WO2020118618A1

WO2020118618A1 - Mammary gland mass image recognition method and device

Info

Publication number: WO2020118618A1
Application number: PCT/CN2018/120885
Authority: WO
Inventors: 李程; 王珊珊; 郑海荣; 刘新; 梁栋
Original assignee: 深圳先进技术研究院
Priority date: 2018-12-13
Filing date: 2018-12-13
Publication date: 2020-06-18

Abstract

A mammary gland mass image recognition method and device. The method comprises: acquiring a mammary gland magnetic resonance image to be recognized (102); inputting said image into a constructed mammary gland mass identification model to obtain a mass identification result in said image; wherein the mass recognition model uses a deep full convolutional neural network model; a coding process of the deep full convolutional neural network model uses a basic convolutional module in a U-shaped convolutional neural network model to perform feature extraction, and a decoding process of the deep full convolutional neural network model uses dense connection to fuse feature maps to be fused after performing size unification (104). According to the method, automatic recognition of the mammary gland mass is implemented, and the recognition accuracy of the mammary gland mass is improved.

Description

Breast mass image recognition method and device

Technical field

This specification belongs to the technical field of image processing, and in particular relates to a method and device for recognizing breast mass images.

Background technique

Breast cancer is the cancer with the highest incidence of women. The cure rate of early breast cancer is much higher than that of advanced breast cancer. Early detection, early diagnosis and early treatment are the keys to reducing the mortality of breast cancer. Imaging examinations including mammography, ultrasound, and magnetic resonance imaging are common techniques for early breast cancer screening. Breast cancer screening can be performed using image examination results to identify breast masses. Breast masses in different patients vary greatly in size and shape, which poses a great challenge to the automatic segmentation method, and mass segmentation is the first priority for all subsequent breast cancer diagnosis and treatment Steps, so the automatic segmentation of breast masses is of great significance to computer-aided diagnosis systems.

In the prior art, breast cancer computer-aided diagnosis systems mainly use manual extraction of features to locate or classify breast masses. Manual feature extraction depends on the professional and empirical knowledge of the researcher, and often has certain limitations and subjectivity, and the results will be greatly affected. The automatic segmentation method of breast masses in the prior art usually only gives the location of the masses, and there is no information on the shape and size of the masses. Therefore, there is an urgent need in the art for a technical solution that can accurately segment a breast mass region.

Summary of the invention

The purpose of this specification is to provide a method and device for breast mass image recognition, which realizes the automatic identification of breast masses and improves the accuracy of breast mass identification.

On the one hand, the embodiments of the present specification provide a method for recognizing breast mass images, including:

Obtain the breast magnetic resonance image to be identified;

Input the breast magnetic resonance image to be identified into the constructed breast mass identification model to obtain the mass identification result in the breast magnetic resonance image to be identified;

Wherein, the mass recognition model uses a deep full convolutional neural network model, and the encoding process of the deep full convolutional neural network model uses the basic convolution module in the U-shaped convolutional neural network model for feature extraction. The decoding process of the convolutional neural network model uses dense connections to unify the feature maps to be fused and then fused.

Further, in another embodiment of the method, the basic convolution module in the U-shaped convolutional neural network model is used to extract features using the following formula:

y ₁ =δ(W ₁₃ *δ(W ₁₂ *δ(W ₁₁ *x ₁ +b ₁₁ )+b ₁₂ )+b ₁₃ )

In the above formula, y ₁ represents the extracted feature map, x ₁ represents the input breast magnetic resonance image, δ represents the first activation function, W ₁₁ , W ₁₂ , W ₁₃ represent the corresponding weights of different convolutional layers, b ₁₁ , b ₁₂ and b ₁₃ represent offset parameters corresponding to different convolutional layers.

Further, in another embodiment of the method, the method further includes: using the deep fully convolutional neural network model to adjust the extracted feature map using the following formula:

y ₂ =δ((W ₂₂ *δ(W ₂₁ *x ₂ +b ₂₁ )+b ₂₂ )+x ₂ )

In the above formula, y ₂ represents the adjusted feature map, x ₂ represents the extracted feature map, δ represents the first activation function, W ₂₁ and W ₂₂ represent the weights corresponding to different convolutional layers, and b ₂₁ and b ₂₂ represent different The offset parameter corresponding to the convolutional layer.

Further, in another embodiment of the method, the method further includes:

Using the deep fully convolutional neural network model to assign different weight values to the extracted feature map;

Feature fusion is performed according to the weight value corresponding to each feature map.

Further, in another embodiment of the method, the assigning the extracted feature map with different weight values includes:

Perform average pooling on the extracted feature maps;

The averaged pooled feature map is connected using a fully connected layer, and is activated using a second activation function to obtain a weight value corresponding to the feature map.

Further, in another embodiment of the method, the assigning the extracted feature map with different weight values includes: using the following formula to obtain the weight value corresponding to the feature map:

In the above formula, Z _c represents the pooling result corresponding to the c-th feature map averaged by the feature map, H represents the height of the c-th feature map, M represents the width of the c-th feature map, and x _c represents the input c-th feature map Feature maps, S represents the weight value vector corresponding to the feature map, Z represents the pooled result vector after the feature map is pooled, σ represents the second activation function, the weights corresponding to different fully connected layers of W ₃₂ and W ₃₁ , b ₃₁ and b ₃₂ represent offset parameters corresponding to different fully connected layers.

Further, in another embodiment of the method, the breast mass identification model is constructed using the following method:

Acquire multiple sample data, the sample data includes: a breast magnetic resonance image and a tumor marker in the breast magnetic resonance image;

Establish the breast mass identification model, use the breast magnetic resonance image in the sample data as the input data of the breast mass identification model, and use the corresponding mass marker in the breast magnetic resonance image as the breast mass identification model Output data, training the breast mass identification model until the breast mass identification model meets the preset requirements.

Further, in another embodiment of the method, the method further includes:

Comparing the image parameters of the breast magnetic resonance image to be identified with the breast magnetic resonance image in the sample data;

If the image parameters of the breast magnetic resonance image to be identified and the breast magnetic resonance image in the sample data are inconsistent, use the breast magnetic resonance image to be identified to adjust the breast mass recognition model;

Using the adjusted breast mass identification model, mass identification of the breast magnetic resonance image to be identified is performed.

On the other hand, this specification provides a breast mass image recognition device, including:

The image acquisition module is used to acquire the breast magnetic resonance image to be identified;

The lump identification module is used to input the breast magnetic resonance image to be identified into the constructed breast mass identification model to obtain the mass recognition result in the breast magnetic resonance image to be identified;

Further, in another embodiment of the device, the basic convolution module in the U-shaped convolutional neural network model uses the following formula for feature extraction:

Further, in another embodiment of the device, the deep fully convolutional neural network model includes a feature adaptation module for adjusting the extracted feature map using the following formula:

y ₂ =δ((W ₂₂ *δ(W ₂₁ *x ₂ +b ₂₁ )+b ₂₂ )+x ₂ )

Further, in another embodiment of the device, the deep fully convolutional neural network model further includes a channel attention module for:

Further, in another embodiment of the device, the channel attention module is specifically used to:

Perform average pooling on the extracted feature maps;

Further, in another embodiment of the device, the channel attention module is specifically used to obtain the weight value corresponding to the feature map using the following formula:

Further, in another embodiment of the device, the device further includes a model building module for building the breast mass identification model using the following method:

Establishing the breast mass identification model, wherein the breast mass identification model includes multiple model parameters;

The breast magnetic resonance image in the sample data is used as the input data of the breast mass identification model, and the corresponding tumor marker in the breast magnetic resonance image is used as the output data of the breast mass identification model to adjust the breast The model parameters of the lump identification model until the breast lump identification model meets the preset requirements.

Further, in another embodiment of the device, the device further includes a model adjustment module for:

In still another aspect, this specification provides a breast mass image recognition processing device, including: at least one processor and a memory for storing processor-executable instructions, and when the processor executes the instructions, the breast in the embodiments of the specification is implemented Lump image recognition method.

In still another aspect, the present specification provides a breast mass image recognition system, including at least one processor and a memory for storing processor executable instructions, and when the processor executes the instructions, the breast in the embodiments of the specification is implemented Lump image recognition method.

The breast mass image recognition method, device, processing equipment and system provided in this specification, based on deep learning, combine the U-shaped convolutional neural network model with the dense convolutional neural network model to construct an asymmetric encoding and decoding breast mass identification model structure. Then input the breast magnetic resonance image to be recognized into the constructed breast mass recognition model, that is, to obtain the breast mass recognition result of the breast magnetic resonance image to be recognized, the automatic identification of the breast mass is realized, and the artificial naked eye recognition is not required, which improves Breast lump identification results. The breast mass image recognition method provided by the implementation of this specification can better retain the beneficial effects of useful features on the segmentation results, and weaken the role of useless features. The effect of the model on segmentation of MRI breast masses is greatly improved, and subsequent network models are not required to further optimize the segmentation results, which reduces the calculation cost, speeds up the image analysis process, and can better assist doctors in real-time image diagnosis.

BRIEF DESCRIPTION

In order to more clearly explain the embodiments of the specification or the technical solutions in the prior art, the following will briefly introduce the drawings required in the embodiments or the description of the prior art. Obviously, the drawings in the following description are only These are some of the embodiments described in this specification. For those of ordinary skill in the art, without paying any creative labor, other drawings can also be obtained based on these drawings.

FIG. 1 is a schematic flowchart of a breast mass image recognition method in an embodiment of this specification;

2 is a schematic diagram of a network architecture of a breast mass identification model in an embodiment of this specification;

FIG. 3 is a schematic diagram of a module structure of an embodiment of a breast mass image recognition device provided in this specification;

4 is a schematic structural diagram of a breast mass image recognition device according to another embodiment of the present specification;

5 is a schematic structural diagram of a breast mass image recognition device according to yet another embodiment of the present specification;

6 is a block diagram of the hardware structure of a breast mass identification server using an embodiment of the present application.

detailed description

In order to enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be described clearly and completely in conjunction with the drawings in the embodiments of this specification. Obviously, the described The embodiments are only a part of the embodiments of this specification, but not all the embodiments. Based on the embodiments in this specification, all other embodiments obtained by those of ordinary skill in the art without creative work shall fall within the protection scope of this specification.

The breast mass image recognition method in this manual can be applied to the client or server. The client can be a smart phone, tablet computer, smart wearable device (smart watch, virtual reality glasses, virtual reality helmet, etc.), smart vehicle equipment and other electronic equipment.

Specifically, FIG. 1 is a schematic flowchart of a breast mass image recognition method in an embodiment of the present specification. As shown in FIG. 1, the overall process of the breast mass image recognition method provided in an embodiment of the present specification may include:

Step 102: Acquire a breast magnetic resonance image to be identified.

Magnetic resonance examination is currently a relatively common medical examination method. In the embodiment of this specification, a user's breast magnetic resonance image can be acquired, and the breast mass can be identified based on the acquired breast magnetic resonance image to be identified.

Step 104: Input the breast magnetic resonance image to be identified into the constructed breast mass identification model to obtain the mass identification result in the breast magnetic resonance image to be identified;

In the specific implementation process, a breast mass identification model can be constructed based on deep learning methods. For example, the existing breast cancer patient’s breast magnetic resonance image data can be used for model training to learn from input magnetic resonance image to output breast A function mapping relationship between the results of mass segmentation to construct a breast mass identification model. 2 is a schematic diagram of a network architecture of a breast mass identification model in an embodiment of the present specification. As shown in FIG. 2, the breast mass identification model in the embodiment of the present specification may be a deep fully convolutional neural network model, which can be U-shaped The convolutional neural network model (ie U-Net) is combined with the dense convolutional neural network model (DenseNet). In the encoding process, the basic convolution module in the U-shaped convolutional neural network model can be used for feature extraction. In the decoding process, the dense connections in the dense convolutional neural network model can be used to unify the feature maps to be fused, and then feature fusion .

U-Net can be understood as a variant of the convolutional neural network, whose structure is mainly like the letter U, hence the name U-Net. The entire neural network of U-Net is mainly composed of two parts: the contraction path and the expansion path. The contraction path is mainly used to capture the context information in the picture, and the expansion path commensurate with it is to segment the need in the picture. For precise positioning. DenseNet can be understood as a convolutional neural network with dense connections. In this network, there is a direct connection between any two layers, that is, the input of each layer of the network is the union of the outputs of all previous layers. , And the feature map learned by this layer will be directly passed to all subsequent layers as input. Dense connections can alleviate the problem of gradient disappearance, strengthen feature propagation, encourage feature reuse, greatly reduce the amount of parameters, and improve the accuracy of image recognition.

As shown in Figure 2, the basic convolution module on the left in Figure 2 can use a basic convolution module similar to the U-shaped convolutional neural network model for feature extraction, and the downward arrow of the basic convolution module can indicate maximum pooling . The channel attention module on the right side in FIG. 2 can be understood as a decoding process, and the upward arrow of the channel attention module can represent bilinear interpolation, where the decoding process can perform feature fusion together on the feature maps of the coding process of the same layer. The connecting line with arrows on the right side of the channel attention module in FIG. 2 may represent a dense connection. In some embodiments of this specification, the network module of the decoding process of the deep fully convolutional neural network model may be densely connected, that is, the The channel attention module makes intensive connections. The channel attention module can uniformly size the feature maps extracted from the coding process of the same layer and the feature maps of other channel attention modules, and then perform feature fusion.

For example: the uppermost channel attention module in Figure 2 can unify the feature map obtained by the encoding process on the left, the feature map output by the three channel attention modules below it, and the feature map output by the lowest feature adaptation module Size, and then merge the feature maps of uniform size. Among them, the feature map obtained by the uppermost channel attention module can be obtained by the encoding process on the left, the feature map output by the three channel attention modules below it, and the feature map output by the lowermost feature adaptation module can be understood as the uppermost channel The feature map of the attention module to be fused.

In the embodiment of the present specification, dense connections are added in the decoding process. These dense connections can separately upsample feature maps with different degrees of abstraction to a uniform size, and then directly merge them. Feature maps are reused. Feature maps with a high degree of abstraction can better guide the classification results, while feature maps with a low degree of abstraction can better retain location information, and the segmentation results are greatly improved.

After the breast mass identification model is constructed, the breast magnetic resonance image to be identified is input into the constructed breast mass identification model, and the breast mass identification result of the breast magnetic resonance image to be identified is obtained using the breast mass identification model. For example, the mass to be identified can be identified Whether there is a lump in the breast magnetic resonance image, and if so, it can also identify the area where the lump is located, or the shape and size of the breast lump. The bright spot in the output image of the rightmost breast mass recognition model in Figure 2 can represent the segmented mass, that is, the area of the mass in the breast magnetic resonance image input on the left, and the doctor can make a further diagnosis based on the identified mass area And treatment, or used for other medical research.

The breast mass image recognition method provided in the embodiment of the present specification combines the U-shaped convolutional neural network model with the dense convolutional neural network model based on deep learning to construct an asymmetric encoding and decoding breast mass identification model structure. Then input the breast magnetic resonance image to be recognized into the constructed breast mass recognition model, that is, to obtain the breast mass recognition result of the breast magnetic resonance image to be recognized, the automatic identification of the breast mass is realized, and the artificial naked eye recognition is not required, which improves Breast lump identification results. The breast mass image recognition method provided by the implementation of this specification can better retain the beneficial effects of useful features on the segmentation results, and weaken the role of useless features. Greatly improve the effect of network model on the segmentation of MRI breast masses, without the need for further network model to further optimize the segmentation results, reduce the calculation cost, speed up the image analysis process, and better assist doctors in real-time image diagnosis.

Based on the above embodiments, in some embodiments of this specification, the basic convolution module in the U-shaped convolutional neural network model can be used to extract features using the following formula:

y ₁ =δ(W ₁₃ *δ(W ₁₂ *δ(W ₁₁ *x ₁ +b ₁₁ )+b ₁₂ )+b ₁₃ ) (1)

In the above formula, y ₁ represents the extracted feature map that is the output of the basic convolution module, x ₁ represents the input breast magnetic resonance image that is the input of the basic convolution module, and δ represents the first activation function (which can be a linear rectification function) , W ₁₁ , W ₁₂ , and W ₁₃ represent weights corresponding to different convolutional layers, and b ₁₁ , b ₁₂ , and b ₁₃ represent offset parameters corresponding to different convolutional layers.

In a specific implementation process, as shown in FIG. 2, the basic convolution module in the figure may include the function of the above formula (1), and use the above formula (1) for feature extraction. The basic convolution module can perform multiple convolution operations, that is, the basic convolution module can include multiple convolutional layers, and different convolutional layers have different weights W ₁₁ , W ₁₂ , W ₁₃ and offset parameters b ₁₁ , b ₁₂ and b ₁₃ . In the embodiment of the present specification, the basic convolution module may include three convolutional layers, and other numbers of convolutional layers may be set according to actual needs. According to the number of convolutional layers, the above formula (1) may be adaptively adjusted or Modifications are not specifically limited in the embodiments of this specification.

The basic convolution modules can be connected by maximum pooling. The maximum pooling can increase the receptive field and to a certain extent achieve the invariance of the input image translation. After the feature map passes through each maximum pooling layer, the resolution is halved, and the channel Number doubled.

The embodiments of this specification use a basic convolution module similar to U-Net for feature extraction, which can reduce the amount of sample data and improve the efficiency of data processing.

Based on the above embodiment, in one embodiment of the present specification, the method may further include: using the deep fully convolutional neural network model to adjust the extracted feature map using the following formula:

y ₂ =δ((W ₂₂ *δ(W ₂₁ *x ₂ +b ₂₁ )+b ₂₂ )+x ₂ ) (2)

In the above formula, y ₂ represents the adjusted feature map, that is, the output of the feature adaptation module, x ₂ represents the extracted feature map, that is, the input of the feature adaptation module, δ represents the first activation function (which may be a linear rectification function), W ₂₁ , W ₂₂ represents weights corresponding to different convolutional layers, and b ₂₁ , b ₂₂ represent offset parameters corresponding to different convolutional layers.

In the specific implementation process, as shown in FIG. 2, the deep fully convolutional neural network model in the embodiment of this specification, that is, the breast mass recognition model, can also include a feature application module, and the feature adaptation module can adjust the basic convolution in the encoding process The feature map generated by the module to enhance the subsequent feature fusion effect. The feature application module may include the function of the above formula (2), and use the above formula (2) to adjust the feature map. The feature adaptation module may also include multiple convolutional layers, and different convolutional layers may correspond to different The weights W ₂₁ , W ₂₂ and b ₂₁ , b ₂₂ . In the embodiment of this specification, the feature adaptation module may include 2 convolutional layers, and other numbers of convolutional layers may be set according to actual needs. The above formula (2) may be adaptively adjusted or modified according to the number of convolutional layers The examples in this specification are not specifically limited. The feature maps with different levels of abstraction generated in the encoding process can be fused with the feature maps in the decoding process in a cascade manner after passing through the feature adaptation module.

The embodiment of the present specification uses the feature adaptation module in the breast mass recognition model to adjust the feature map, improve the effect of feature fusion, and further improve the accuracy of breast mass identification.

Based on the above embodiments, in an embodiment of this specification, the method further includes:

In the specific implementation process, the embodiments of this specification can assign different feature maps with different weight values, such as: setting the weight value of some feature maps with more information to be higher, and reducing the amount of information or useless The weight value of the feature map is set relatively low. Then perform feature fusion according to the weight value corresponding to the feature map, for example: you can multiply the weight value corresponding to each feature map to the feature map to be fused, increase the impact of the important feature map on the mass recognition result, and reduce the unimportant feature map Impact on the results of mass identification.

In an embodiment of the present specification, the extracted feature maps can be averagely pooled, for example: the feature maps adjusted by the feature adaptation module are averagely pooled, then through two fully connected layers, and finally through the second activation function as : The Sigmoid function is activated to generate the weight of each feature map, and then the weight value is multiplied back to the fused feature map. Average pooling can mean averaging all the values in the local acceptance domain. The Sigmoid function is a common S-shaped function in biology, and it can also be called an S-shaped growth curve.

In an embodiment of this specification, the following formula (3) may be used to obtain the weight value corresponding to each feature map:

In the above formula, Z _c represents the pooling result corresponding to the c-th feature map averaged by the feature map, H represents the height of the c-th feature map, M represents the width of the c-th feature map, and i can represent 1-H Numerical value, j can represent the value of 1-M, x _c can represent the c-th feature map input by the channel attention module, S represents the weight value vector corresponding to the feature map (which can include the weight values corresponding to multiple feature maps), Z represents the pooling result vector after the feature map is pooled (which may include pooling results corresponding to multiple feature maps, such as: Z _c ), σ represents the second activation function (which may be a Sigmoid function), W ₃₂ , Weights corresponding to different fully connected layers of W ₃₁ , b ₃₁ and b ₃₂ represent offset parameters corresponding to different fully connected layers.

After obtaining the weight values corresponding to different feature maps, the product of the weight values corresponding to each feature map and the feature map can be used as an output for feature fusion. Such as:

y ₃ =S _c ·x _c (4)

In the above formula, y ₃ can represent the output of the channel attention module, S _c can represent the weight value corresponding to the c-th feature map, and x _c can represent the c-th feature map input by the channel attention module.

It should be noted that the formulas in the embodiments of the present specification are only schematic representations, and the above formulas can be adjusted, transformed, or modified according to actual needs, and the embodiments of the present specification are not specifically limited.

In the embodiments of the present specification, the channel attention module is used to assign different feature maps with different weight values and then perform feature fusion to realize the screening of the feature maps, which can improve the influence of important feature maps and further improve the accuracy of the mass recognition results.

Based on the above embodiments, in the embodiments of the present specification, a breast mass identification model can be constructed in the following manner:

In a specific embodiment, the breast magnetic resonance image of a historical user may be acquired as sample data, and the sample data may be a breast magnetic resonance image of a user who has been diagnosed with breast cancer. The sample data may also include the acquired mass markers in the breast magnetic resonance image as training labels. The specific number of sample data may be selected according to actual needs, which is not specifically limited in the embodiments of this specification. In one embodiment of the present specification, after acquiring multiple breast magnetic resonance images, the acquired breast magnetic resonance images can be normalized, that is, the pixels of the breast magnetic resonance images in the sample data are processed in a unified manner, which is convenient Follow-up model training. Then, the normalized breast magnetic resonance image is labeled with a mass, which can be labeled by a professional doctor or based on the user's diagnosis, and the location, size, etc. of the mass can be marked.

After the sample data preparation is completed, a breast mass identification model may be constructed, such as: constructing a network architecture of the breast mass identification model, etc. For the network architecture of the breast mass identification model, reference may be made to the records of the foregoing embodiments, and details are not described here. As shown in FIG. 2, in some embodiments of this specification, the lump recognition model may include a basic convolution module, a feature adaptation module, and a channel attention module, where the basic convolution module and feature adaptation module may be understood as an encoding process, channel attention The module can be understood as the decoding process. The breast mass identification model may also include multiple model parameters, such as: the size of the convolution kernel and the number of convolution layers. After the breast mass identification model is constructed, the breast magnetic resonance image in the sample data can be used as the input data of the breast mass identification model, and the corresponding mass marker in the breast magnetic resonance image can be used as the output data of the breast mass identification model. The recognition model performs model training until the breast mass recognition model meets the preset requirements, such as: the model output accuracy meets the requirements or the model training times meet the requirements, that is, the model training is concluded.

The embodiment of the present specification uses deep learning training to construct a breast mass identification model, which can realize the automatic identification of breast masses, without the need for manual identification, and improves the accuracy of breast mass identification.

The breast mass identification in the embodiment of this specification mainly includes two stages of model training and prediction. On the basis of the above embodiment, in one embodiment of this specification, in the training stage, existing breast magnetic resonance data can be used to design The deep fully convolutional network performs optimization and parameter learning. In the testing stage, the trained network model is used to analyze new data that has not been seen during training, that is, the breast magnetic resonance image to be identified. When performing mass identification of the breast magnetic resonance image to be identified, the image parameters of the breast magnetic resonance image to be identified and the breast magnetic resonance image in the sample data of the training stage (such as: main field intensity, relaxation time T1, T2, etc.) ) Compare and compare whether the distribution of the two image data is uniform, that is, whether the image contrast, signal to noise ratio, etc. are consistent. If the breast magnetic resonance image and sample data to be identified conform to a uniform distribution, the breast mass identification model constructed by training can be used directly for mass identification. When the breast magnetic resonance image to be identified and the sample data do not satisfy the uniform distribution, a small amount of new data, that is, the breast magnetic resonance image to be identified (or an image with the same parameters as the breast magnetic resonance image to be identified) can be used to construct the breast Quickly fine-tune the parameters of the lump identification model. The adjusted breast mass identification model is used to identify the mass of the breast magnetic resonance image to be identified, and the breast mass identification model is adjusted through continuous updates to improve the applicability of the model and further improve the accuracy of the mass identification results.

The embodiment of the present specification proposes an asymmetric codec main frame structure, designs a new feature fusion and screening mechanism, and introduces dense connections. The segmentation method in the embodiment of the present specification can better retain the beneficial influence of useful features on the segmentation result, and weaken the role of useless features. As a result, the effect of the network on the segmentation of the magnetic resonance breast mass can be greatly improved, and the subsequent network is not required to further optimize the segmentation result, which reduces the calculation cost, speeds up the image analysis process, and can better assist the doctor in real-time image diagnosis.

It should be noted that the breast mass image recognition method in the embodiment of the present specification may not be limited to identifying breast masses, but may also be used in other image recognition processes, such as: identifying other lesion areas (such as brain tumors) and the like. You can use the magnetic resonance image training of other parts to build a corresponding recognition model to complete the automatic recognition of other lesion areas.

The embodiments of the above method in this specification are described in a progressive manner. The same or similar parts between the embodiments can be referred to each other. Each embodiment focuses on the differences from other embodiments. For the relevant parts, please refer to the description of the method embodiments.

Based on the breast mass image recognition method described above, one or more embodiments of the present specification further provide a breast mass image recognition device. The device may include a system (including a distributed system), software (applications), modules, components, servers, clients, etc. using the method described in the embodiments of the present specification in combination with necessary hardware implementation devices. Based on the same innovative concept, the devices in one or more embodiments provided by the embodiments of this specification are as described in the following embodiments. Since the implementation solution of the device to solve the problem is similar to the method, the implementation of the specific device in the embodiments of the present specification may refer to the implementation of the foregoing method, and the repetition is not repeated. As used below, the term "unit" or "module" may implement a combination of software and/or hardware that achieves a predetermined function. Although the devices described in the following embodiments are preferably implemented in software, implementation of hardware or a combination of software and hardware is also possible and conceived.

Specifically, FIG. 3 is a schematic diagram of a module structure of an embodiment of the breast mass image recognition device provided in this specification. As shown in FIG. 3, the breast mass image recognition device provided in this specification includes: an image acquisition module 31 and a mass identification module 32 ,among them:

The image acquisition module 31 can be used to acquire a breast magnetic resonance image to be identified;

The mass identification module 32 may be used to input the breast magnetic resonance image to be identified into the constructed breast mass identification model to obtain the mass identification result in the breast magnetic resonance image to be identified;

The breast mass image recognition device provided in the embodiment of the present specification combines the U-shaped convolutional neural network model with the dense convolutional neural network model based on deep learning to construct an asymmetric encoding and decoding breast mass identification model structure. Then input the breast magnetic resonance image to be recognized into the constructed breast mass recognition model, that is, to obtain the breast mass recognition result of the breast magnetic resonance image to be recognized, the automatic identification of the breast mass is realized, and the artificial naked eye recognition is not required, which improves Breast lump identification results. The breast mass image recognition method provided by the implementation of this specification can better retain the beneficial effects of useful features on the segmentation results, and weaken the role of useless features. It can greatly improve the effect of the network on the segmentation of magnetic resonance breast masses, without the need for further network optimization of the segmentation results, reducing the calculation cost, speeding up the image analysis process, and better assisting doctors in real-time image diagnosis.

Based on the above embodiments, the basic convolution module in the U-shaped convolutional neural network model uses the following formula for feature extraction:

The breast mass image recognition device provided by the embodiment of the present specification uses a basic convolution module similar to U-Net for feature extraction, which can reduce the amount of sample data and improve the efficiency of data processing.

Based on the above embodiment, the deep fully convolutional neural network model includes a feature adaptation module for adjusting the extracted feature map using the following formula:

y ₂ =δ((W ₂₂ *δ(W ₂₁ *x ₂ +b ₂₁ )+b ₂₂ )+x ₂ )

Based on the above embodiments, the deep fully convolutional neural network model further includes a channel attention module for:

In the embodiment of the present specification, different feature maps are given different weight values, the influence of the important feature map on the mass recognition result is increased, and the influence of the unimportant feature map on the mass recognition result is reduced.

Based on the above embodiments, the channel attention module is specifically used to:

Perform average pooling on the extracted feature maps;

Based on the above embodiment, the channel attention module is specifically used to obtain the weight value corresponding to the feature map using the following formula:

FIG. 4 is a schematic structural diagram of a breast mass image recognition device in another embodiment of the present specification. As shown in FIG. 4, on the basis of the above embodiment, the device further includes a model building module 41 for constructing the The breast mass identification model:

In the embodiment of the present specification, the breast mass identification model is constructed by using deep learning training, which can realize the automatic identification of the breast mass without manual identification, and improves the accuracy of the breast mass identification.

FIG. 5 is a schematic structural diagram of a breast mass image recognition device in another embodiment of the present specification. As shown in FIG. 5, based on the above embodiment, the device further includes a model adjustment module 51 for:

In the embodiment of the present specification, the breast mass identification model is adjusted through continuous updates to improve the applicability of the model and further improve the accuracy of the mass identification results.

It should be noted that the above description of the device according to the method embodiment may also include other implementations. For a specific implementation manner, reference may be made to the description of related method embodiments, and details are not repeated herein.

An embodiment of the present specification also provides a breast mass image recognition processing device, including: at least one processor and a memory for storing processor executable instructions, and the processor implements the instructions to implement the breast mass image of the above embodiment Identification methods, such as:

Obtain the breast magnetic resonance image to be identified;

The storage medium may include a physical device for storing information, usually after the information is digitized and then stored in a medium using electrical, magnetic, or optical means. The storage medium may include: devices that use electrical energy to store information, such as various types of memory, such as RAM, ROM, etc.; devices that use magnetic energy to store information, such as hard disks, floppy disks, magnetic tapes, magnetic core memories, and bubble memories, U disk; a device that uses optical means to store information such as CD or DVD. Of course, there are other ways of readable storage media, such as quantum memory, graphene memory, and so on.

It should be noted that the above description of the processing device according to the method embodiment may also include other implementation manners. For a specific implementation manner, reference may be made to the description of related method embodiments, and details are not repeated herein.

The breast mass identification system provided in this manual can be a separate breast mass identification system, or it can be applied to various data analysis and processing systems. The system may include any breast mass image recognition device in the above embodiments. The system may be a separate server, or it may include a server cluster, system (including distributed system), software (application) using one or more of the methods or one or more embodiments of this specification. Terminal devices that actually operate devices, logic gate devices, quantum computers, etc., combined with the necessary implementation hardware. The detection system for checking the difference data may include at least one processor and a memory storing computer-executable instructions. When the processor executes the instructions, the steps of the method in any one or more of the above embodiments are implemented.

The method embodiments provided in the embodiments of this specification can be executed in a mobile terminal, a computer terminal, a server, or a similar computing device. Taking running on a server as an example, FIG. 6 is a block diagram of a hardware structure of a breast mass identification server using an embodiment of the present application. As shown in FIG. 6, the server 10 may include one or more (only one is shown in the figure) processor 100 (the processor 100 may include but is not limited to a processing device such as a microprocessor MCU or a programmable logic device FPGA), A memory 200 for storing data, and a transmission module 300 for communication functions. A person of ordinary skill in this neighborhood can understand that the structure shown in FIG. 6 is merely an illustration, which does not limit the structure of the foregoing electronic device. For example, the server 10 may also include more or fewer components than those shown in FIG. 6, for example, it may also include other processing hardware, such as a database or a multi-level cache, a GPU, or have a configuration different from that shown in FIG.

The memory 200 may be used to store software programs and modules of application software, such as program instructions/modules corresponding to the breast mass image recognition method in the embodiments of the present specification, and the processor 100 executes the software programs and modules stored in the memory 200 to execute Various functional applications and data processing. The memory 200 may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some examples, the memory 200 may further include memories remotely provided with respect to the processor 100, and these remote memories may be connected to a computer terminal through a network. Examples of the above network include but are not limited to the Internet, intranet, local area network, mobile communication network, and combinations thereof.

The transmission module 300 is used to receive or send data via a network. The specific example of the network described above may include a wireless network provided by a communication provider of computer terminals. In one example, the transmission module 300 includes a network adapter (Network Interface Controller, NIC), which can be connected to other network devices through the base station to communicate with the Internet. In one example, the transmission module 300 may be a radio frequency (RF) module, which is used to communicate with the Internet in a wireless manner.

The foregoing describes specific embodiments of the present specification. Other embodiments are within the scope of the following claims. In some cases, the actions or steps recited in the claims may be performed in a different order than in the embodiments and still achieve the desired results. In addition, the processes depicted in the drawings do not necessarily require the particular order shown or sequential order to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

The method or apparatus described in the above embodiments provided in this specification can implement business logic through a computer program and be recorded on a storage medium, and the storage medium can be read and executed by a computer to achieve the effects of the solutions described in the embodiments of this specification.

The above-mentioned breast mass image recognition method or device provided by the embodiment of the present specification can be implemented by a processor executing corresponding program instructions in a computer, such as using the Windows operating system C++ language to implement on the PC side, Linux system, or other, for example, using Android and iOS system programming languages are implemented in smart terminals, and quantum logic-based processing logic.

It should be noted that the description of the device, computer storage medium, and system described above in the specification according to the related method embodiments may also include other implementation manners. For specific implementation manners, reference may be made to the description of the corresponding method embodiments, and details are not repeated here. .

The embodiments in this specification are described in a progressive manner. The same or similar parts between the embodiments can be referred to each other. Each embodiment focuses on the differences from other embodiments. In particular, for the hardware + program embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant part can be referred to the description of the method embodiment.

The embodiments of this specification are not limited to those that must comply with industry communication standards, standard computer data processing and data storage rules, or those described in one or more embodiments of this specification. Some industry standards or implementations described in a custom manner or embodiments based on slightly modified implementations can also achieve the same, equivalent, or similar, or predictable implementation effects of the foregoing embodiments. Examples obtained by applying these modified or deformed data acquisition, storage, judgment, processing methods, etc., can still fall within the scope of optional implementations of the examples in this specification.

In the 1990s, the improvement of a technology can be clearly distinguished from the improvement in hardware (for example, the improvement of circuit structures such as diodes, transistors, and switches) or the improvement in software (the improvement of the process flow). However, with the development of technology, the improvement of many methods and processes can be regarded as a direct improvement of the hardware circuit structure. Designers almost get the corresponding hardware circuit structure by programming the improved method flow into the hardware circuit. Therefore, it cannot be said that the improvement of a method flow cannot be realized by hardware physical modules. For example, a programmable logic device (Programmable Logic Device, PLD) (such as a field programmable gate array (Field Programmable Gate Array, FPGA)) is such an integrated circuit, and its logic function is determined by the user programming the device. Designers can program themselves to "integrate" a digital system on a PLD without having to ask chip manufacturers to design and make dedicated integrated circuit chips. Moreover, nowadays, instead of manually making integrated circuit chips, this kind of programming is also mostly implemented with "logic compiler" software, which is similar to the software compiler used in program development and writing, but before compilation The original code must also be written in a specific programming language, which is called hardware description language (Hardware Description Language, HDL), and HDL is not only one kind, but there are many kinds, such as ABEL (Advanced Boolean Expression) Language , AHDL (AlteraHardwareDescriptionLanguage), Confluence, CUPL (CornellUniversityProgrammingLanguage), HDCal, JHDL (JavaHardwareDescriptionLanguage), Lava, Lola, MyHDL, PALASM, RHDL (RubyHardwareDescription) It is VHDL (Very-High-Speed Integrated Circuit Hardware Description) and Verilog. Those skilled in the art should also be clear that by simply programming the method flow in the above hardware description languages and programming into the integrated circuit, the hardware circuit that implements the logic method flow can be easily obtained.

The controller may be implemented in any suitable manner, for example, the controller may take a microprocessor or processor and a computer-readable medium storing computer-readable program code (such as software or firmware) executable by the (micro)processor , Logic gates, switches, application specific integrated circuits (Application Specific Integrated Circuit, ASIC), programmable logic controllers and embedded microcontrollers. Examples of controllers include but are not limited to the following microcontrollers: ARC625D, Atmel AT91SAM, Microchip PIC18F26K20 and Silicon Labs C8051F320, the memory controller can also be implemented as part of the control logic of the memory. Those skilled in the art also know that, in addition to implementing the controller in the form of pure computer-readable program code, it is entirely possible to logically program method steps to make the controller use logic gates, switches, application specific integrated circuits, programmable logic controllers and embedded To achieve the same function in the form of a microcontroller, etc. Therefore, such a controller can be regarded as a hardware component, and the device for implementing various functions included therein can also be regarded as a structure within the hardware component. Or even, the means for realizing various functions can be regarded as both a software module of an implementation method and a structure within a hardware component.

The system, device, module or unit explained in the above embodiments may be specifically implemented by a computer chip or entity, or implemented by a product with a certain function. A typical implementation device is a computer. Specifically, the computer may be, for example, a personal computer, a laptop computer, an on-board human-machine interaction device, a cellular phone, a camera phone, a smart phone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet A computer, a wearable device, or any combination of these devices.

Although one or more embodiments of this specification provide method operation steps as described in the embodiments or flowcharts, more or fewer operation steps may be included based on conventional or non-inventive means. The order of the steps listed in the embodiment is only one way among the order of execution of many steps, and does not represent a unique order of execution. When the actual device or terminal product is executed, it may be executed sequentially or in parallel according to the method shown in the embodiments or the drawings (for example, a parallel processor or multi-threaded processing environment, or even a distributed data processing environment). The terms "include", "include" or any other variation thereof are intended to cover non-exclusive inclusion, so that a process, method, product, or device that includes a series of elements includes not only those elements, but also others that are not explicitly listed Elements, or also include elements inherent to such processes, methods, products, or equipment. Without more restrictions, it does not exclude that there are other identical or equivalent elements in the process, method, product or equipment including the elements. The first and second words are used to indicate names, but do not indicate any particular order.

For the convenience of description, when describing the above device, the functions are divided into various modules and described separately. Of course, when implementing one or more of this specification, the functions of each module may be implemented in the same or more software and/or hardware, or the modules that achieve the same function may be implemented by a combination of multiple submodules or subunits, etc. . The device embodiments described above are only schematic. For example, the division of the unit is only a division of logical functions. In actual implementation, there may be another division manner, for example, multiple units or components may be combined or integrated To another system, or some features can be ignored, or not implemented. In addition, the displayed or discussed mutual coupling or direct coupling or communication connection may be indirect coupling or communication connection through some interfaces, devices or units, and may be in electrical, mechanical or other forms.

The present invention is described with reference to flowcharts and/or block diagrams of methods, apparatuses (systems), and computer program products according to embodiments of the present invention. It should be understood that each flow and/or block in the flowchart and/or block diagram and a combination of the flow and/or block in the flowchart and/or block diagram may be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, special-purpose computer, embedded processing machine, or other programmable data processing device to produce a machine that enables the generation of instructions executed by the processor of the computer or other programmable data processing device An apparatus for realizing the functions specified in one block or multiple blocks of one flow or multiple flows of a flowchart and/or one block or multiple blocks of a block diagram.

These computer program instructions may also be stored in a computer-readable memory that can guide a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture including an instruction device, the instructions The device implements the functions specified in one block or multiple blocks of the flowchart one flow or multiple flows and/or block diagrams.

These computer program instructions can also be loaded onto a computer or other programmable data processing device, so that a series of operating steps are performed on the computer or other programmable device to produce computer-implemented processing, which is executed on the computer or other programmable device The instructions provide steps for implementing the functions specified in one block or multiple blocks of the flowchart one flow or multiple flows and/or block diagrams.

In a typical configuration, the computing device includes one or more processors (CPUs), input/output interfaces, network interfaces, and memory.

The memory may include non-permanent memory, random access memory (RAM) and/or non-volatile memory in computer-readable media, such as read only memory (ROM) or flash memory (flash RAM). Memory is an example of computer-readable media.

Computer-readable media, including permanent and non-permanent, removable and non-removable media, can store information by any method or technology. The information may be computer readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, read-only compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, Magnetic cassette tapes, magnetic tape magnetic disk storage, graphene storage or other magnetic storage devices or any other non-transmission media can be used to store information that can be accessed by computing devices. As defined in this article, computer-readable media does not include temporary computer-readable media (transitory media), such as modulated data signals and carrier waves.

Those skilled in the art should understand that one or more embodiments of this specification may be provided as a method, system, or computer program product. Therefore, one or more embodiments of this specification may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware. Moreover, one or more embodiments of this specification may employ computer programs implemented on one or more computer usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer usable program code The form of the product.

One or more embodiments of this specification may be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types. One or more embodiments of this specification can also be practiced in distributed computing environments in which tasks are performed by remote processing devices connected through a communication network. In a distributed computing environment, program modules may be located in local and remote computer storage media including storage devices.

The embodiments in this specification are described in a progressive manner. The same or similar parts between the embodiments can be referred to each other. Each embodiment focuses on the differences from other embodiments. In particular, for the system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant part can be referred to the description of the method embodiment. In the description of this specification, the description referring to the terms "one embodiment", "some embodiments", "examples", "specific examples", or "some examples" means specific features described in conjunction with the embodiment or examples , Structure, material or characteristic is included in at least one embodiment or example of this specification. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples. In addition, without contradicting each other, those skilled in the art may combine and combine different embodiments or examples and features of the different embodiments or examples described in this specification.

The above is only an embodiment of one or more embodiments of this specification, and is not intended to limit one or more embodiments of this specification. For those skilled in the art, various modifications and changes can be made to one or more embodiments of this specification. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of this specification shall be included in the scope of the claims.

Claims

A breast mass image recognition method, characterized in that it includes:

Obtain the breast magnetic resonance image to be identified;

Input the breast magnetic resonance image to be identified into the constructed breast mass identification model to obtain the mass identification result in the breast magnetic resonance image to be identified;

Wherein, the mass recognition model uses a deep full convolutional neural network model, and the encoding process of the deep full convolutional neural network model uses the basic convolution module in the U-shaped convolutional neural network model for feature extraction. The decoding process of the convolutional neural network model uses dense connections to unify the feature maps to be fused and then fused.
The method according to claim 1, characterized in that the basic convolution module in the U-shaped convolutional neural network model is used for feature extraction using the following formula:

y 1 =δ(W 13 *δ(W 12 *δ(W 11 *x 1 +b 11 )+b 12 )+b 13 )

In the above formula, y 1 represents the extracted feature map, x 1 represents the input breast magnetic resonance image, δ represents the first activation function, W 11 , W 12 , W 13 represent the corresponding weights of different convolutional layers, b 11 , b 12 and b 13 represent offset parameters corresponding to different convolutional layers.
The method of claim 1, wherein the method further comprises: using the deep fully convolutional neural network model to adjust the extracted feature map using the following formula:

y 2 =δ((W 22 *δ(W 21 *x 2 +b 21 )+b 22 )+x 2 )

In the above formula, y 2 represents the adjusted feature map, x 2 represents the extracted feature map, δ represents the first activation function, W 21 and W 22 represent the weights corresponding to different convolutional layers, and b 21 and b 22 represent different The offset parameter corresponding to the convolutional layer.
The method of claim 1, wherein the method further comprises:

Using the deep fully convolutional neural network model to assign different weight values to the extracted feature map;

Feature fusion is performed according to the weight value corresponding to each feature map.
The method according to claim 4, wherein the assigning the extracted feature maps with different weight values includes:

Perform average pooling on the extracted feature maps;

The averaged pooled feature map is connected using a fully connected layer, and is activated using a second activation function to obtain a weight value corresponding to the feature map.
The method according to claim 5, wherein the assigning the extracted feature maps with different weight values includes: using the following formula to obtain the weight values corresponding to the feature maps:

In the above formula, Z c represents the pooling result corresponding to the c-th feature map averaged by the feature map, H represents the height of the c-th feature map, M represents the width of the c-th feature map, and x c represents the input c-th feature map Feature maps, S represents the weight value vector corresponding to the feature map, Z represents the pooled result vector after the feature map is pooled, σ represents the second activation function, the weights corresponding to different fully connected layers of W 32 and W 31 , b 31 and b 32 represent offset parameters corresponding to different fully connected layers.
The method according to claim 1, wherein the breast mass identification model is constructed using the following method:

Acquire multiple sample data, the sample data includes: a breast magnetic resonance image and a tumor marker in the breast magnetic resonance image;

Establish the breast mass identification model,

The breast magnetic resonance image in the sample data is used as the input data of the breast mass identification model, and the corresponding tumor marker in the breast magnetic resonance image is used as the output data of the breast mass identification model. The lump identification model is trained until the breast lump identification model meets the preset requirements.
The method of claim 7, wherein the method further comprises:

Comparing the image parameters of the breast magnetic resonance image to be identified with the breast magnetic resonance image in the sample data;

If the image parameters of the breast magnetic resonance image to be identified and the breast magnetic resonance image in the sample data are inconsistent, use the breast magnetic resonance image to be identified to adjust the breast mass recognition model;

Using the adjusted breast mass identification model, mass identification of the breast magnetic resonance image to be identified is performed.
A breast mass image recognition device, characterized in that it includes:

The image acquisition module is used to acquire the breast magnetic resonance image to be identified;

The lump identification module is used to input the breast magnetic resonance image to be identified into the constructed breast mass identification model to obtain the mass recognition result in the breast magnetic resonance image to be identified;

Wherein, the mass recognition model uses a deep full convolutional neural network model, and the encoding process of the deep full convolutional neural network model uses the basic convolution module in the U-shaped convolutional neural network model for feature extraction. The decoding process of the convolutional neural network model uses dense connections to unify the feature maps to be fused and then fused.
The device according to claim 9, wherein the basic convolution module in the U-shaped convolutional neural network model uses the following formula for feature extraction:

y 1 =δ(W 13 *δ(W 12 *δ(W 11 *x 1 +b 11 )+b 12 )+b 13 )

In the above formula, y 1 represents the extracted feature map, x 1 represents the input breast magnetic resonance image, δ represents the first activation function, W 11 , W 12 , W 13 represent the corresponding weights of different convolutional layers, b 11 , b 12 and b 13 represent offset parameters corresponding to different convolutional layers.
The device according to claim 9, wherein the deep fully convolutional neural network model includes a feature adaptation module for adjusting the extracted feature map using the following formula:

y 2 =δ((W 22 *δ(W 21 *x 2 +b 21 )+b 22 )+x 2 )

In the above formula, y 2 represents the adjusted feature map, x 2 represents the extracted feature map, δ represents the first activation function, W 21 and W 22 represent the weights corresponding to different convolutional layers, and b 21 and b 22 represent different The offset parameter corresponding to the convolutional layer.
The apparatus of claim 9, wherein the deep fully convolutional neural network model further includes a channel attention module for:

Using the deep fully convolutional neural network model to assign different weight values to the extracted feature map;

Feature fusion is performed according to the weight value corresponding to each feature map.
The device of claim 12, wherein the channel attention module is specifically used to:

Perform average pooling on the extracted feature maps;

The averaged pooled feature map is connected using a fully connected layer, and is activated using a second activation function to obtain a weight value corresponding to the feature map.
The apparatus according to claim 13, wherein the channel attention module is specifically configured to obtain the weight value corresponding to the feature map using the following formula:

In the above formula, Z c represents the pooling result corresponding to the c-th feature map averaged by the feature map, H represents the height of the c-th feature map, M represents the width of the c-th feature map, and x c represents the input c-th feature map Feature maps, S represents the weight value vector corresponding to the feature map, Z represents the pooled result vector after the feature map is pooled, σ represents the second activation function, the weights corresponding to different fully connected layers of W 32 and W 31 , b 31 and b 32 represent offset parameters corresponding to different fully connected layers.
The device according to claim 9, characterized in that the device further comprises a model building module for building the breast mass identification model by the following method:

Acquire multiple sample data, the sample data includes: a breast magnetic resonance image and a tumor marker in the breast magnetic resonance image;

Establish the breast mass identification model, use the breast magnetic resonance image in the sample data as the input data of the breast mass identification model, and use the corresponding mass marker in the breast magnetic resonance image as the breast mass identification model Output data, training the breast mass identification model until the breast mass identification model meets the preset requirements.
The apparatus of claim 15, wherein the apparatus further comprises a model adjustment module for:

Comparing the image parameters of the breast magnetic resonance image to be identified with the breast magnetic resonance image in the sample data;

If the image parameters of the breast magnetic resonance image to be identified and the breast magnetic resonance image in the sample data are inconsistent, use the breast magnetic resonance image to be identified to adjust the breast mass recognition model;

Using the adjusted breast mass identification model, mass identification of the breast magnetic resonance image to be identified is performed.
A breast mass image recognition processing device, characterized in that it includes: at least one processor and a memory for storing processor executable instructions, and the processor implements the instructions to implement any of claims 1-8 The method described.
A breast mass image recognition system, characterized in that it includes: at least one processor and a memory for storing processor executable instructions, and the processor implements the instructions to implement any one of claims 1-8 Methods.