Breast cancer risk assessment method and system

Abstract

The present invention relates to a method, system, computer device and storage medium for assessing the risk of breast cancer, which obtains original data based on breast cancer including gene sequencing data and medical imaging data. Then, feature extraction is performed on the original data, mutation site features and gene expression level features are obtained from the gene sequencing data, and shape and texture features of the tumor area are extracted from the medical imaging data. A breast cancer risk assessment model based on a Transformer neural network model is constructed using the extracted features. Finally, the initial model is trained to obtain a trained assessment model. The method achieves efficient and accurate assessment of the risk of breast cancer by fusing gene sequencing data and medical imaging data and combining the feature extraction and learning capabilities of the Transformer neural network.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and in particular to a method, system, computer device and storage medium for assessing the risk of breast cancer. Background Art

[0002] Breast cancer is one of the most common malignant tumors in women, with millions of new cases worldwide each year. With the advancement of medical technology, the importance of early screening and diagnosis of breast cancer has become increasingly prominent. Although traditional breast cancer screening methods, such as mammography, have a certain effect in early detection, their accuracy and sensitivity are affected by many factors, resulting in some high-risk groups being unable to receive timely intervention. In addition, the complexity of genetic factors, lifestyle habits, and environmental factors increases the difficulty of breast cancer risk assessment. The existing risk assessment models still have much room for improvement in terms of prediction accuracy and individualized risk assessment.

[0003] Most current breast cancer risk assessment methods are based on statistical models and epidemiological data. Although these methods have shown certain application value in large-sample studies, they often show limitations when faced with individual differences. For example, the common Gail model and Claus model take into account family history and some risk factors, but do not take into account individual gene mutations. This results in many high-risk groups not being effectively identified, and low-risk groups may be over-screened, increasing unnecessary medical burdens and psychological pressure.

[0004] Existing methods for identifying breast cancer cannot effectively address individual differences, resulting in technical problems such as failure to accurately identify high-risk groups and over-screening of low-risk groups, and no effective solution has yet been found. Summary of the invention

[0005] The main purpose of the present invention is to provide a breast cancer risk assessment method, system, computer device and storage medium to solve the technical problem that existing breast cancer identification methods cannot effectively address individual differences, resulting in inaccurate identification of high-risk groups and over-screening of low-risk groups.

[0006] To achieve the above-mentioned object, the present invention provides a method for assessing the risk of breast cancer, comprising: obtaining original data based on breast cancer, wherein the original data based on breast cancer at least includes: gene sequencing data and medical imaging data; performing feature extraction on the original data to obtain a feature set, wherein the feature set at least includes: mutation site features and gene expression level features in the gene sequencing data, and shape and texture features of the tumor area in the medical impact data; using the feature set to construct a breast cancer risk assessment model based on a Transformer neural network model, and training the initial breast cancer risk assessment model based on the Transformer neural network model to obtain a trained breast cancer risk assessment model based on the Transformer neural network model.

[0007] Optionally, after obtaining the original data based on breast cancer, the method for assessing the risk of breast cancer also includes: adjusting the image pixels in the medical image data to preset pixel values ​​to obtain first medical image data; using the Z-SCORE method to map the pixel values ​​of the first medical image data to a preset interval to obtain second medical image data; using an anisotropic diffusion filter to remove image noise in the second medical image data to obtain third medical image data; using contrast-limited adaptive histogram equalization to adjust the contrast of the third medical image data to obtain fourth medical image data; using a neural network U-NET to segment the fourth medical image and extract tumor areas and organ boundaries in the fourth medical image data; marking the tumor areas and organ boundaries in the fourth medical image data to obtain preprocessed medical image data, wherein the preprocessed medical image data is used to execute the step of "extracting features from the original data to obtain a feature set".

[0008] Optionally, after obtaining the original data based on breast cancer, the disease risk assessment method also includes: filtering the reads containing joint contamination in the gene sequencing data, and removing duplicate reads in the gene sequencing data to obtain second gene sequencing data; aligning the reads in the second gene sequencing data to the reference genome to obtain third gene sequencing data; annotating the aligned reads in the third gene sequencing data to obtain preprocessed gene sequencing data, wherein the preprocessed gene sequencing data is used to execute the step of "extracting features from the original data to obtain a feature set".

[0009] Optionally, feature extraction is performed on the original data to obtain a feature set, including: dividing the image in the preprocessed medical imaging data into multiple small blocks, and flattening each small block into a one-dimensional vector to obtain a first vector; using a linear layer in a Transformer neural network model to embed the first vector into a preset dimension to obtain a second vector; using a learned position encoding to generate a code for each position of the second vector to obtain a third vector; connecting the attention head in the Transformer neural network model to the feedforward neural network in the Transformer neural network model, and feeding the third vector into the feedforward neural network in the Transformer neural network model to obtain shape and texture features of the tumor area in the medical imaging data.

[0010] Optionally, feature extraction is performed on the original data to obtain a feature set, including: extracting mutation site information of each sample in the preprocessed gene sequencing data, counting the type and frequency of each sample mutation site based on the mutation site information, and constructing a mutation feature matrix based on the type and frequency of each sample mutation site; calculating the expression level of the gene in the preprocessed gene sequencing data, and constructing an expression feature matrix based on the expression level of the gene; performing feature transformation on the mutation feature matrix and the expression feature matrix to convert them into dense vectors to obtain fifth gene sequencing data; feeding the vectorized fifth gene sequencing data into a preset Transformer-based neural network model; using a preset convolutional layer in the Transformer-based neural network model to extract initial features of mutation sites and initial features of gene expression levels in the fifth gene sequencing data, and outputting a feature map; using a preset pooling layer in the Transformer-based neural network model to perform dimensionality reduction on the feature map to obtain a first important feature map in the mutation site and a second important feature map of the gene expression level in the fifth gene sequencing data; flattening the first important feature map and the second important feature map, and connecting them to a fully connected layer to obtain mutation site features and gene expression level features of the gene sequencing data.

[0011] Optionally, the feature set is used to construct a breast cancer disease assessment model based on the Transformer neural network model, including: inputting the feature set into the normalization layer in the Transformer neural network model, normalizing the shape and texture features of the tumor area in each layer of the input medical imaging data and the mutation site features and gene expression level features of the gene sequencing data to obtain a first fusion feature; adding the first fusion feature to the vectorized fifth gene sequencing data and the first vector to obtain a first fusion feature; inputting the first fusion feature into the normalization layer in the Transformer neural network model to obtain a second fusion feature; adding the second fusion feature to the vectorized fifth gene sequencing data and the first vector to obtain a first fusion feature; The combined feature is added to the vectorized fifth gene sequencing data and the first vector to form a third fusion feature; the third fusion feature is input into the normalization layer in the Transformer neural network model to obtain a fourth fusion feature; the fourth fusion feature is added to the vectorized fifth gene sequencing data and the first vector to form a fifth fusion feature; the fifth fusion feature is input into the normalization layer in the Transformer neural network model to obtain a sixth fusion feature; the sixth fusion feature is input into the Softmax layer in the Transformer neural network model to obtain a breast cancer morbidity assessment model based on the Transformer neural network model.

[0012] Optionally, the initial model for breast cancer prevalence assessment based on the Transformer neural network model is trained to obtain a trained model for breast cancer prevalence assessment based on the Transformer neural network model, including: using a cross entropy loss function to calculate the loss of the initial model for breast cancer prevalence assessment based on the Transformer neural network model; using an RMSprop optimizer to adjust the parameters of the initial model for breast cancer prevalence assessment based on the Transformer neural network model; using an ultra-bandwidth optimization method to optimize the parameter combination of the initial model for breast cancer prevalence assessment based on the Transformer neural network model, repeatedly screening and allocating resources, and screening out the best hyperparameter combination; dividing the original data into multiple small batches, using a small batch of data each time for forward propagation and back propagation, repeating the above steps for multiple training cycles, and training the initial model for breast cancer prevalence assessment based on the Transformer neural network model multiple times to obtain the trained model for breast cancer prevalence assessment based on the Transformer neural network model.

[0013] The present invention also provides a breast cancer risk assessment system, comprising: an acquisition unit, used to acquire original data based on breast cancer, wherein the original data based on breast cancer includes at least: gene sequencing data and medical imaging data; a feature extraction unit, used to extract features from the original data to obtain a feature set, wherein the feature set includes at least: mutation site features and gene expression level features in the gene sequencing data, and shape and texture features of the tumor area in the medical impact data; a training unit, used to use the feature set to construct a breast cancer risk assessment model based on the Transformer neural network model, and train the initial breast cancer risk assessment model based on the Transformer neural network model to obtain a trained breast cancer risk assessment model based on the Transformer neural network model.

[0014] The present invention also provides a computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of any one of the above methods when executing the computer program.

[0015] The present invention also provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the steps of any of the above-mentioned methods are implemented.

[0016] The present invention provides a breast cancer risk assessment method, system, computer device and storage medium, which collect and integrate the patient's gene sequencing data and medical imaging data. These data contain important information related to breast cancer, such as gene mutations and expression levels, as well as tumor morphology and texture features captured by medical images (such as MRI or ultrasound). These data provide a basis for subsequent feature extraction and model construction.

[0017] In order to extract useful features from the raw data, the present invention adopts a preset neural network algorithm. The algorithm can effectively identify and retain the most representative and predictive features, extract mutation site features and gene expression level features from gene sequencing data, and extract the shape and texture features of the tumor area from medical imaging data. Through feature extraction, the data dimension can be significantly reduced, and the training efficiency and prediction accuracy of the model can be improved. Using the features obtained from feature extraction, a breast cancer assessment model based on the Transformer neural network is constructed. The Transformer model, with its powerful self-attention mechanism, can process complex high-dimensional data, capture the association between different features, and improve the accuracy of prediction. Next, the model is trained with a large-scale breast cancer data set so that it can learn and identify the characteristic patterns of breast cancer risk, thereby achieving accurate individualized risk assessment.

[0018] In general, the present invention integrates gene sequencing data and medical imaging data, and provides a more comprehensive and accurate disease risk assessment through comprehensive analysis of multi-dimensional data. Secondly, the use of neural network algorithms can effectively identify and retain the most predictive features, improving the training efficiency and prediction performance of the model. Compared with traditional statistical models and shallow neural network models, the Transformer model has a stronger self-attention mechanism and the ability to process high-dimensional data, which can capture the interaction between complex features, improve the accuracy of prediction and the accuracy of individualized risk assessment. In addition, the present invention overcomes the limitations of existing models in the identification of individualized differences by conducting a comprehensive evaluation of gene mutations and medical images.

[0019] In summary, the present invention not only improves the accuracy of identifying people at high risk of breast cancer and reduces the possibility of over-screening in people at low risk, but also achieves a significant breakthrough in individualized difference assessment. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 This is a schematic diagram of the steps of a method for assessing the risk of breast cancer in one embodiment of the present invention;

[0021] Figure 2 is a structural block diagram of a breast cancer risk assessment system in one embodiment of the present invention;

[0022] Figure 3 It is a schematic block diagram of the structure of a computer device according to an embodiment of the present invention.

[0023] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0024] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0025] Reference Figure 1 The embodiment of the present invention provides a method for assessing the risk of breast cancer, comprising:

[0026] S1, obtaining original data based on breast cancer, wherein the original data based on breast cancer at least includes: gene sequencing data and medical imaging data.

[0027] S2, extracting features from the original data to obtain a feature set, wherein the feature set includes at least: mutation site features and gene expression level features in gene sequencing data, and shape and texture features of the tumor area in medical impact data.

[0028] S3, using the feature set to construct a breast cancer morbidity assessment model based on the Transformer neural network model.

[0029] S4, training the initial model of breast cancer prevalence assessment based on the Transformer neural network model to obtain a trained breast cancer prevalence assessment model based on the Transformer neural network model.

[0030] The present invention provides a method for assessing the risk of breast cancer, which aims to solve the problem of insufficient recognition of individual differences in the prior art, ensure that high-risk groups can be accurately identified and receive timely intervention, and avoid unnecessary medical burden and psychological pressure caused by excessive screening of low-risk groups. First, the method collects and integrates the patient's gene sequencing data and medical imaging data. These data contain important information related to breast cancer, such as gene mutations and expression levels, as well as tumor morphology and texture features captured by medical images (such as MRI or ultrasound). These data provide a basis for subsequent feature extraction and model construction.

[0031] The present invention extracts mutation site features and gene expression level features from gene sequencing data, and extracts shape and texture features of tumor areas from medical imaging data by identifying and retaining the most representative and predictive features. Through feature extraction, the data dimension can be significantly reduced, and the training efficiency and prediction accuracy of the model can be improved. Using the features obtained from feature extraction, a breast cancer assessment model based on a Transformer neural network is constructed. The Transformer model, with its powerful self-attention mechanism, can process complex high-dimensional data, capture the association between different features, and improve the accuracy of prediction. Next, the model is trained using a large-scale breast cancer data set so that it can learn and identify the characteristic patterns of breast cancer risk, thereby achieving accurate individualized risk assessment.

[0032] In general, the present invention integrates gene sequencing data and medical imaging data, and provides a more comprehensive and accurate disease risk assessment through comprehensive analysis of multi-dimensional data. Secondly, the use of neural network algorithms can effectively identify and retain the most predictive features, improving the training efficiency and predictive performance of the model. Compared with traditional statistical models and shallow neural network models, the Transformer model has a stronger self-attention mechanism and the ability to process high-dimensional data, which can capture the interactions between complex features, improve the accuracy of predictions and the precision of individualized risk assessments.

[0033] In summary, through the above method, the present invention provides a clear, complete and feasible method for assessing the risk of breast cancer. This method not only improves the accuracy of identifying people at high risk of breast cancer, reduces the possibility of overscreening of people at low risk, but also achieves a significant breakthrough in individualized difference assessment.

[0034] In one example, after obtaining the original data based on breast cancer, the breast cancer risk assessment method also includes: adjusting the image pixels in the medical image data to preset pixel values ​​to obtain first medical image data; using the Z-SCORE method to map the pixel values ​​of the first medical image data to a preset interval to obtain second medical image data; using an anisotropic diffusion filter to remove image noise in the second medical image data to obtain third medical image data; using contrast-limited adaptive histogram equalization to adjust the contrast of the third medical image data to obtain fourth medical image data; using a neural network U-NET to segment the fourth medical image and extract tumor areas and organ boundaries in the fourth medical image data; marking the tumor areas and organ boundaries in the fourth medical image data to obtain preprocessed medical image data, wherein the preprocessed medical image data is used to execute the step of "extracting features from the original data to obtain a feature set".

[0035] In this example, the breast cancer risk assessment method performs detailed image processing steps after obtaining the raw data to ensure that high-quality feature information is extracted from the medical image data. The following is an explanation of each step: First, after obtaining the raw data based on breast cancer, the method includes adjusting the image pixels in the medical image data to a preset pixel value to obtain the first medical image data. This step is to standardize the resolution and size of the image so that subsequent processing and analysis can be performed on a unified basis. Next, the pixel values ​​of the first medical image data are mapped to a preset interval using the Z-SCORE method to obtain the second medical image data. The Z-SCORE method is a standardization technique that adjusts the mean value of the data to 0 and the standard deviation to 1 so that the data distribution of different images can be compared on the same scale. After that, an anisotropic diffusion filter is used to remove image noise in the second medical image data and retain the detail information in the image. This filter can retain the edges and details in the image while removing noise, ensuring that the quality of the image is not over-smoothed. Next, contrast-limited adaptive histogram equalization (CLAHE) is used to increase the contrast of the image to obtain the third medical image data. CLAHE is a technology that enhances image contrast, which can make details clearer and thus improve the effect of feature extraction.

[0036] After the image preprocessing is completed, the neural network U-NET is used to segment the third medical image to extract the tumor area and organ boundaries in the image. U-NET is a neural network model widely used in medical image segmentation. It has high accuracy and robustness and can accurately segment the area of ​​interest. After that, the tool LABEL ME can be used to annotate the tumor area and organ boundaries in the labeled medical image data. LABEL ME is an image annotation tool that can accurately annotate important areas in the image manually. These labeled medical image data are used in the subsequent feature extraction steps.

[0037] In summary, the above image processing steps ensure the accuracy and quality of the shape and texture features of the tumor area extracted from the medical imaging data. These high-quality feature data play a key role in the subsequent breast cancer risk assessment model, which can significantly improve the prediction accuracy of the model and the accuracy of individualized risk assessment.

[0038] In one example, after obtaining the original data based on breast cancer, the disease risk assessment method also includes: filtering the reads containing joint contamination in the gene sequencing data, and removing the duplicate reads in the gene sequencing data to obtain second gene sequencing data; aligning the reads in the second gene sequencing data to the reference genome to obtain third gene sequencing data; annotating the aligned reads in the third gene sequencing data to obtain preprocessed gene sequencing data, wherein the preprocessed gene sequencing data is used to execute the step of "extracting features from the original data to obtain a feature set".

[0039] In this example, the breast cancer risk assessment method performs detailed processing steps on the gene sequencing data after obtaining the raw data to ensure that high-quality feature information is extracted from the gene sequencing data. The following is an explanation of each step: First, after obtaining the raw data based on breast cancer, the FASTQC tool can be used to filter out reads containing adapter contamination in the gene sequencing data. FASTQC is a quality control tool used to detect various quality problems in gene sequencing data, including adapter contamination. This step ensures that the data used in subsequent analysis is of high quality. Next, the Trimmomatic tool is used to remove duplicate reads to obtain the second gene sequencing data. Trimmomatic is a flexible read trimming tool that can effectively remove duplicate reads and low-quality reads in sequencing data, further improving the quality and reliability of the data.

[0040] Afterwards, the BWA tool is used to align the reads in the second gene sequencing data to the reference genome to obtain the third gene sequencing data. BWA (Burrows-Wheeler Aligner) is an efficient alignment tool that can accurately align the reads in the gene sequencing data to the reference genome to ensure that each read is in the correct position in the genome. Next, the Ensembl tool is used to annotate the aligned reads in the third gene sequencing data to obtain the fourth gene sequencing data. Ensembl is a comprehensive genome annotation platform that provides rich genome annotation information. Through annotation, genes, mutation sites and other functional elements in gene sequencing data can be identified and annotated to enrich the biological significance of the data.

[0041] In a word, through the above steps, the fourth gene sequencing data obtained has high-quality alignment and annotation information. These data are used for subsequent feature extraction steps, and a preset neural network algorithm is adopted to obtain mutation site characteristics and gene expression level characteristics from gene sequencing data. These characteristics play a key role in the breast cancer risk assessment model, and can significantly improve the prediction accuracy of the model and the accuracy of individualized risk assessment. This processing flow ensures the high quality and high reliability of gene sequencing data, and provides a solid foundation for breast cancer risk assessment. Through meticulous quality control, alignment and annotation steps, the characteristic information extracted from gene sequencing data is more accurate and abundant, which helps to build a more accurate and personalized breast cancer risk assessment model. And this is also one of the key points of the present invention.

[0042] In one example, feature extraction is performed on the original data to obtain a feature set, including: extracting mutation site information of each sample in the preprocessed gene sequencing data, counting the type and frequency of each sample mutation site based on the mutation site information, and constructing a mutation feature matrix based on the type and frequency of each sample mutation site; calculating the expression amount of the gene in the preprocessed gene sequencing data, and constructing an expression feature matrix based on the expression amount of the gene; performing feature transformation on the mutation feature matrix and the expression feature matrix, converting them into dense vectors, to obtain fifth gene sequencing data; feeding the vectorized fifth gene sequencing data into a preset Transformer-based neural network model; using a preset convolutional layer in the Transformer-based neural network model to extract initial features of mutation sites and initial features of gene expression levels in the fifth gene sequencing data, and outputting a feature map; using a preset pooling layer in the Transformer-based neural network model to perform dimensionality reduction on the feature map to obtain a first important feature map in the mutation site and a second important feature map of the gene expression level in the fifth gene sequencing data; flattening the first important feature map and the second important feature map, and connecting them to a fully connected layer to obtain mutation site features and gene expression level features of the gene sequencing data.

[0043] In this example, the breast cancer risk assessment method uses a preset neural network algorithm to perform detailed feature extraction steps on the original gene sequencing data to obtain key information on gene mutations and expression levels. The following is an explanation of each step: First, the preset neural network algorithm is used to extract features from the original data. The mutation site information of each sample is extracted from the fourth gene sequencing data, and the type and frequency of the mutation site of each sample are counted based on this information. This step ensures a comprehensive understanding of gene mutations and lays the foundation for subsequent feature construction. Next, a mutation feature matrix is ​​constructed based on the type and frequency of the mutation site of each sample. This matrix records the occurrence of various types of mutations in different samples and is the core data of mutation features. At the same time, the expression of genes in the fourth gene sequencing data is calculated, and an expression feature matrix is ​​constructed based on these expression levels. The expression feature matrix records the expression level of genes in each sample, providing a basis for the analysis of expression features.

[0044] Then, the mutation feature matrix and the expression feature matrix are feature transformed and converted into dense vectors to obtain the fifth gene sequencing data. Feature transformation is to convert the data in matrix form into vector form so that it is suitable for the input requirements of the neural network. The vectorized fifth gene sequencing data is fed into the preset Transformer-based neural network model. In the neural network, the convolution layer is first used to extract the initial features of the mutation site and the initial features of the gene expression level in the fifth gene sequencing data, and the feature map is output. The convolution layer can capture the local patterns and important features in the data by applying the convolution kernel. Next, the pooling layer is used to reduce the dimension of the feature map to obtain the first important feature map in the mutation site and the second important feature map of the gene expression level. The pooling layer reduces the size of the feature map, retains the most important information, and reduces the computational complexity. Then, the first important feature map and the second important feature map are flattened and connected to the fully connected layer. The flattening operation is to convert the two-dimensional feature map into a one-dimensional vector, which is convenient for connection with the fully connected layer. The fully connected layer is a basic component of a neural network, which converts the input features into the final output features through weighting and activation functions. Through this process, the mutation site characteristics and gene expression level characteristics of the fifth gene sequencing data are obtained.

[0045] In summary, this example extracts key information from gene sequencing data through a series of detailed steps and analyzes it using a Transformer-based neural network model. This method can accurately capture the characteristics of gene mutations and expression levels, and improve the accuracy and individuality of breast cancer risk assessment.

[0046] In one example, feature extraction is performed on the original data to obtain a feature set, including: dividing the image in the preprocessed medical imaging data into multiple small blocks, and flattening each small block into a one-dimensional vector to obtain a first vector; using a linear layer in a Transformer neural network model to embed the first vector into a preset dimension to obtain a second vector; using a learned position encoding to generate a code for each position of the second vector to obtain a third vector; connecting the attention head in the Transformer neural network model to the feedforward neural network in the Transformer neural network model, and feeding the third vector into the feedforward neural network in the Transformer neural network model to obtain shape and texture features of the tumor area in the medical imaging data.

[0047] In this example, the breast cancer risk assessment method uses a preset neural network algorithm to extract the shape and texture features of the tumor area from the medical imaging data. The following is a detailed explanation of each step: First, the image in the medical imaging data is divided into multiple small blocks. Each small block represents a small part of the image. This division method can be used to analyze and process the image data in a more detailed manner. Next, each small block is flattened into a one-dimensional vector to obtain a first vector. Flattening into a one-dimensional vector is to convert the image data into a form suitable for neural network processing. The first vector is embedded into a preset dimension using a linear layer in the Transformer neural network model to obtain a second vector. This step maps the data of each small block to a high-dimensional space through a linear transformation, making it more expressive. Next, a learned position encoding is used to generate an encoding for each position of the second vector to obtain a third vector. Position encoding is a key step in the Transformer model. It assigns a unique encoding to each position in the input data, enabling the model to recognize and process position information in sequence information. The attention head in the Transformer neural network model is connected to the feedforward neural network in the Transformer neural network model, and the third vector is fed into the feedforward neural network in the Transformer neural network model. The attention mechanism is the core component of the Transformer model. It can assign attention weights according to the correlation in the input data, thereby capturing important features in the data. The feedforward neural network further processes these features and extracts higher-level information. Through this series of processing steps, the shape and texture features of the tumor area in the medical imaging data are finally obtained. These features play a key role in the risk assessment of breast cancer, because the shape and texture features can reflect the specific morphology and internal structure of the tumor, which helps to improve the prediction accuracy and individualized assessment ability of the model.

[0048] In summary, this example shows how to use a Transformer-based neural network model to extract key shape and texture features from medical imaging data. Through steps such as image segmentation, embedding dimensions, position encoding, and attention mechanisms, the features extracted from the imaging data are ensured to be highly expressive and accurate, thereby enhancing the effect of breast cancer risk assessment.

[0049] In one example, the feature set is used to construct a breast cancer assessment model based on a Transformer neural network model, including: inputting the feature set into a normalization layer in the Transformer neural network model, normalizing the shape and texture features of the tumor area in each layer of the input medical imaging data and the mutation site features and gene expression level features of the gene sequencing data to obtain a first fusion feature; adding the first fusion feature to the vectorized fifth gene sequencing data and the first vector to obtain a first fusion feature; inputting the first fusion feature into the normalization layer in the Transformer neural network model to obtain a second fusion feature; and adding the first fusion feature to the vectorized fifth gene sequencing data and the first vector to obtain a first fusion feature. The second fusion feature is added to the vectorized fifth gene sequencing data and the first vector to form a third fusion feature; the third fusion feature is input into the normalization layer in the Transformer neural network model to obtain a fourth fusion feature; the fourth fusion feature is added to the vectorized fifth gene sequencing data and the first vector to form a fifth fusion feature; the fifth fusion feature is input into the normalization layer in the Transformer neural network model to obtain a sixth fusion feature; the sixth fusion feature is input into the Softmax layer in the Transformer neural network model to obtain a breast cancer morbidity assessment model based on the Transformer neural network model.

[0050] In this example, the breast cancer risk assessment method is to conduct a comprehensive assessment by constructing features extracted from medical imaging data and gene sequencing data based on a Transformer neural network model. The following is a detailed explanation of each step: First, the shape and texture features of the tumor area in the medical imaging data, as well as the mutation site features and gene expression level features of the fifth gene sequencing data are input into the normalization layer based on the Transformer neural network model. The function of the normalization layer is to standardize the input features so that the distribution of features in each layer is more uniform, thereby improving the stability and convergence speed of model training. After normalization, the first fusion feature is obtained. Then, the first fusion feature is added to the vectorized fifth gene sequencing data and the first vector to form an updated first fusion feature. The purpose of this step is to integrate data features from different sources to ensure the comprehensiveness and diversity of information. The updated first fusion feature is input into the normalization layer again to obtain the second fusion feature. Then, the second fusion feature is added to the vectorized fifth gene sequencing data and the first vector to form the third fusion feature. In this process, through multiple addition and normalization operations, the association between different features is gradually fused and strengthened. The third fused feature is input into the normalization layer to obtain the fourth fused feature.

[0051] Then add the fourth fusion feature to the vectorized fifth gene sequencing data and the first vector to form the fifth fusion feature. This series of fusion steps, through repeated feature fusion and normalization processing, fully utilizes the characteristics of various types of data and improves the model's ability to express complex features. The fifth fusion feature is input into the normalization layer to obtain the sixth fusion feature. Finally, the sixth fusion feature is input into the Softmax layer based on the Transformer neural network model. The Softmax layer is an activation function layer used in classification tasks. It converts the input features into probability distributions to output probability values ​​for each category. In this example, the Softmax layer will output the probability of breast cancer, and finally form a breast cancer assessment model based on the Transformer neural network model.

[0052] In summary, through these steps, this method uses a Transformer-based neural network model, combined with medical imaging data and gene sequencing data, to comprehensively extract and fuse various key features, thereby building an efficient breast cancer risk assessment model. This model can accurately capture the shape and texture characteristics of the tumor area as well as the characteristics of gene mutations and expression levels, improving the accuracy and individualization of breast cancer risk assessment.

[0053] In one example, a training process is performed on an initial breast cancer prevalence assessment model based on a Transformer neural network model to obtain a trained breast cancer prevalence assessment model based on a Transformer neural network model, including: using a cross entropy loss function to calculate the loss of the initial breast cancer prevalence assessment model based on the Transformer neural network model; using an RMSprop optimizer to adjust the parameters of the initial breast cancer prevalence assessment model based on the Transformer neural network model; using an ultra-bandwidth optimization method to optimize the parameter combination of the initial breast cancer prevalence assessment model based on the Transformer neural network model, repeatedly screening and allocating resources, and screening out the best hyperparameter combination; dividing the original data into multiple small batches, performing forward propagation and back propagation with a small batch of data each time, repeating the above steps for multiple training cycles, and training the initial breast cancer prevalence assessment model based on the Transformer neural network model multiple times to obtain the trained breast cancer prevalence assessment model based on the Transformer neural network model.

[0054] In this example, the breast cancer risk assessment method obtains an efficient breast cancer risk assessment model by training the initial model based on the Transformer neural network. The following is a detailed explanation of each step: First, the loss of the initial breast cancer risk assessment model based on the Transformer neural network model is calculated using the cross entropy loss function. The cross entropy loss function is a loss function used for classification problems. It can measure the difference between the model's prediction results and the actual labels. The smaller the loss value, the closer the model's prediction results are to the actual situation. Next, the RMSprop optimizer is used to adjust the parameters of the Transformer neural network model. The RMSprop optimizer is a gradient descent optimization method that accelerates the model's training process by adjusting the learning rate. It is particularly suitable for dealing with non-stationary objective functions. In each iteration, the RMSprop optimizer updates the model's parameters according to the gradient of the loss function, thereby gradually reducing the loss value and improving the model's prediction performance. After that, the ultra-bandwidth optimization method is used to optimize the parameter combination based on the Transformer neural network model. Ultra-bandwidth optimization is a method for hyperparameter optimization that repeatedly screens and allocates resources to find the best hyperparameter combination. In model training, hyperparameters (such as learning rate, batch size, etc.) have an important impact on the performance of the model. Optimizing these hyperparameters can significantly improve the effect of the model. Through the ultra-bandwidth optimization method, different hyperparameter combinations can be efficiently explored, and the optimal configuration can be selected to further improve the performance of the model.

[0055] Then, the original data is divided into multiple small batches, and each small batch of data is used for forward propagation and back propagation. Forward propagation is to pass the input data through the neural network model to calculate the prediction result; back propagation is to adjust the parameters of the model according to the gradient of the loss function. Through this batch-by-batch training method, memory resources can be effectively utilized while improving training efficiency. Repeat the above steps for multiple training cycles. In each training cycle, new batches of data are continuously used for forward propagation and back propagation to gradually optimize the parameters of the model and reduce the loss value. By repeatedly training the initial model of breast cancer assessment based on the Transformer neural network model, a trained breast cancer assessment model based on the Transformer neural network model is finally obtained. In this process, the model continuously learns and optimizes, extracts the most useful features from the data, and improves the prediction accuracy and individualized assessment ability of breast cancer risk.

[0056] In summary, this example shows how to train the Transformer-based breast cancer risk assessment initial model into an efficient and accurate assessment model through a series of optimization and training steps. The cross entropy loss function, RMSprop optimizer, and ultra-bandwidth optimization method are used to ensure that the model can continuously improve over multiple training cycles, thereby providing reliable breast cancer risk assessment results.

[0057] In one example, after the initial breast cancer disease assessment model based on the Transformer neural network model is trained to obtain a trained breast cancer disease assessment model based on the Transformer neural network model, the disease risk assessment method further includes: saving the trained breast cancer disease assessment model based on the Transformer neural network model and model parameters to a file, and the user can load the trained breast cancer disease assessment model based on the Transformer neural network model and model parameters. After loading, the user inputs test data to predict whether the patient has breast cancer.

[0058] In this example, the breast cancer risk assessment method also includes the steps of saving and using the trained model for prediction after completing the training of the Transformer neural network model. The following is a detailed explanation of each step: First, after the initial model of breast cancer assessment based on the Transformer neural network model is trained, a trained model is obtained. This means that the model has fully learned and optimized the features extracted from the data through multiple training cycles and can accurately predict the risk of breast cancer. Next, the trained Transformer neural network model and model parameters are saved to a file. The purpose of saving the model and parameters is to store the trained model so that it can be easily loaded and used in the future. Usually, this file will include the model's structure, weight parameters, and other necessary information to ensure that the model can reproduce its prediction performance in different environments. Users can load the trained Transformer neural network model and model parameters. The process of loading a model usually involves reading the saved file and restoring the model's structure and parameters to memory so that it can be used for inference and prediction. This step ensures that users can directly use the optimized model for evaluation without retraining the model. After loading the model, the user enters the test data. Test data is usually medical imaging data and gene sequencing data that are similar to training data, but these data have not participated in the model training process. By inputting test data, users hope to verify the model's predictive ability on new data.

[0059] In short, the model will perform forward propagation and calculation based on the input test data and output the prediction results. Specifically, the model will analyze the shape and texture features of the tumor area in the test data, as well as the gene mutation and expression level features, and use these features for comprehensive evaluation to predict whether the patient has breast cancer.

[0060] Reference Figure 2 The embodiment of the present invention provides a breast cancer risk assessment system, comprising:

[0061] The acquisition unit 1 is used to acquire original data based on breast cancer, wherein the original data based on breast cancer at least includes: gene sequencing data and medical imaging data.

[0062] The feature extraction unit 2 is used to extract features from the original data to obtain a feature set, wherein the feature set at least includes: mutation site features and gene expression level features in gene sequencing data, and shape and texture features of the tumor area in medical impact data.

[0063] The construction unit 3 is used to use the feature set to construct a breast cancer morbidity assessment model based on the Transformer neural network model.

[0064] The training unit 4 is used to train the initial breast cancer prevalence assessment model based on the Transformer neural network model to obtain a trained breast cancer prevalence assessment model based on the Transformer neural network model.

[0065] In this embodiment, for the specific implementation of each unit in the above system embodiment, please refer to the above method embodiment, which will not be repeated here.

[0066] Reference Figure 3 In an embodiment of the present invention, a computer device is also provided. The computer device may be a server, and its internal structure may be as follows: Figure 3 As shown. The computer device includes a processor, a memory, a display screen, an input system, a network interface and a database connected through a system bus. Among them, the processor designed by the computer is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store the corresponding data in this embodiment. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, the above method is implemented.

[0067] Those skilled in the art will understand that Figure 3 The structure shown in the figure is merely a block diagram of a portion of the structure related to the solution of the present invention, and does not constitute a limitation on the computer device to which the solution of the present invention is applied.

[0068] An embodiment of the present invention further provides a computer-readable storage medium on which a computer program is stored, and when the computer program is executed by a processor, the above method is implemented. It can be understood that the computer-readable storage medium in this embodiment can be a volatile readable storage medium or a non-volatile readable storage medium.

[0069] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media provided by the present invention and used in the embodiments may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double-speed data rate SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM.

[0070] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, system, article or method including a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, system, article or method. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the presence of other identical elements in the process, system, article or method including the element.

[0071] The above description is only a preferred embodiment of the present invention, and does not limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made by using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A method for assessing the risk of breast cancer, characterized in that: The following steps are involved: Acquiring original data based on breast cancer, wherein the original data based on breast cancer at least includes: gene sequencing data and medical imaging data; The raw data is subjected to feature extraction to obtain a feature set, wherein the feature set includes at least: mutation site features and gene expression level features in the gene sequencing data, and shape and texture features of the tumor area in the medical imaging data; specifically including: extracting mutation site information of each sample in the preprocessed gene sequencing data, counting the type and frequency of each sample mutation site based on the mutation site information, and constructing a mutation feature matrix based on the type and frequency of each sample mutation site; calculating the expression amount of the gene in the preprocessed gene sequencing data, and constructing an expression feature matrix based on the expression amount of the gene; performing feature transformation on the mutation feature matrix and the expression feature matrix, converting them into dense vectors, and obtaining a fifth gene sequencing data. The method comprises the following steps: feeding the vectorized fifth gene sequencing data into a preset Transformer-based neural network model; using a preset convolutional layer in the Transformer-based neural network model to extract initial features of mutation sites and initial features of gene expression levels in the fifth gene sequencing data, and outputting a feature map; using a preset pooling layer in the Transformer-based neural network model to perform a dimensionality reduction operation on the feature map to obtain a first important feature map of mutation sites in the fifth gene sequencing data and a second important feature map of gene expression levels; flattening the first important feature map and the second important feature map, and connecting them to a fully connected layer to obtain mutation site features and gene expression level features of the gene sequencing data; The feature set is used to construct a breast cancer disease assessment model based on the Transformer neural network model; specifically, the feature set is divided into multiple small blocks in the preprocessed medical imaging data, and each small block is flattened into a one-dimensional vector to obtain a first vector; the feature set is input into the normalization layer in the Transformer neural network model, and the shape and texture features of the tumor area in each layer of the input medical imaging data and the mutation site features and gene expression level features of the gene sequencing data are normalized to obtain a first fusion feature; the first fusion feature is added to the vectorized fifth gene sequencing data and the first vector to obtain a first fusion feature; the first fusion feature is input into the normalization layer in the Transformer neural network model. The second fusion feature is obtained by adding the second fusion feature to the vectorized fifth gene sequencing data and the first vector to form a third fusion feature; the third fusion feature is input into the normalization layer in the Transformer neural network model to obtain a fourth fusion feature; the fourth fusion feature is added to the vectorized fifth gene sequencing data and the first vector to form a fifth fusion feature; the fifth fusion feature is input into the normalization layer in the Transformer neural network model to obtain a sixth fusion feature; the sixth fusion feature is input into the Softmax layer in the Transformer neural network model to obtain a breast cancer morbidity assessment model based on the Transformer neural network model; The initial model of breast cancer prevalence assessment based on the Transformer neural network model is trained to obtain a trained breast cancer prevalence assessment model based on the Transformer neural network model.

2. The disease risk assessment method according to claim 1, characterized in that: After obtaining the original data based on breast cancer, the method for assessing the risk of breast cancer further includes: Adjusting image pixels in the medical image data to preset pixel values ​​to obtain first medical image data; Mapping the pixel values ​​of the first medical image data into a preset interval using a Z-SCORE method to obtain second medical image data; Using an anisotropic diffusion filter to remove image noise in the second medical image data to obtain third medical image data; adjusting the contrast of the third medical image data using contrast limited adaptive histogram equalization to obtain fourth medical image data; Segmenting the fourth medical image using a neural network U-NET to extract tumor regions and organ boundaries in the fourth medical image data; The tumor area and the organ boundary in the fourth medical image data are marked to obtain preprocessed medical image data, wherein the preprocessed medical image data is used to execute the step of "extracting features from the original data to obtain a feature set".

3. The disease risk assessment method according to claim 1, characterized in that: After obtaining the original data based on breast cancer, the disease risk assessment method further includes: Filtering reads containing adapter contamination in the gene sequencing data, and removing duplicate reads in the gene sequencing data to obtain second gene sequencing data; Aligning the read segments in the second gene sequencing data to the reference genome to obtain third gene sequencing data; Annotate the aligned reads in the third gene sequencing data to obtain preprocessed gene sequencing data, wherein the preprocessed gene sequencing data is used to perform the step of "extracting features from the original data to obtain a feature set".

4. The disease risk assessment method according to claim 1, characterized in that: Performing feature extraction on the original data to obtain a feature set, including: Dividing the image in the preprocessed medical imaging data into a plurality of small blocks, and flattening each small block into a one-dimensional vector to obtain a first vector; Embed the first vector into a preset dimension using a linear layer in a Transformer-based neural network model to obtain a second vector; Generate a code for each position of the second vector using a learned position code to obtain a third vector; The attention head in the Transformer neural network model is connected to the feedforward neural network in the Transformer neural network model, and the third vector is fed into the feedforward neural network in the Transformer neural network model to obtain shape and texture features of the tumor area in the medical imaging data.

5. The disease risk assessment method according to claim 1, characterized in that: The initial model of breast cancer disease assessment based on the Transformer neural network model is trained to obtain a trained breast cancer disease assessment model based on the Transformer neural network model, including: Using a cross entropy loss function to calculate the loss of the initial model for breast cancer assessment based on the Transformer neural network model; Using the RMSprop optimizer to adjust the parameters of the initial model for breast cancer assessment based on the Transformer neural network model; Using an ultra-bandwidth optimization method to optimize the parameter combination of the initial model for breast cancer assessment based on the Transformer neural network model, repeatedly screening and allocating resources, and screening out the best ultra-parameter combination; The original data is divided into multiple small batches, and forward propagation and back propagation are performed with a small batch of data each time. The above steps are repeated for multiple training cycles, and the initial breast cancer prevalence assessment model based on the Transformer neural network model is trained multiple times to obtain the trained breast cancer prevalence assessment model based on the Transformer neural network model.

6. A breast cancer risk assessment system, characterized in that: include: An acquisition unit, configured to acquire original data based on breast cancer, wherein the original data based on breast cancer at least includes: gene sequencing data and medical imaging data; The feature extraction unit is used to extract features from the original data to obtain a feature set, wherein the feature set at least includes: mutation site features and gene expression level features in gene sequencing data, and shape and texture features of tumor areas in medical imaging data; specifically includes: extracting mutation site information of each sample in the preprocessed gene sequencing data, counting the type and frequency of each sample mutation site based on the mutation site information, and constructing a mutation feature matrix based on the type and frequency of each sample mutation site; calculating the expression amount of the gene in the preprocessed gene sequencing data, and constructing an expression feature matrix based on the expression amount of the gene; performing feature transformation on the mutation feature matrix and the expression feature matrix to convert them into dense vectors to obtain a fifth Gene sequencing data; feeding the vectorized fifth gene sequencing data into a preset Transformer-based neural network model; using a preset convolutional layer in the Transformer-based neural network model to extract initial features of mutation sites and initial features of gene expression levels in the fifth gene sequencing data, and outputting a feature map; using a preset pooling layer in the Transformer-based neural network model to perform a dimensionality reduction operation on the feature map to obtain a first important feature map of mutation sites in the fifth gene sequencing data and a second important feature map of gene expression levels; flattening the first important feature map and the second important feature map, and connecting them to a fully connected layer to obtain mutation site features and gene expression level features of the gene sequencing data; A construction unit is used to use the feature set to construct a breast cancer disease assessment model based on a Transformer neural network model; specifically comprising: dividing the image in the preprocessed medical imaging data into multiple small blocks, and flattening each small block into a one-dimensional vector to obtain a first vector; inputting the feature set into the normalization layer in the Transformer neural network model, normalizing the shape and texture features of the tumor area in each layer of the input medical imaging data and the mutation site features and gene expression level features of the gene sequencing data to obtain a first fusion feature; adding the first fusion feature to the vectorized fifth gene sequencing data and the first vector to obtain a first fusion feature; inputting the first fusion feature into the normalization layer in the Transformer neural network model; The second fusion feature is input into the normalization layer in the Transformer neural network model to obtain a second fusion feature; the second fusion feature is added to the vectorized fifth gene sequencing data and the first vector to form a third fusion feature; the third fusion feature is input into the normalization layer in the Transformer neural network model to obtain a fourth fusion feature; the fourth fusion feature is added to the vectorized fifth gene sequencing data and the first vector to form a fifth fusion feature; the fifth fusion feature is input into the normalization layer in the Transformer neural network model to obtain a sixth fusion feature; the sixth fusion feature is input into the Softmax layer in the Transformer neural network model to obtain a breast cancer disease assessment model based on the Transformer neural network model; The training unit is used to train the initial breast cancer prevalence assessment model based on the Transformer neural network model to obtain a trained breast cancer prevalence assessment model based on the Transformer neural network model.

7. A computer device comprising a memory and a processor, wherein a computer program is stored in the memory, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 5 are implemented.

8. 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 according to any one of claims 1 to 5 are implemented.

Images