Intelligent breast cancer detection system based on deep learning

Abstract

The application discloses a kind of breast cancer intelligent detection systems based on deep learning, it is related to deep learning and intelligent detection field;Including image acquisition and pretreatment module, multi-modal data set construction module, multi-task classification module, breast cancer intelligent detection module, evaluation and feedback module, user efficient interaction module, multi-task classification module, by integrating attention mechanism and multi-task learning method, system not only improves the accuracy of breast density classification, also provides more abundant prediction information for benign and malignant detection of breast cancer, help to improve the application effect of breast cancer intelligent detection system in clinic.The application greatly improves the performance of breast cancer detection system, especially in the accuracy and efficiency of lesion positioning, feature extraction and classification, greatly improves the practicability of the system in clinical auxiliary diagnosis.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of deep learning and intelligent detection, and particularly relates to a breast cancer intelligent detection system based on deep learning. BACKGROUND

[0002] Attention network technology is a deep learning method designed to address information redundancy in breast image processing and feature extraction, by focusing on lesion areas, the system can automatically identify important feature regions in the image, thereby improving the accuracy of detection, attention network technology enables the breast cancer intelligent detection system to more effectively distinguish key lesions, thereby improving the efficiency of early diagnosis.

[0003] Existing breast cancer intelligent detection systems cannot fully identify and distinguish subtle breast lesions in complex images, resulting in the risk of false negatives and false positives, in addition, traditional image processing methods often rely on manual feature extraction, lack sufficient intelligence, limiting the effectiveness and reliability of the system in practical applications. SUMMARY

[0004] The purpose of the present application is to provide a breast cancer intelligent detection system based on deep learning to solve the problem of the existing breast cancer intelligent detection system based on deep learning in the background art, which cannot fully identify and distinguish subtle breast lesions in complex images, resulting in the risk of false negatives and false positives, and traditional image processing methods often rely on manual feature extraction, lack sufficient intelligence, limiting the effectiveness and reliability of the system in practical applications.

[0005] To achieve the above purpose, the present application provides the following technical scheme:

[0006] A breast cancer intelligent detection system based on deep learning, comprising an image acquisition and preprocessing module, a multi-modal data set construction module, a multi-task classification module, a breast cancer intelligent detection module, an evaluation and feedback module, and a user efficient interaction module;

[0007] The image acquisition and preprocessing module acquires breast molybdenum target images from mammography and performs preprocessing operations on the images;

[0008] The multi-modal data set construction module integrates different types of breast image data, including breast molybdenum target images and patient clinical information, forming a multi-modal data set for subsequent module use;

[0009] The multi-task classification module includes a breast data labeling unit and an attention mechanism unit, the breast data labeling unit is used for labeling lesions in breast molybdenum target images; the attention mechanism unit is based on an attention network breast data classification algorithm, focusing on key lesion areas in breast images by applying channel and spatial attention mechanisms;

[0010] The breast cancer intelligent detection module comprises a target detection unit and an iteration and optimization unit. The target detection unit is based on an intelligent target detection algorithm of deep learning, and is used for automatically positioning and intelligently identifying a lesion in a breast molybdenum target image. By using a two-stage target detection network in deep learning, the model can identify the position of the lesion and distinguish the type of the lesion. The iteration and optimization unit iteratively trains breast data samples and adjusts model parameters.

[0011] The evaluation and feedback module evaluates the output result of the breast cancer intelligent detection module, and performs a comparative experiment according to a test result and a public data set.

[0012] The user efficient interaction module displays the detection result and analysis information to the user through a user interface and generates a report.

[0013] Preferably, the breast data classification algorithm based on the attention network is specifically as follows:

[0014] For the image data processed by the breast data labeling unit, a convolutional neural network based on an attention mechanism is designed, and the structure is specifically represented as:

[0015]

[0016] wherein, is a convolutional neural network model, and represent the input of the left and right breast images, respectively; and represent the output feature maps of the left and right breast images processed by the residual network, respectively, is an adaptive spatial attention module, is an adaptive channel attention module, represents a 1x1 convolutional layer, is a weight set of the convolutional neural network . represents the output feature map data of the convolutional layer processed by the spatial attention module and the channel attention module in sequence, is a fully connected layer, which is used to map the learned features to the final classification result.

[0017] Preferably, the adaptive spatial attention module is based on an adaptive spatial attention strategy, and the specific formula is represented as:

[0018]

[0019] wherein, Let be the th element of the final output feature map after processing with the adaptive spatial attention strategy. Each pixel is a weighted aggregation of the input feature map. Represented as the index of the number of pixels in the final output feature map. It is represented as a learnable scale parameter. This represents the total number of pixels in the feature map after processing with the adaptive spatial attention strategy. This represents the index of the number of pixels in the input feature map. Represented as spatial attention weights, The output feature map of the left breast image, which is the input feature map processed by the residual network, is represented by the first feature map. 1 pixel, The term represents the output feature map of the convolutional layer after being processed by the spatial attention module and the channel attention module, respectively. 1 pixel;

[0020] The adaptive channel attention module Based on the adaptive channel attention strategy, the specific formula is expressed as follows:

[0021]

[0022] in, The term is represented by the th element of the final output feature map after processing with the adaptive channel attention strategy. 1 pixel, It is represented as a learnable scale parameter. This represents the total number of pixels in the feature map after processing with the adaptive channel attention strategy. This represents the index of the number of pixels in the input feature map. Represented as channel attention weights, The feature map after 1x1 convolution is represented as the first... 1 pixel;

[0023] The fully connected layer acts as a classifier, mapping the feature maps processed by the spatial and channel attention modules onto the classification results of breast density. The specific formula is as follows:

[0024]

[0025] in, This is represented as the classification result, corresponding to the predicted probability of different breast density categories. Represented as the weight matrix of the fully connected layer. This is represented as the result after processing by an adaptive spatial attention strategy and an adaptive channel attention strategy, followed by a 1*1 convolutional layer. The integrated feature map Represented as the bias term of the fully connected layer. This is represented as an activation function.

[0026] Preferably, the attention network-based breast data classification algorithm also employs a multi-task learning strategy to enable the convolutional neural network model to... While performing breast density classification, it can differentiate between benign and malignant breast cancer. The specific formula is as follows:

[0027]

[0028] in, This represents an optimized model that performs multi-task learning. This is represented by a process function that represents multi-task learning. Represented as a convolutional neural network model The convolutional neural network model after iterative optimization using adaptive spatial attention and adaptive channel attention strategies. This is represented as an auxiliary task that includes the localization of the mass and the classification of the benign or malignant nature of the tumor.

[0029] Preferably, the deep learning-based intelligent target detection algorithm is as follows:

[0030] First, an image segmentation convolutional neural network is constructed to segment the mammogram image after it has been processed by the attention mechanism module, so as to distinguish between cancerous areas and normal tissue.

[0031] After image segmentation, in order to extract effective breast cancer lesion features from the image, a locust optimization search strategy is constructed. The locust optimization algorithm is used to simulate the swarm behavior of locusts to optimize the search process for the solution to the problem, thereby optimizing the feature extraction process.

[0032] Then feature selection is performed to reduce the dimensionality of breast cancer lesion features and improve classification accuracy;

[0033] Finally, the processed breast cancer lesion features are classified and the classification results are output, specifically as follows:

[0034]

[0035] in, This represents the category label for predicted benign or malignant breast cancer lesions. Represents probability Maximize the choice function, This represents the input breast cancer lesion characteristics. Category The conditional probability, This index represents the categories of benign and malignant breast cancer lesions.

[0036] Preferably, the image segmentation convolutional neural network includes an activation function and a linear transformation, specifically expressed as follows:

[0037]

[0038] in, Represented as the first The linear output of the layer, This represents the index of the layer in the image segmentation convolutional neural network. Represented as the first The weight matrix of the layer, Represented as the first The activation output of a layer represents the output of the previous layer after processing by the activation function. Represented as the first Layer bias vector, Represented as the first Layer activation output, This represents the activation function.

[0039] Preferably, the feature selection employs a support vector machine.

[0040] Preferably, the preprocessing operation includes denoising and normalizing the image.

[0041] Preferably, the annotation content of the breast data annotation unit includes the location, type, and benign or malignant characteristics of the lesion, and the type includes lumps, nodules, and calcifications.

[0042] Compared with the prior art, the beneficial effects of the present invention are:

[0043] 1. The attention mechanism unit proposes a breast data classification algorithm based on attention networks. First, by designing a convolutional neural network model BASCNet based on attention mechanism, the algorithm can utilize information from bilateral breast images and adaptively focus on features in spatial and channel dimensions, thereby improving the accuracy of breast density classification. Specifically, the BASCNet convolutional neural network model captures global information in the spatial dimension through an adaptive spatial attention strategy and adaptively focuses on discriminative features; the adaptive channel attention module strategy highlights information-rich features in the channel dimension. Through these two strategies, the system can perform in-depth analysis of key regions in the image, thereby effectively improving the accuracy and robustness of lesion detection. In addition, the attention mechanism of this algorithm, through adaptive spatial and channel attention strategies, enables the model to focus on regions and channels that are crucial to the classification task, significantly improving classification performance. By constructing a multi-task learning strategy, the BASCNet convolutional neural network model can not only perform the main classification task of breast density, but also simultaneously distinguish between benign and malignant tumor features. This multi-task learning method, through joint training, improves the detection and classification capabilities of the entire network, making it more powerful and comprehensive in complex clinical applications. By integrating these attention mechanisms and multi-task learning methods, the system not only improves the accuracy of breast density classification, but also provides richer predictive information for the detection of benign and malignant breast cancer, which helps to improve the clinical application effect of intelligent breast cancer detection systems.

[0044] 2. The target detection unit proposes a deep learning-based intelligent target detection algorithm. First, a convolutional neural network is used to segment mammograms, automatically identifying and distinguishing cancerous areas from normal tissue. This algorithm adaptively learns image features, capturing the morphological characteristics of breast cancer lesions, especially the pattern recognition of tumor areas, which is crucial for subsequent lesion detection. The segmented image information provides accurate input for subsequent detection. Second, to further optimize the feature extraction process, a locust-based optimization search strategy is constructed. This strategy enhances the efficiency and accuracy of feature extraction by simulating the social behavior of locust swarms. By introducing the influence of multiple factors such as gravity, wind, and social interaction, the optimization algorithm can effectively find the optimal feature set in the search space, thereby reducing redundant information and improving the model's detection performance. Next, to improve accuracy, Support Vector Machine (SVM) is used to select extracted features to reduce dimensionality and improve classification accuracy. SVM performs well in high-dimensional spaces and can find the optimal decision boundary for different types of breast lesions, ensuring effective differentiation between benign and malignant tumors in complex feature spaces. Finally, the data after feature extraction and selection will be used for the benign and malignant classification of breast cancer lesions. The system calculates the probability distribution of features and outputs a clear classification result. This process directly determines the effectiveness of the system in actual clinical applications. Overall, the deep learning-based intelligent target detection algorithm greatly improves the performance of the breast cancer detection system, especially in the accuracy and efficiency of lesion localization, feature extraction and classification, and greatly enhances the practicality of the system in clinical auxiliary diagnosis. Attached Figure Description

[0045] The invention will be further illustrated with reference to the accompanying drawings, but the embodiments in the drawings do not constitute any limitation on the invention. For those skilled in the art, other drawings can be obtained based on the following drawings without any creative effort.

[0046] Figure 1 This is a schematic diagram of the structure of the present invention. Detailed Implementation

[0047] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0048] Please see Figure 1This invention provides a deep learning-based intelligent breast cancer detection system, including an image acquisition and preprocessing module, a multimodal dataset construction module, a multi-task classification module, an intelligent breast cancer detection module, an evaluation and feedback module, and an efficient user interaction module.

[0049] The image acquisition and preprocessing module is used to acquire mammograms from mammography and perform preprocessing operations on the images, including denoising and normalization, to improve image quality and ensure that lesion areas are clearly visible, thereby improving the accuracy of the system in subsequent detection.

[0050] The multimodal dataset construction module integrates different types of breast imaging data, including mammograms and patient clinical information, to form a multimodal dataset for use by subsequent modules. It collects different types of mammograms, including positive and negative samples, and generates a standardized dataset format to ensure that the data acquired during system training can be effectively used.

[0051] The multi-task classification module includes a breast data annotation unit and an attention mechanism unit. The breast data annotation unit accurately labels lesions in mammograms, including their location, type, and benign / malignant characteristics. Types include masses, nodules, and calcifications. Accurate annotation ensures that subsequent modules can identify and classify different types of lesions, thereby improving the model's detection and classification accuracy. The attention mechanism unit proposes a breast data classification algorithm based on attention networks. During breast image processing and feature extraction, it focuses on key lesion regions in the breast image by applying channel and spatial attention mechanisms, ensuring the system can effectively identify lesions in complex structures and improving the detection capability of dense breast lesions.

[0052] The intelligent breast cancer detection module comprises a target detection unit and an iteration and optimization unit. The target detection unit proposes a deep learning-based intelligent target detection algorithm for automatically locating and intelligently identifying lesions in mammograms. By using a two-stage target detection network in deep learning, the model can identify the location of lumps, nodules, and calcified lesions and distinguish between lesion types. The iteration and optimization unit is used to iteratively train breast data samples and adjust the weight parameters of the image segmentation convolutional neural network to continuously optimize detection and classification performance, ensuring that the accuracy, sensitivity, and specificity of the model are gradually improved during training to adapt to different types of breast images.

[0053] The evaluation and feedback module is used to evaluate the output results of the breast cancer intelligent detection module, analyze the system's key indicators such as detection accuracy, sensitivity, and specificity, and conduct comparative experiments with public datasets based on the test results to ensure the system's reliability and clinical applicability in practical applications. It also provides feedback to optimize the system's performance.

[0054] The efficient user interaction module provides an intuitive and easy-to-use user interface, displaying test results and analysis information and generating detailed classification reports. This ensures that users can quickly understand and use the system's test results, thereby improving the efficiency of clinical decision-making.

[0055] Furthermore, the attention-based breast data classification algorithm is as follows: For the image data processed by the breast data annotation unit, a convolutional neural network based on an attention mechanism is designed. By utilizing bilateral breast image information, it adaptively focuses on features in the spatial and channel dimensions, thereby improving the accuracy of breast density classification. The specific convolutional neural network structure is represented as follows:

[0056]

[0057] in, Represented as a convolutional neural network model, and These represent the input images of the left and right breasts, respectively. and These are the output feature maps of the left breast image after processing by the residual network and then processed by the adaptive spatial attention module, and the right breast image after processing by the adaptive channel attention module, respectively. Represented as an adaptive spatial attention module, it is used to focus on important features in the spatial dimension; Represented as an adaptive channel attention module, it is used to highlight information-rich features along the channel dimension; It is represented as a 1x1 convolutional layer, used to further integrate features after the adaptive spatial attention module and the adaptive channel attention module; Represented as a convolutional neural network The set of weights includes the parameters of all convolutional layers, 1x1 convolutional layers, and fully connected layers; This represents the output feature map data of the convolutional layer after being processed by the spatial attention module and the channel attention module in sequence. Represented as a fully connected layer, it is used to map the learned features to the final classification result.

[0058] The convolutional neural network The adaptive spatial attention module in By constructing an adaptive spatial attention strategy, global spatial information is captured while adaptively focusing on discriminative information. The specific formula is expressed as follows:

[0059]

[0060] in, Let be the th element of the final output feature map after processing with the adaptive spatial attention strategy. Each pixel is a weighted aggregation of the input feature map. Represented as the index of the number of pixels in the final output feature map. This is represented as a learnable scale parameter used to adjust the weights of spatial attention features. This represents the total number of pixels in the feature map after processing with the adaptive spatial attention strategy. This represents the index of the number of pixels in the input feature map. Represented as spatial attention weights, these weights reflect the interdependencies between different spatial locations. The output feature map of the left breast image, which is the input feature map processed by the residual network, is represented by the first feature map. 1 pixel, The term represents the output feature map of the convolutional layer after being processed by the spatial attention module and the channel attention module, respectively. Each pixel. The constructed adaptive spatial attention module can capture global spatial information and adaptively focus on discriminative features in the image, enabling the attention mechanism network to focus on regions important to the classification task, thereby improving classification performance.

[0061] The adaptive channel attention module By constructing an adaptive channel attention strategy, leveraging the interdependencies between channels, information-rich channels are highlighted. The specific formula is as follows:

[0062]

[0063] in, The term is represented as the th element of the final output feature map after processing by the adaptive channel attention strategy. 1 pixel,

[0064] This is represented as a learnable scaling parameter used to adjust the weights of the channel attention features. This represents the total number of pixels in the feature map after processing with the adaptive channel attention strategy. This represents the index of the number of pixels in the input feature map. Represented as channel attention weights, these weights reflect the interdependencies between different channel positions. The feature map after 1x1 convolution is represented as the first... Each pixel is then used as a classifier by constructing a fully connected layer. The feature map processed by the spatial and channel attention modules is mapped onto the classification result of breast density. The specific formula is as follows:

[0065]

[0066] in, This is represented as the classification result, corresponding to the predicted probability of different breast density categories. Represented as the weight matrix of a fully connected layer, it is used to weight the feature map. Mapped to the class probability space, This is represented as the result after processing by an adaptive spatial attention strategy and an adaptive channel attention strategy, followed by a 1*1 convolutional layer. The integrated feature map is the final output of the convolutional neural network model. This represents the bias term of the fully connected layer, used to adjust the output. This is represented as an activation function.

[0067] Finally, a multi-task learning strategy is constructed to enable the network to distinguish between benign and malignant breast cancer while performing breast density classification. The specific formula is as follows:

[0068]

[0069] in, This represents an optimized model that performs multi-task learning, capable of performing the main breast density classification task and distinguishing between benign and malignant tumors. This is represented by a process function for multi-task learning, which improves the overall performance of the model by jointly training tasks. Represented as a convolutional neural network model The convolutional neural network model after iterative optimization using adaptive spatial attention and adaptive channel attention strategies. This can be represented as an auxiliary task that includes the localization of the mass and the classification of the benign or malignant nature of the tumor. The convolutional neural network model optimized through multi-task learning methods can not only improve the performance in breast density classification, but also help to build a comprehensive and powerful intelligent breast cancer detection system by simultaneously learning the recognition capabilities of the auxiliary tasks of localization of the mass and classification of the benign or malignant nature of the tumor. This system can provide richer detection information and thus efficiently and stably serve clinical auxiliary diagnosis and treatment planning.

[0070] The deep learning-based intelligent target detection algorithm is as follows: First, an image segmentation convolutional neural network is constructed to segment the mammogram image after it has been processed by the attention mechanism module, in order to distinguish between cancerous areas and normal tissue. Each layer of the image segmentation convolutional neural network includes an activation function and a linear transformation, which can automatically learn the features of the image and identify the pattern of the tumor area, which is crucial for subsequent lesion detection. The specific formula is expressed as follows:

[0071]

[0072] in, Represented as the first The linear output of a layer is the value of the input data after it has been processed by the weights and biases of that layer. This represents the index of the layer in the image segmentation convolutional neural network. Represented as the first The layer's weight matrix is ​​used to transform the activation output of the previous layer into the linear output of the current layer. Represented as the first The activation output of a layer represents the output of the previous layer after processing by the activation function. Represented as the first Layer bias vector, Represented as the first Layer activation output, Represented as the Rectified Linear Unit activation function, it is used to introduce non-linearity, helping convolutional neural networks for image segmentation learn complex patterns.

[0073] After image segmentation, to extract effective features from the image, a grasshopper optimization search strategy is constructed. The Grasshopper Optimization Algorithm (GOA) is used to simulate the swarm behavior of grasshoppers to optimize the search process for the solution, incorporating social interaction factors. Considering the interactions among locusts, the component of gravity and wind direction component The attraction of the target to locusts and the influence of environmental wind on locust flight were simulated respectively. By combining forces, the locusts' optimized search strategy can guide the locust swarm towards the optimal solution, thereby optimizing the feature extraction process. Then, feature selection is performed to reduce dimensionality and improve classification accuracy. To select representative features from different breast lesion features, support vector machines are used for feature selection, and the specific formula is expressed as follows:

[0074]

[0075] in, This represents the final classification result of the breast lesions. Represented as a symbolic function, it is used to return the symbol of the input. This represents the total number of breast lesion samples used for training. This represents the index used to count the number of breast lesion samples for training. Represented as Lagrange multipliers, these represent data points that have a decisive impact on classification decisions. Represented as the first Labels of each training sample Represented as a kernel function, it is used to compute the input vector. With training samples Similarity between them, input vector Represented as and The joint output vector, This represents the raw mammogram data input to the target detection unit by the system. Represented as a bias term, it controls the threshold for classification decisions. By maximizing the margin between two classes, the support vector machine can find the optimal decision boundary, enabling efficient classification in high-dimensional spaces.

[0076] Finally, the processed breast cancer lesion characteristics are classified to determine the benign or malignant type of the lesions, and the classification results are output. This directly affects the practicality and accuracy of the system, specifically as follows:

[0077]

[0078] in, This represents the category label of benign or malignant breast cancer lesions predicted by the algorithm. Let represent the selection function that maximizes the subsequent probability expression. This represents the input breast cancer lesion characteristics. Category The conditional probability, The system represents the benign and malignant categories of breast cancer lesions. By calculating the probability distribution of features, the system outputs a clear classification result. This process directly determines the effectiveness of the system in actual clinical applications.

[0079] In practical use, firstly, the image acquisition and preprocessing module acquires mammograms from mammography and performs preprocessing operations, including denoising and normalization, to improve image quality and ensure clear visibility of lesion areas. Secondly, the multimodal dataset construction module integrates different types of breast imaging data, including mammograms and patient clinical information, to form a multimodal dataset for subsequent modules. Then, the multi-task classification module includes a breast data annotation unit and an attention mechanism unit. The breast data annotation unit accurately annotates lesions in mammograms, including their location, type, and benign / malignant characteristics. The attention mechanism unit proposes a breast data classification algorithm based on an attention network, which focuses on lesion areas during breast image processing and feature extraction. The processed data is then output to the breast cancer intelligent detection module, enabling the system to automatically identify lesions. The system focuses on key feature regions in the image to improve detection accuracy. Secondly, the intelligent breast cancer detection module includes a target detection unit and an iteration and optimization unit. The target detection unit proposes a deep learning-based intelligent target detection algorithm for automatically locating and intelligently identifying lesions in mammograms. Through a two-stage target detection network, the model can identify the location of masses, nodules, and calcified lesions and differentiate lesion types. The iteration and optimization unit iteratively trains breast data samples and adjusts network weight parameters to continuously optimize detection and classification performance. Finally, the evaluation and feedback module evaluates the output results of the intelligent breast cancer detection module, analyzes the system's key indicators such as detection accuracy, sensitivity, and specificity, and provides feedback based on the test results to further optimize system performance. The efficient user interaction module displays detection results and analysis information to users and generates reports, helping users quickly obtain lesion detection and classification results.

[0080] Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A deep learning-based intelligent breast cancer detection system, characterized in that, The system comprises an image acquisition and preprocessing module, a multi-modal data set construction module, a multi-task classification module, a breast cancer intelligent detection module, an evaluation and feedback module, and a user efficient interaction module. The image acquisition and preprocessing module acquires mammography images from breast X-ray photography and performs preprocessing operations on the images. The multi-modal data set construction module integrates different types of breast imaging data, including mammography images and patient clinical information, to form a multi-modal data set for subsequent module use. The multi-task classification module includes a breast data labeling unit and an attention mechanism unit. The breast data labeling unit is used to label lesions in mammography images. The attention mechanism unit is based on an attention network breast data classification algorithm that focuses on key lesion areas in breast images by applying channel and spatial attention mechanisms. The breast cancer intelligent detection module includes a target detection unit and an iteration and optimization unit. The target detection unit is based on a deep learning intelligent target detection algorithm that automatically locates and intelligently identifies lesions in mammography images and distinguishes between lesion types. The iteration and optimization unit iteratively trains breast data samples to adjust model parameters. The evaluation and feedback module evaluates the output results of the breast cancer intelligent detection module and conducts comparative experiments based on test results and public data sets. The user efficient interaction module displays detection results and analysis information to users through a user interface and generates reports. The attention network-based breast data classification algorithm is as follows: For image data processed by the breast data labeling unit, a convolutional neural network based on attention mechanism is designed, and its structure is specifically represented as: ； wherein, is represented as a convolutional neural network model, is represented as an input of a left breast image; is represented as an input of a right breast image; is represented as an output feature map of the left breast image processed by the residual network; is represented as an output feature map of the right breast image processed by the residual network, is represented as an adaptive spatial attention module, is represented as an adaptive channel attention module, is represented as a 1x1 convolutional layer, is represented as a weight set of a convolutional neural network ; is represented as output feature map data of the convolutional layer processed by the spatial attention module and the channel attention module in sequence, is represented as a fully connected layer for mapping the learned features to the final classification result. 2.The deep learning-based breast cancer intelligent detection system according to claim 1, wherein, The adaptive spatial attention module Based on the adaptive spatial attention strategy, the specific formula is expressed as: ； in, Let be the th element of the final output feature map after processing with the adaptive spatial attention strategy. Each pixel is a weighted aggregation of the input feature map. Represented as the index of the number of pixels in the final output feature map. It is represented as a learnable scale parameter. This represents the total number of pixels in the feature map after processing with the adaptive spatial attention strategy. This represents the index of the number of pixels in the input feature map. Represented as spatial attention weights, The output feature map of the left breast image, which is the input feature map processed by the residual network, is represented by the first feature map. 1 pixel, The term represents the output feature map of the convolutional layer after being processed by the spatial attention module and the channel attention module, respectively. 1 pixel; The adaptive channel attention module Based on the adaptive channel attention strategy, the specific formula is expressed as: ； in, The term is represented as the th element of the final output feature map after processing by the adaptive channel attention strategy. 1 pixel, It is represented as a learnable scale parameter. This represents the total number of pixels in the feature map after processing with the adaptive channel attention strategy. This represents the index of the number of pixels in the input feature map. Represented as channel attention weights, The output feature map of the right breast image is represented by the first... 1 pixel; The fully connected layer serves as a classifier that maps feature maps processed by the spatial and channel attention modules to breast density classification results, and the specific formula is represented as: ； wherein, are represented as classification results, corresponding to the predicted probabilities of different breast density categories, are represented as weight matrices of the fully connected layers, are represented as 1*1 convolutional layers are the integrated feature maps, are represented as bias terms of the fully connected layers, are represented as activation functions. 3.The deep learning-based breast cancer intelligent detection system according to claim 2, characterized in that, The breast data classification algorithm based on the attention network also enables the convolutional neural network model to learn the features of the breast density and the benign and malignant breast cancer through a multi-task learning strategy While performing the breast density classification, the differentiation between the benign and malignant breast cancer can be performed, and the specific formula is represented as: ； wherein, denotes an optimization model performing multi-task learning, denotes a process function representing multi-task learning, denotes a convolutional neural network model a convolutional neural network model after iterative optimization by an adaptive spatial attention strategy and an adaptive channel attention strategy, denotes auxiliary tasks including localization of a mass and classification of benign and malignant properties of a tumor. 4.The deep learning-based breast cancer intelligent detection system according to claim 3, wherein, The deep learning-based intelligent target detection algorithm is as follows: First, an image segmentation convolutional neural network is constructed to segment mammography images processed by the attention mechanism module to distinguish cancerous regions and normal tissues. After image segmentation, a locust optimization search strategy is constructed to extract effective breast cancer lesion features from the images. The locust optimization algorithm simulates the group behavior of locusts to optimize the search process of problem solutions, optimizing the feature extraction process. Then, feature selection is performed to reduce the dimensionality of breast cancer lesion features and improve classification accuracy. Finally, the processed breast cancer lesion features are classified and the classification results are output, and the specific representation is: ； wherein, a class label denoted as predicted breast cancer lesion, a probability a selection function maximized, a breast cancer lesion feature denoted as input a conditional probability of belonging to a class, a class index denoted as benign vs. malignant breast cancer lesion. 5.The deep learning-based breast cancer intelligent detection system according to claim 4, wherein, The image segmentation convolutional neural network includes an activation function and a linear transformation, and the specific formula is represented as: ； wherein, represents the index of a layer in the image segmentation convolutional neural network, represents the linear output of the layer, represents the index of a layer in the image segmentation convolutional neural network, represents the weight matrix of the layer, represents the index of a layer in the image segmentation convolutional neural network, represents the activation output of the layer, represents the index of a layer in the image segmentation convolutional neural network, represents the bias vector of the layer, represents the index of a layer in the image segmentation convolutional neural network, represents the activation output of the layer, represents the index of a layer in the image segmentation convolutional neural network, represents the Rectified Linear Unit activation function, which is used to introduce non-linearity and help the image segmentation convolutional neural network learn complex patterns. 6.The deep learning-based breast cancer intelligent detection system according to claim 5, wherein, The feature selection uses a support vector machine.

7. The deep learning-based intelligent breast cancer detection system of claim 6, wherein, The preprocessing operations include denoising and normalization of the images. 8.The deep learning-based breast cancer intelligent detection system according to claim 7, wherein, The labeling content of the breast data labeling unit includes the location, type, and benign and malignant characteristics of the lesion, and the type includes mass, nodule, and calcification.

Images