting breast cancer radiotherapy sensitivity based on deep learning according to embodiment 1 of the present invention, including the following process:

[0061] S1. Obtain histopathological image data and miRNA data of female breast cancer patients who received radiotherapy from the TCGA database;

[0062] S2. Screen the acquired tissue pathology image data and miRNA data, remove samples with a survival time of less than 30 days, and divide them into positive and negative sample data sets based on the three-year survival time;

[0063] The division of positive and negative sample data sets includes:

[0064] Based on the three-year survival time, female breast cancer patients who received radiotherapy were divided into long-term survivors and short-term survivors. Long-term survivors who survived for more than three years after radiotherapy were marked as 1, which is the positive sample data set; short-term survivors who survived for less than three years after radiotherapy were marked as 0, which is the negative sample data set.

[0065] Based on their three-year survival time, these patients are divided into long-term survivors and short-term survivors. Long-term survivors (those who survived for more than three years after radiotherapy) are labeled as 1, representing the positive sample dataset. Patients in the positive sample dataset are sensitive to radiotherapy and can benefit from it.

[0066] For patients with short survival (less than 3 years after radiotherapy), we mark them as 0, which is the negative sample data set. Patients in the negative sample data set are not sensitive to radiotherapy and cannot benefit from radiotherapy.

[0067] S3. Extract tissue pathology image features and miRNA data features from the positive and negative sample data sets. Extract local binary pattern (LBP) image features that can reflect image texture information from the tissue pathology images of breast cancer patients, that is, the tissue pathology image features of the 10-dimensional LBP texture feature vector. For miRNA, use the R package DESeq2 to screen out 13 differentially expressed miRNA data features in the positive and negative sample data sets.

[0068] Extracting histopathology image features from positive and negative sample datasets includes the following processes:

[0069] S311, segmenting the histopathological images in the positive and negative sample data sets, wherein each patient's histopathological image is segmented into a plurality of histopathological image sub-blocks, each of which has a size of 1024×1024;

[0070] S312, discarding the histopathology image sub-blocks in which the tissue area accounts for less than 70% of the histopathology image;

[0071] S313, using the LBP function in the Skimage library to extract several 10-dimensional feature vectors from several tissue pathology image sub-blocks;

[0072] LBP, or local binary pattern, is a texture feature of an image. Its function formula is:

[0073]

[0074] Where c is the given position for LBP solution, with it as the center and R as the radius, a circle is drawn, and n points are evenly taken out on the circumference. I is the indicator function, i p is the grayscale value, i c It is the grayscale value of adjacent pixels. After the R and p required for calculation are given, LBP becomes a function that is only related to position. It is a quantity that describes the relationship between the center point and the surrounding points.

[0075] S314. Combining several 10-dimensional feature vectors into a feature matrix, and removing the median of each column;

[0076] S315 , extracting a tissue pathology image feature of a 10-dimensional LBP texture feature vector from the feature matrix.

[0077] Acquisition of tissue pathology image features: First, the tissue pathology image is segmented. The tissue pathology image of each patient is divided into several 1024*1024 patches (sub-blocks of the tissue pathology image after segmentation). Then, in order to reduce computational overhead, patches with tissue areas accounting for less than 70% are discarded. Through the above operations, the original tissue pathology image of each breast cancer patient is segmented into several 1024*1024 patches. First, the Local binary pattern function in the Skimage library is used to extract a 10-dimensional feature vector from each patch. Then, similarly, several 10-dimensional feature vectors can be extracted from a patient. These feature vectors are combined into a feature matrix, and the median of each column is removed. Finally, a 10-dimensional LBP texture feature vector is extracted from the tissue pathology image of each patient.

[0078] Differential expression analysis is used as a method to extract miRNA data features from positive and negative sample data sets. The process includes:

[0079] The miRNAs in the positive and negative sample datasets were analyzed using the R package DESeq2 (|log2 fold change|>1 and P<0.05), and 13 differentially expressed miRNA data features were finally obtained;

[0080] Differential Expression Analysis: TCGA data contains 1,881 miRNAs per breast cancer patient. We used differential expression analysis as a feature extraction method for miRNA data. We used the R package DESeq2 (|log2 fold change|>1 and P<0.05) to analyze the miRNAs in the positive and negative sample patients in the training set, ultimately identifying a total of 13 differentially expressed miRNAs.

[0081] S4. A radiotherapy sensitivity prediction model was constructed based on the characteristics of tissue pathology images and miRNA data. The final radiotherapy sensitivity prediction model was jointly constructed by BiLSTM, Batch Normalization, Attention, and Neural Network. The parameters of the radiotherapy sensitivity prediction model were adjusted using five-fold cross-validation to obtain the optimal model.

[0082] The step of building a radiotherapy sensitivity prediction model also includes:

[0083] S41, inputting the histopathological image features and the miRNA data features into two identical bidirectional long short-term memory networks, respectively, and concatenating the histopathological image features and the miRNA data features encoded by the bidirectional long short-term memory networks;

[0084] S42, using batch normalization method to normalize the spliced ​​and combined tissue pathology image features and miRNA data features;

[0085] S43, using the attention mechanism to assign weights to the normalized histopathological image features and miRNA data features;

[0086] S44. Input the weighted tissue pathology image features and miRNA data features into a three-layer neural network to obtain a radiotherapy sensitivity prediction model.

[0087] S5. Extracting histopathological image features and miRNA data features from patient samples and inputting them into a radiotherapy sensitivity prediction model to predict and determine whether the patient has radiotherapy sensitivity;

[0088] When a breast cancer patient needs to be predicted, tissue pathology images and features and miRNA data features are extracted from the breast cancer patient's samples, normalized, and then input into the radiotherapy sensitivity prediction model to determine whether the patient has radiotherapy sensitivity.

[0089] If the probability value output by the radiotherapy sensitivity prediction model is greater than or equal to the preset threshold, it will be predicted as a positive sample;

[0090] If the probability value output by the radiotherapy sensitivity prediction model is less than the preset threshold, it is predicted as a negative sample.

[0091] When the probability value output by the radiotherapy sensitivity prediction model is greater than or equal to the threshold, it is predicted as a positive sample; otherwise, it is predicted as a negative sample. The threshold is typically set to 0.5 and can be adjusted appropriately based on the actual problem. However, AUC, as a comprehensive indicator, is not affected by the threshold and ranges from 0 to 1. The closer the value is to 1, the better the performance of the radiotherapy sensitivity prediction model.

[0092] This embodiment further provides a system applied to the above-mentioned method for predicting breast cancer radiotherapy sensitivity, comprising:

[0093] The data acquisition module 100 is used to acquire histopathological image data and miRNA data of female breast cancer patients who received radiotherapy from the TCGA database;

[0094] The data screening module 200 is used to screen the acquired tissue pathology image data and miRNA data, remove samples with a survival time of less than 30 days, and divide them into positive and negative sample data sets according to the three-year survival time;

[0095] The data feature extraction module 300 is used to extract histopathological image features and miRNA data features from the positive and negative sample data sets;

[0096] The prediction model construction module 400 is used to construct a radiotherapy sensitivity prediction model based on the characteristics of tissue pathology images and miRNA data;

[0097] The sensitivity prediction and judgment module 500 is used to extract tissue pathology image features and miRNA data features from patient samples and input them into the radiotherapy sensitivity prediction model to predict and judge whether the patient has radiotherapy sensitivity.

[0098] Through this embodiment, a breast cancer radiotherapy sensitivity prediction model is constructed by combining tissue pathology image data and miRNA data, which can more accurately determine whether a patient needs radiotherapy, thereby enabling precise treatment of the patient and avoiding waste of time and harm caused by radiotherapy to the patient.

[0099] Example 2

[0100] Please refer to Figure 5-Figure 6 , shows a method for predicting breast cancer radiotherapy sensitivity based on deep learning according to embodiment 2 of the present invention, including the following process:

[0101] S1. Obtain histopathological image data and miRNA data of female breast cancer patients who received radiotherapy from the TCGA database;

[0102] S2. Screen the acquired tissue pathology image data and miRNA data, remove samples with a survival time of less than 30 days, and divide them into positive and negative sample data sets based on the three-year survival time;

[0103] The division of positive and negative sample data sets includes:

[0104] Based on the three-year survival time, female breast cancer patients who received radiotherapy were divided into long-term survivors and short-term survivors. Long-term survivors who survived for more than three years after radiotherapy were marked as 1, which is the positive sample data set; short-term survivors who survived for less than three years after radiotherapy were marked as 0, which is the negative sample data set.

[0105] Based on their three-year survival time, these patients are divided into long-term survivors and short-term survivors. Long-term survivors (those who survived for more than three years after radiotherapy) are labeled as 1, representing the positive sample dataset. Patients in the positive sample dataset are sensitive to radiotherapy and can benefit from it.

[0106] For patients with short survival (less than 3 years after radiotherapy), we mark them as 0, which is the negative sample data set. Patients in the negative sample data set are not sensitive to radiotherapy and cannot benefit from radiotherapy.

[0107] S3. Extract tissue pathology image features and miRNA data features from the positive and negative sample data sets. Extract local binary pattern (LBP) image features that can reflect image texture information from the tissue pathology images of breast cancer patients, that is, the tissue pathology image features of the 10-dimensional LBP texture feature vector. For miRNA, use the R package DESeq2 to screen out 13 differentially expressed miRNA data features in the positive and negative sample data sets.

[0108] S31. StandardScaler method was used to standardize the features of tissue pathology images and miRNA data and transform them into normal distribution;

[0109] After the above processing, 10-dimensional histopathology image features and 13-dimensional miRNA data features can be obtained respectively. Then, the StandardScaler method is used to standardize the 10-dimensional histopathology image features and 13-dimensional miRNA data features. StandardScaler can transform the histopathology image feature and miRNA data feature matrices into a normal distribution. Its calculation formula is as follows:

[0110]

[0111] Where x is the characteristic value of the input sample, μ and δ represent the mean and variance of the sample characteristics respectively, and x' represents the standardized sample characteristics.

[0112] S4. A radiotherapy sensitivity prediction model was constructed based on the characteristics of tissue pathology images and miRNA data. The final radiotherapy sensitivity prediction model was jointly constructed by BiLSTM, Batch Normalization, Attention, and Neural Network. The parameters of the radiotherapy sensitivity prediction model were adjusted using five-fold cross-validation to obtain the optimal model.

[0113] S5. Extracting histopathological image features and miRNA data features from patient samples and inputting them into a radiotherapy sensitivity prediction model to predict and determine whether the patient has radiotherapy sensitivity;

[0114] When a breast cancer patient needs to be predicted, tissue pathology images and features and miRNA data features are extracted from the breast cancer patient's samples, normalized, and then input into the radiotherapy sensitivity prediction model to determine whether the patient has radiotherapy sensitivity.

[0115] If the probability value output by the radiotherapy sensitivity prediction model is greater than or equal to the preset threshold, it will be predicted as a positive sample;

[0116] If the probability value output by the radiotherapy sensitivity prediction model is less than the preset threshold, it is predicted as a negative sample.

[0117] When the probability value output by the radiotherapy sensitivity prediction model is greater than or equal to the threshold, it is predicted as a positive sample; otherwise, it is predicted as a negative sample. The threshold is typically set to 0.5 and can be adjusted appropriately based on the actual problem. However, AUC, as a comprehensive indicator, is not affected by the threshold and ranges from 0 to 1. The closer the value is to 1, the better the performance of the radiotherapy sensitivity prediction model.

[0118] This embodiment further provides a system applied to the above-mentioned method for predicting breast cancer radiotherapy sensitivity, comprising:

[0119] The data acquisition module 100 is used to acquire histopathological image data and miRNA data of female breast cancer patients who received radiotherapy from the TCGA database;

[0120] The data screening module 200 is used to screen the acquired tissue pathology image data and miRNA data, remove samples with a survival time of less than 30 days, and divide them into positive and negative sample data sets according to the three-year survival time;

[0121] The data feature extraction module 300 is used to extract histopathological image features and miRNA data features from the positive and negative sample data sets;

[0122] Normal distribution conversion module 301, which is used to use the StandardScaler method to perform standardization processing on the tissue pathology image features and miRNA data features and convert them into normal distribution;

[0123] The prediction model construction module 400 is used to construct a radiotherapy sensitivity prediction model based on the characteristics of tissue pathology images and miRNA data;

[0124] The sensitivity prediction and judgment module 500 is used to extract tissue pathology image features and miRNA data features from patient samples and input them into the radiotherapy sensitivity prediction model to predict and judge whether the patient has radiotherapy sensitivity.

[0125] Through this embodiment, tissue pathology image data and miRNA data are deeply integrated, and a deep learning model is constructed to predict the radiotherapy sensitivity of breast cancer patients, contributing to the personalized treatment of breast cancer patients.

[0126] This embodiment is made on the basis of the first embodiment and has the same beneficial effects as the first embodiment. The same or similar parts can be referred to each other and will not be described in detail here.

[0127] Example 3

[0128] Please refer to Figure 5 and Figure 7 , shows a method for predicting breast cancer radiotherapy sensitivity based on deep learning according to embodiment 3 of the present invention, including the following process:

[0129] S1. Obtain histopathological image data and miRNA data of female breast cancer patients who received radiotherapy from the TCGA database;

[0130] S2. Screen the acquired tissue pathology image data and miRNA data, remove samples with a survival time of less than 30 days, and divide them into positive and negative sample data sets based on the three-year survival time;

[0131] The division of positive and negative sample data sets includes:

[0132] Based on the three-year survival time, female breast cancer patients who received radiotherapy were divided into long-term survivors and short-term survivors. Long-term survivors who survived for more than three years after radiotherapy were marked as 1, which is the positive sample data set; short-term survivors who survived for less than three years after radiotherapy were marked as 0, which is the negative sample data set.

[0133] Based on their three-year survival time, these patients are divided into long-term survivors and short-term survivors. Long-term survivors (those who survived for more than three years after radiotherapy) are labeled as 1, representing the positive sample dataset. Patients in the positive sample dataset are sensitive to radiotherapy and can benefit from it.

[0134] For patients with short survival (less than 3 years after radiotherapy), we mark them as 0, which is the negative sample data set. Patients in the negative sample data set are not sensitive to radiotherapy and cannot benefit from radiotherapy.

[0135] S3. Extract tissue pathology image features and miRNA data features from the positive and negative sample data sets. Extract local binary pattern (LBP) image features that can reflect image texture information from the tissue pathology images of breast cancer patients, that is, the tissue pathology image features of the 10-dimensional LBP texture feature vector. For miRNA, use the R package DESeq2 to screen out 13 differentially expressed miRNA data features in the positive and negative sample data sets.

[0136] S31. StandardScaler method was used to standardize the features of tissue pathology images and miRNA data and transform them into normal distribution;

[0137] After the above processing, 10-dimensional histopathology image features and 13-dimensional miRNA data features can be obtained respectively. Then, the StandardScaler method is used to standardize the 10-dimensional histopathology image features and 13-dimensional miRNA data features. StandardScaler can transform the histopathology image feature and miRNA data feature matrices into a normal distribution. Its calculation formula is as follows:

[0138]

[0139] Where x is the characteristic value of the input sample, μ and δ represent the mean and variance of the sample characteristics respectively, and x' represents the standardized sample characteristics.

[0140] S4. A radiotherapy sensitivity prediction model was constructed based on the characteristics of tissue pathology images and miRNA data. The final radiotherapy sensitivity prediction model was jointly constructed by BiLSTM, Batch Normalization, Attention, and Neural Network. The parameters of the radiotherapy sensitivity prediction model were adjusted using five-fold cross-validation to obtain the optimal model.

[0141] The step of building a radiotherapy sensitivity prediction model also includes:

[0142] S41, inputting the histopathological image features and the miRNA data features into two identical bidirectional long short-term memory networks, respectively, and concatenating the histopathological image features and the miRNA data features encoded by the bidirectional long short-term memory networks;

[0143] S42, using batch normalization method to normalize the spliced ​​and combined tissue pathology image features and miRNA data features;

[0144] S43, using the attention mechanism to assign weights to the normalized histopathological image features and miRNA data features;

[0145] S44. Input the weighted tissue pathology image features and miRNA data features into a three-layer neural network to obtain a radiotherapy sensitivity prediction model.

[0146] S45. Use the five-fold cross-validation method to randomly divide the positive and negative sample data sets into training sets and validation sets;

[0147] S46. inputting the training set into a radiotherapy sensitivity prediction model for training;

[0148] S47. Input the validation set into the radiotherapy sensitivity prediction model to adjust parameters, and finally obtain the optimal radiotherapy sensitivity prediction model.

[0149] Through this embodiment, the constructed radiotherapy sensitivity prediction model can be trained, and the optimal radiotherapy sensitivity prediction model can be obtained using a validation set, thereby improving the accuracy of breast cancer radiotherapy sensitivity prediction.

[0150] Please refer to Figure 8 The principle block diagram of the entire breast cancer radiotherapy sensitivity prediction method clearly and completely explains the core technical solution of the present invention. From it, we can fully understand the use process of the entire breast cancer radiotherapy sensitivity prediction method. At the same time, according to the above implementation plan, patients with breast cancer can be predicted and judged whether they have radiotherapy sensitivity.

[0151] This embodiment is made on the basis of the first embodiment and has the same beneficial effects as the first embodiment. The same or similar parts can be referred to each other and will not be described in detail here.

[0152] Experimental example

[0153] The dataset for this experiment consists of two parts: a training set and a test set.

[0154] The team downloaded histopathological image data and miRNA data from female breast cancer patients who had undergone radiotherapy from the TCGA database. Samples with survival times less than 30 days were removed, resulting in a total of 276 breast cancer patient samples. Based on three-year survival, these patients were divided into positive samples (survival time greater than 3 years after radiotherapy) and negative samples (survival time less than 3 years after radiotherapy). These samples were then randomly divided into a training set (N = 202) and a test set (N = 74).

[0155] In order to select the optimal model and adjust the model parameters, we divide the obtained training set into training set and validation set according to the five-fold cross-validation method. The training set is used to build and train the model, and the validation set is used to adjust the model parameters to finally obtain the optimal model.

[0156] We first input the 10-dimensional histopathology image features and the 13-dimensional miRNA features into two identical Bidirectional Long Short-Term Memory (BiLSTM) networks. Next, we concatenate the two features encoded by the BiLSTM and perform batch normalization on the combined features to obtain the encoded features. Finally, we weight the combined features using an attention mechanism. Finally, we input the weighted features into a three-layer neural network to obtain the final prediction results.

[0157] The validation criteria we used include sensitivity (SEN), specificity (SPE), precision (PRE), accuracy (ACC), and F1 score (F1), which are calculated as follows:

[0158]

[0159]

[0160]

[0161]

[0162]

[0163] Among them, TP (True Positive) represents the number of patients correctly predicted to be sensitive to radiotherapy; TN (True Negative) represents the number of patients correctly predicted to be insensitive to radiotherapy; FP (False Positive) represents the number of patients incorrectly predicted to be sensitive to radiotherapy; FN (False Negative) represents the number of patients incorrectly predicted to be insensitive to radiotherapy.

[0164] In addition, this article also uses AUC (The Area Under the Receiver Operating Characteristic (ROC) Curve) to evaluate the overall performance of the model. Generally speaking, the five indicators of SEN, SPE, PRE, F1 and ACC will be affected by the threshold. When the probability value output by the model is greater than or equal to the threshold, it will be predicted as a positive sample, otherwise it will be predicted as a negative sample. Generally, the threshold is set to 0.5 and can be adjusted appropriately according to the actual problem. However, as a comprehensive indicator, AUC is not affected by the threshold. Its range is between 0 and 1. The closer its value is to 1, the better the performance of the model.

[0165] Verification Example

[0166] In order to verify the superiority of our model, we selected some excellent tools for comparison, including:

[0167] Two deep learning models, GPDBN and MDNNMD, and three machine learning models, support vector machine (SVM), logistic regression (LR), and random forest (RF), were used. We named the breast cancer radiosensitivity prediction model WMBAN. Table 1 fully demonstrates the performance comparison of the WMBAN model with the other five models using six evaluation metrics (SEN, SPE, PRE, F1, ACC, and AUC).

[0168]

[0169] Table 1 Performance comparison of WMBAN model and existing models

[0170] As shown in Table 1, the WMBAN model achieves the best performance in terms of PRE, F1, ACC, and AUC, and also performs well in SEN and SPE. As a comprehensive metric for evaluating a model, the WMBAN model significantly outperforms the other five models, demonstrating its effectiveness.

[0171] The present invention combines tissue pathology image data and miRNA data to construct a breast cancer radiotherapy sensitivity prediction model, which can more accurately determine whether a patient needs radiotherapy, thereby enabling precise treatment of the patient and avoiding waste of time and harm caused by radiotherapy to the patient.

[0172] This paper deeply integrates tissue pathology image data and miRNA data, and constructs a deep learning model to predict the radiotherapy sensitivity of breast cancer patients, contributing to the personalized treatment of breast cancer patients.

[0173] Each embodiment in this specification is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to in detail. For the above embodiments, since they are basically similar to the method embodiments, the description is relatively simple. For relevant parts, please refer to the partial description of the method embodiments.

[0174] The above embodiments provide a detailed introduction to the present invention. Specific examples are used herein to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core ideas. At the same time, for those skilled in the art, according to the ideas of the present invention, there may be changes in the specific implementation methods and application scopes. In summary, the contents of this specification should not be understood as limiting the present invention.

Claims

1. A method for predicting breast cancer radiotherapy sensitivity based on deep learning, characterized in that: The following processes are included: Obtain histopathological image data and miRNA data of female breast cancer patients who received radiotherapy from the TCGA database; The acquired tissue pathology image data and miRNA data were screened, samples with a survival time of less than 30 days were removed, and the data were divided into positive and negative sample data sets based on the three-year survival time; Extract histopathological image features and miRNA data features from positive and negative sample datasets; A radiotherapy sensitivity prediction model was constructed based on histopathological image features and miRNA data features; Extracting histopathological image features and miRNA data features from patient samples and inputting them into the radiotherapy sensitivity prediction model to predict and determine whether the patient has radiotherapy sensitivity; The process of building a radiotherapy sensitivity prediction model also includes: The histopathological image features and miRNA data features are input into two identical bidirectional long short-term memory networks respectively, and the histopathological image features and miRNA data features encoded by the bidirectional long short-term memory network are spliced; Batch normalization method is used to normalize the spliced ​​and combined tissue pathology image features and miRNA data features; The attention mechanism is used to assign weights to the normalized histopathological image features and miRNA data features; The weighted tissue pathology image features and miRNA data features were input into a three-layer neural network to obtain a radiotherapy sensitivity prediction model.

2. The method for predicting breast cancer radiotherapy sensitivity according to claim 1, wherein: The division of positive and negative sample data sets includes: Female breast cancer patients who received radiotherapy were divided into long-term survivors and short-term survivors based on three-year survival time; Patients with long-term survival of more than 3 years after radiotherapy are marked as 1, which is a positive sample data set; Patients with short survival time less than 3 years after radiotherapy are marked as 0, which is a negative sample data set.

3. The method for predicting breast cancer radiotherapy sensitivity according to claim 1, wherein: After extracting the tissue pathology image features and the miRNA data features, the method also includes: using the StandardScaler method to standardize the tissue pathology image features and the miRNA data features and convert them into a normal distribution.

4. The method for predicting breast cancer radiotherapy sensitivity according to claim 1, wherein: Extracting histopathological image features from positive and negative sample datasets The following processes are included: The histopathological images in the positive and negative sample data sets are segmented, and the histopathological image of each patient is segmented into several histopathological image sub-blocks; Discard the histopathology image sub-blocks where the tissue area accounts for less than 70% of the histopathology image; Use the LBP function in the Skimage library to extract several 10-dimensional feature vectors from several tissue pathology image sub-blocks; Combine several 10-dimensional feature vectors into a feature matrix and remove the median of each column; A 10-dimensional LBP texture feature vector of tissue pathology image features is extracted from the feature matrix.

5. The method for predicting breast cancer radiotherapy sensitivity according to claim 4, wherein: The size of the histopathology image sub-block is 1024×1024.

6. The method for predicting breast cancer radiotherapy sensitivity according to claim 1, wherein: Differential expression analysis is used as a method to extract miRNA data features from positive and negative sample data sets. The process includes: The miRNAs in the positive and negative sample datasets were analyzed using the R package DESeq2 (|log2 fold change|>1 and P <0.05), and 13 differentially expressed miRNA data features were finally obtained.

7. The method for predicting breast cancer radiotherapy sensitivity according to claim 1, wherein: The step of building a radiotherapy sensitivity prediction model also includes: The five-fold cross-validation method is used to randomly divide the positive and negative sample data sets into training sets and validation sets; The training set is input into the radiotherapy sensitivity prediction model for training; The validation set was input into the radiotherapy sensitivity prediction model for parameter adjustment, and finally the optimal radiotherapy sensitivity prediction model was obtained.

8. A system for implementing the method for predicting breast cancer radiotherapy sensitivity according to any one of claims 1 to 7, characterized in that: include: A data acquisition module (100), wherein the data acquisition module (100) is used to acquire histopathological image data and miRNA data of female breast cancer patients who received radiotherapy from the TCGA database; A data screening module (200), the data screening module (200) is used to screen the acquired tissue pathology image data and miRNA data, remove samples with a survival time of less than 30 days, and divide them into positive and negative sample data sets according to the three-year survival time; A data feature extraction module (300), the data feature extraction module (300) is used to extract tissue pathology image features and miRNA data features from positive and negative sample data sets; A prediction model construction module (400), wherein the prediction model construction module (400) is used to construct a radiotherapy sensitivity prediction model based on tissue pathology image features and miRNA data features; The sensitivity prediction and judgment module (500) is used to extract tissue pathology image features and miRNA data features from patient samples and input them into a radiotherapy sensitivity prediction model to predict and judge whether the patient has radiotherapy sensitivity.

9. The system according to claim 8, characterized in that Also includes: A normal distribution conversion module (301) is used to use the StandardScaler method to perform standardization processing on the tissue pathology image features and the miRNA data features to convert them into normal distribution.

Images