CN118644485B

CN118644485B - Breast cancer axillary lymph node metastasis state analysis system based on ultrasonic radiography image

Info

Publication number: CN118644485B
Application number: CN202411116902.1A
Authority: CN
Inventors: 王晓珍
Original assignee: First Hospital Jinlin University
Current assignee: First Hospital Jinlin University
Priority date: 2024-08-15
Filing date: 2024-08-15
Publication date: 2024-10-18
Anticipated expiration: 2044-08-15
Also published as: CN118644485A

Abstract

The application discloses a breast cancer axillary lymph node metastasis state analysis system based on an ultrasonic radiography image, which relates to the field of intelligent analysis, wherein the breast cancer ultrasonic radiography image is acquired, and an image processing and analysis algorithm based on artificial intelligence and deep learning is introduced at the rear end to analyze the breast cancer ultrasonic radiography image so as to learn and capture CEUS characteristics therein, and breast cancer axillary lymph node metastasis state analysis and judgment are performed based on the CEUS characteristics, so that whether breast cancer axillary lymph node metastasis occurs is detected. Therefore, the method can automatically learn and capture the slight changes of blood flow distribution, microvascular perfusion condition and the like of the axillary lymph nodes by utilizing the ultrasonic contrast image characteristics, thereby realizing more intelligent analysis of the breast cancer axillary lymph node metastasis state, and assisting in diagnosing breast axillary lymph node metastasis so as to better guide clinical diagnosis and decision.

Description

Breast cancer axillary lymph node metastasis state analysis system based on ultrasonic radiography image

Technical Field

The application relates to the field of intelligent analysis, and in particular relates to a breast cancer axillary lymph node metastasis state analysis system based on ultrasonic contrast images.

Background

Breast cancer is one of the most common malignant tumors in women, and axillary lymph node metastasis is an important prognostic factor for breast cancer, with an important impact on patient survival and treatment regimen selection. Early diagnosis of breast cancer axillary lymph node metastasis status is helpful for making more accurate treatment plan.

However, traditional imaging examinations such as molybdenum targets, ultrasound, CT and conventional MRI scans, etc., focus mainly on the anatomical morphology level of lymph nodes, such as the short diameter of lymph nodes, but these morphological changes may lag behind the histological changes, resulting in failure to detect early metastases. In addition, although fine needle penetration under ultrasound guidance can increase the preoperative diagnostic rate, this approach is invasive and there is still a high false negative rate in combination with clinical studies and various imaging examinations. Moreover, the traditional breast cancer axillary lymph node metastasis state analysis mode usually only depends on the experience of doctors and a small amount of image features to analyze, detailed information about tissue microvascular perfusion and cell density cannot be provided, and the mode is greatly influenced by subjective judgment and experience of the doctors, is easy to generate individual difference and lacks objectivity and standardization.

Thus, an optimized breast cancer axillary lymph node metastasis status analysis system is desired.

Disclosure of Invention

The present application has been made to solve the above-mentioned technical problems. The application provides a breast cancer axillary lymph node metastasis state analysis system based on an ultrasonic radiography image.

According to one aspect of the present application, there is provided a breast cancer axillary lymph node metastasis status analysis system based on ultrasound contrast images, comprising:

The ultrasonic contrast image acquisition module is used for acquiring an ultrasonic contrast image of the breast cancer;

the ultrasonic contrast image statistical feature extraction module is used for extracting HOG features and LBP features of the breast cancer ultrasonic contrast image to obtain a breast cancer ultrasonic contrast HOG feature vector and a breast cancer ultrasonic contrast LBP feature vector;

The breast cancer ultrasound contrast multi-mode statistical feature fusion module is used for inputting the breast cancer ultrasound contrast LBP feature vector and the breast cancer ultrasound contrast HOG feature vector into the feature vector dynamic interaction fusion module based on gating response to obtain a breast cancer ultrasound contrast multi-mode gating interaction fusion vector;

The ultrasonic contrast image implicit characteristic extraction optimization module is used for extracting image characteristics of the breast cancer ultrasonic contrast image and inputting the image characteristics into the compression-inhibition optimization module selected based on the characteristic attention so as to obtain an optimized breast cancer ultrasonic contrast characteristic image;

The cross-domain joint coding module is used for inputting the optimized breast cancer ultrasonic imaging feature map and the breast cancer ultrasonic imaging multi-mode gating interactive fusion vector into a cross-domain joint coder based on a meta-network so as to obtain a breast cancer ultrasonic imaging modulation feature map under the constraint of statistical features;

And the transfer state analysis module is used for determining a transfer state analysis result based on the ultrasonic contrast modulation characteristic diagram of the breast cancer under the statistical characteristic constraint, wherein the transfer state analysis result is used for indicating whether the probability of transfer exceeds a preset threshold value.

In the above system for analyzing the metastasis state of the axillary lymph node of breast cancer based on the ultrasound contrast image, the statistical feature extraction module of the ultrasound contrast image is configured to:

Extracting the HOG characteristic of the breast cancer ultrasonic radiography image based on the directional gradient histogram to obtain the HOG characteristic vector of the breast cancer ultrasonic radiography;

And extracting LBP characteristics of the breast cancer ultrasonic radiography image based on an LBP mode operator to obtain the breast cancer ultrasonic radiography LBP characteristic vector.

In the above system for analyzing the metastasis state of the axillary lymph node of breast cancer based on ultrasound contrast images, the module for fusion of the statistical characteristics of the ultrasound contrast of the breast cancer comprises:

The feature combination unit is used for inputting the breast cancer ultrasound radiography LBP feature vector and the breast cancer ultrasound radiography HOG feature vector into a feature combination module for cascade processing so as to obtain a breast cancer ultrasound radiography LBP feature-HOG feature combination feature vector;

The linear transformation unit is used for calculating the matrix multiplication of the LBP characteristic-HOG characteristic combined characteristic vector and the parameter matrix of the breast cancer ultrasonic radiography, and then carrying out position-based addition on the obtained characteristic vector and the offset vector to obtain the linear transformation breast cancer ultrasonic radiography LBP characteristic-HOG characteristic combined characteristic vector;

an activation unit for using Activating the linear transformation breast cancer ultrasound radiography LBP characteristic-HOG characteristic combined characteristic vector by a function to obtain a breast cancer ultrasound radiography LBP characteristic-HOG characteristic dynamic information fusion response gating value;

The LBP characteristic weight modulation unit is used for carrying out the weight modulation on the breast cancer ultrasonic radiography LBP characteristic weight, calculating the position-based product between the breast cancer ultrasonic radiography LBP characteristic vector and the breast cancer ultrasonic radiography LBP characteristic-HOG characteristic dynamic information fusion response gating value, and obtaining a weight modulation breast cancer ultrasonic radiography LBP characteristic vector;

The HOG feature weight modulation unit is used for calculating a subtracted dynamic information fusion response gating value of the breast cancer ultrasound imaging LBP feature-HOG feature, and multiplying the obtained weight value with the breast cancer ultrasound imaging HOG feature vector according to the position to obtain a weight-modulated breast cancer ultrasound imaging HOG feature vector;

the multi-mode gating interaction fusion unit is used for obtaining the breast cancer ultrasound contrast multi-mode gating interaction fusion vector by carrying out position point-based on the weight modulation breast cancer ultrasound contrast LBP feature vector and the weight modulation breast cancer ultrasound contrast HOG feature vector.

In the above system for analyzing the metastasis state of the axillary lymph node of breast cancer based on the ultrasound contrast image, the implicit feature extraction optimization module of the ultrasound contrast image comprises:

the breast cancer ultrasound radiography implicit characteristic extraction unit is used for inputting the breast cancer ultrasound radiography image into an image characteristic extractor based on AlexNet model to obtain a breast cancer ultrasound radiography characteristic diagram;

And the ultrasonic contrast characteristic attention selection optimizing unit is used for inputting the breast cancer ultrasonic contrast characteristic image into a compression-inhibition optimizing module based on characteristic attention selection to obtain the optimized breast cancer ultrasonic contrast characteristic image.

In the above system for analyzing the metastasis state of the axillary lymph node of breast cancer based on ultrasound contrast images, the ultrasound contrast feature attention selection optimizing unit includes:

The breast cancer ultrasound contrast characteristic compression information representation subunit is used for calculating the global average value of each characteristic matrix of the breast cancer ultrasound contrast characteristic map along the channel dimension to obtain a breast cancer ultrasound contrast characteristic compression information representation vector;

The convolution coding subunit is used for carrying out one-dimensional convolution coding on the breast cancer ultrasonic contrast characteristic compressed information representation vector so as to obtain a correlation representation characteristic vector among breast cancer ultrasonic contrast compressed information;

The breast cancer ultrasound contrast compressed information multi-scale representation subunit is used for cascading the breast cancer ultrasound contrast characteristic compressed information representation vector and the breast cancer ultrasound contrast compressed information association representation characteristic vector to obtain a breast cancer ultrasound contrast compressed information multi-scale representation vector;

The compressed information feature extraction subunit is used for inputting the compressed information multiscale representation vector of the breast cancer ultrasound contrast imaging to a compressed information feature extraction module to obtain a compressed information multiscale associated feature vector of the breast cancer ultrasound contrast imaging, wherein the compressed information feature extraction module is a multi-layer perceptron comprising two full-connection layers and SiLU activation functions;

The normalization subunit is used for performing normalization operation on the breast cancer ultrasound contrast compression information multi-scale associated feature vector by using a Sigmoid function so as to obtain a breast cancer ultrasound contrast weight feature vector;

and the optimization subunit is used for carrying out characteristic amplification and inhibition operation on the breast cancer ultrasonic radiography characteristic map based on the breast cancer ultrasonic radiography weight characteristic vector so as to obtain the optimized breast cancer ultrasonic radiography characteristic map.

In the above system for analyzing the metastasis state of the axillary lymph node of breast cancer based on ultrasound contrast images, the compressed information feature extraction subunit is configured to:

Taking the negative numbers of the characteristic values of each position in the breast cancer ultrasound contrast compression information multi-scale associated characteristic vector as indexes of natural constants to calculate index function values based on the natural constants according to the positions so as to obtain the breast cancer ultrasound contrast compression information multi-scale associated support characteristic vector;

and calculating the inverse of the sum of the characteristic value of each position in the breast cancer ultrasonic contrast compression information multi-scale association support characteristic vector and a constant one to obtain the breast cancer ultrasonic contrast weight characteristic vector.

In the above system for analyzing the metastasis state of the axillary lymph node of breast cancer based on ultrasonic radiography images, the optimizing subunit is configured to: and multiplying the feature value of each position in the breast cancer ultrasonic radiography weight feature vector with each feature matrix of the breast cancer ultrasonic radiography feature map along the channel dimension according to the position to obtain the optimized breast cancer ultrasonic radiography feature map.

In the above breast cancer axillary lymph node metastasis state analysis system based on ultrasound contrast image, the metastasis state analysis module is configured to: inputting the ultrasonic contrast modulation characteristic diagram of the breast cancer under the constraint of the statistical characteristic into a transfer state analysis result generator based on a classifier to obtain a transfer state analysis result, wherein the transfer state analysis result is used for indicating whether the probability of occurrence of transfer exceeds a preset threshold value.

The breast cancer axillary lymph node metastasis state analysis system based on the ultrasonic contrast image further comprises a feature vector dynamic interaction fusion module based on the gating response, an image feature extractor based on the AlexNet model, a compression-suppression optimization module based on feature attention selection, a cross-domain joint encoder based on a meta-network and a training module for training a metastasis state analysis result generator based on a classifier.

In the above breast cancer axillary lymph node metastasis state analysis system based on ultrasound contrast image, the training module includes:

the training data acquisition unit is used for acquiring training data, wherein the training data comprises training breast cancer ultrasonic contrast images;

The training image statistical feature extraction unit is used for extracting HOG features and LBP features of the training breast cancer ultrasonic radiography images to obtain training breast cancer ultrasonic radiography HOG feature vectors and training breast cancer ultrasonic radiography LBP feature vectors;

The training multi-mode statistical feature fusion unit is used for inputting the training breast cancer ultrasound contrast LBP feature vector and the training breast cancer ultrasound contrast HOG feature vector into a feature vector dynamic interaction fusion module based on gating response to obtain a training breast cancer ultrasound contrast multi-mode gating interaction fusion vector;

The training image implicit characteristic extraction optimization unit is used for extracting image characteristics of the training breast cancer ultrasonic contrast image and inputting the image characteristics into the compression-inhibition optimization module selected based on the characteristic attention so as to obtain an optimized training breast cancer ultrasonic contrast characteristic image;

The training multi-mode feature cross-domain joint coding unit is used for inputting the optimized training breast cancer ultrasonic radiography feature map and the training breast cancer ultrasonic radiography multi-mode gating interaction fusion vector into a cross-domain joint coder based on a meta-network so as to obtain a breast cancer ultrasonic radiography modulation feature map under the constraint of training statistical features;

the classification loss unit is used for enabling the breast cancer ultrasonic contrast modulation characteristic diagram under the training statistical characteristic constraint to pass through a transfer state analysis result generator based on a classifier so as to obtain a classification loss function value;

The loss calculation unit is used for calculating the ultrasonic contrast modulation loss function value of the breast cancer under the statistical characteristic constraint based on the ultrasonic contrast modulation characteristic loss item of the breast cancer under the statistical characteristic constraint;

The weighting unit is used for calculating a weighted sum of the classified loss function value and the breast cancer ultrasonic contrast modulation loss function value under the constraint of the statistical characteristics to obtain a final loss function value;

The training unit is used for training the feature vector dynamic interaction fusion module based on the gating response, the image feature extractor based on the AlexNet model, the compression-suppression optimization module based on feature attention selection, the cross-domain joint encoder based on the meta-network and the transfer state analysis result generator based on the classifier based on the final loss function value.

Compared with the prior art, the breast cancer axillary lymph node metastasis state analysis system based on the ultrasonic radiography image provided by the application is characterized in that the breast cancer ultrasound radiography image is acquired, and an image processing and analysis algorithm based on artificial intelligence and deep learning is introduced at the rear end to analyze the breast cancer ultrasound radiography image, so that the CEUS characteristics are learned and captured, and the breast cancer axillary lymph node metastasis state analysis and judgment are carried out based on the CEUS characteristics, so that whether the breast cancer axillary lymph node metastasis occurs is detected. Therefore, the method can automatically learn and capture the slight changes of blood flow distribution, microvascular perfusion condition and the like of the axillary lymph nodes by utilizing the ultrasonic contrast image characteristics, thereby realizing more intelligent analysis of the breast cancer axillary lymph node metastasis state, and assisting in diagnosing breast axillary lymph node metastasis so as to better guide clinical diagnosis and decision.

Drawings

The above and other objects, features and advantages of the present application will become more apparent by describing embodiments of the present application in more detail with reference to the attached drawings. The accompanying drawings are included to provide a further understanding of embodiments of the application and are incorporated in and constitute a part of this specification, illustrate the application and together with the embodiments of the application, and not constitute a limitation to the application. In the drawings, like reference numerals generally refer to like parts or steps.

FIG. 1 is a block diagram of a breast cancer axillary lymph node metastasis status analysis system based on ultrasound contrast images in accordance with an embodiment of the present application;

FIG. 2 is a data flow diagram of an ultrasound contrast image based breast cancer axillary lymph node metastasis status analysis system according to an embodiment of the present application;

FIG. 3 is a block diagram of a training module of a breast cancer axillary lymph node metastasis status analysis system based on ultrasound contrast images in accordance with an embodiment of the present application;

fig. 4 is a block diagram of an ultrasound contrast image implicit feature extraction optimization module in a breast cancer axillary lymph node metastasis status analysis system based on ultrasound contrast images according to an embodiment of the present application.

Detailed Description

Hereinafter, exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. It should be apparent that the described embodiments are only some embodiments of the present application and not all embodiments of the present application, and it should be understood that the present application is not limited by the example embodiments described herein.

As used in the specification and in the claims, the terms "a," "an," "the," and/or "the" are not specific to a singular, but may include a plurality, unless the context clearly dictates otherwise. In general, the terms "comprises" and "comprising" merely indicate that the steps and elements are explicitly identified, and they do not constitute an exclusive list, as other steps or elements may be included in a method or apparatus.

Although the present application makes various references to certain modules in a system according to embodiments of the present application, any number of different modules may be used and run on a user terminal and/or server. The modules are merely illustrative, and different aspects of the systems and methods may use different modules.

A flowchart is used in the present application to describe the operations performed by a system according to embodiments of the present application. It should be understood that the preceding or following operations are not necessarily performed in order precisely. Rather, the various steps may be processed in reverse order or simultaneously, as desired. Also, other operations may be added to or removed from these processes.

Traditional imaging examinations, such as molybdenum targets, ultrasound, CT and conventional MRI scans, focus mainly on the anatomical aspects of lymph nodes, such as the minor diameter of lymph nodes, but these morphological changes may lag behind the histological changes, resulting in failure to detect metastasis early. In addition, although fine needle penetration under ultrasound guidance can increase the preoperative diagnostic rate, this approach is invasive and there is still a high false negative rate in combination with clinical studies and various imaging examinations. Moreover, the traditional breast cancer axillary lymph node metastasis state analysis mode usually only depends on the experience of doctors and a small amount of image features to analyze, detailed information about tissue microvascular perfusion and cell density cannot be provided, and the mode is greatly influenced by subjective judgment and experience of the doctors, is easy to generate individual difference and lacks objectivity and standardization. In view of the above technical problems, in the technical solution of the present application, an analysis system for the metastasis state of axillary lymph nodes of breast cancer based on ultrasound contrast images is provided, which can diagnose and evaluate whether the axillary lymph nodes of breast cancer patients metastasize by using ultrasound contrast technology (CEUS).

It should be understood that the ultrasound imaging technique is a non-invasive examination method that uses microbubbles as a contrast agent, and observes the distribution and dynamic changes of microbubbles in tissue by the ultrasound imaging technique, thereby obtaining information such as tissue blood flow perfusion and cell density. The principle is that microbubbles vibrate in an ultrasonic field to generate echo enhancement, so that the microvascular structure and the hemodynamics are more clearly visible, and the diagnosis and the state analysis of breast cancer axillary lymph node metastasis are assisted. CEUS is able to more sensitively detect minor changes in axillary lymph nodes, such as blood flow distribution and microvascular perfusion, than conventional ultrasound, thereby aiding in the diagnosis of breast axillary lymph node metastasis. By observing the perfusion pattern, degree of enhancement, and time characteristics of microbubbles in the lymph nodes, the clinician can be provided with more clues about the metastatic status of the lymph nodes.

In particular, CEUS characteristics are closely related to breast cancer axillary lymph node metastasis, and a number of parameters including peripheral areola, blood flow classification, perfusion order, etc. are all related to metastasis. In addition, in combination with ultrasonic multiparameter evaluation, prediction of breast cancer axillary lymph node metastasis can be better aided. For example, parameters such as maximum diameter of lesions, lesion location, enhancement of anterior fat echo, etc. have been shown to be independent risk factors for predicting breast cancer axillary lymph node metastasis.

In the technical scheme of the application, a breast cancer axillary lymph node metastasis state analysis system based on ultrasonic radiography images is provided. Fig. 1 is a block diagram of a breast cancer axillary lymph node metastasis status analysis system based on ultrasound contrast images in accordance with an embodiment of the present application. Fig. 2 is a data flow diagram of an ultrasound contrast image based breast cancer axillary lymph node metastasis status analysis system according to an embodiment of the present application. As shown in fig. 1 and 2, a breast cancer axillary lymph node metastasis status analysis system based on an ultrasound contrast image according to an embodiment of the present application includes: an ultrasound contrast image acquisition module 310 for acquiring ultrasound contrast images of breast cancer; the ultrasound contrast image statistical feature extraction module 320 is configured to extract HOG features and LBP features of the ultrasound contrast image of breast cancer to obtain an ultrasound contrast HOG feature vector of breast cancer and an ultrasound contrast LBP feature vector of breast cancer; the breast cancer ultrasound contrast multi-mode statistical feature fusion module 330 is configured to input the breast cancer ultrasound contrast LBP feature vector and the breast cancer ultrasound contrast HOG feature vector into a feature vector dynamic interaction fusion module based on gating response to obtain a breast cancer ultrasound contrast multi-mode gating interaction fusion vector; the ultrasound contrast image implicit feature extraction optimizing module 340 is configured to perform image feature extraction on the breast cancer ultrasound contrast image, and input the image feature extraction to the compression-suppression optimizing module based on feature attention selection to obtain an optimized breast cancer ultrasound contrast feature map; the cross-domain joint coding module 350 for the multi-modal feature of the breast cancer ultrasound imaging is used for inputting the optimized breast cancer ultrasound imaging feature map and the cross-domain joint coder based on the meta-network to obtain a breast cancer ultrasound imaging modulation feature map under the constraint of statistical features; the transition state analysis module 360 is configured to determine a transition state analysis result based on the ultrasound contrast modulation feature map of breast cancer under the statistical feature constraint, where the transition state analysis result is used to indicate whether the probability of occurrence of transition exceeds a preset threshold.

In particular, the ultrasound contrast image acquisition module 310 and the ultrasound contrast image statistics feature extraction module 320 are configured to acquire an ultrasound contrast image of breast cancer; and extracting HOG features and LBP features of the breast cancer ultrasonic radiography images to obtain breast cancer ultrasonic radiography HOG feature vectors and breast cancer ultrasonic radiography LBP feature vectors. It will be appreciated that texture and statistical features in the image play an important role in the analysis of breast cancer axillary lymph node metastasis status, and can help identify and analyze the blood flow patterns of tumor areas, which is important for distinguishing benign and malignant lymph nodes and for performing breast axillary lymph node metastasis status analysis. Based on the above, in the technical scheme of the application, HOG features of the breast cancer ultrasound contrast image are extracted based on the directional gradient histogram to obtain a breast cancer ultrasound contrast HOG feature vector; and extracting LBP characteristics of the breast cancer ultrasonic radiography image based on an LBP mode operator to obtain a breast cancer ultrasonic radiography LBP characteristic vector. It should be appreciated that HOG is an algorithm that describes local features of an image, by counting the direction and intensity of pixel gradients in the image, to generate feature descriptors. The HOG features can capture gradient information in different directions in the image, are helpful for distinguishing texture and shape features of different areas, and have better expression capability for microstructure in breast cancer ultrasound contrast images. In addition, HOG features can effectively describe edge information in images, and play an important role in describing structural features and texture information in breast cancer ultrasound contrast images, which helps to distinguish between tumor and normal tissue. LBP is a simple and efficient texture analysis method that generates features by comparing the relative intensities of pixels and their surrounding neighbors. The LBP can capture the contrast and texture information of local areas, is helpful for identifying the micro-blood vessels and cell structure changes in the image, and has better depicting ability for tissue structures and texture features in breast cancer ultrasound contrast images. In addition, LBP features have rotational invariance, can identify texture patterns in an image, and remain stable even in the event of image rotation or deformation.

Specifically, the breast cancer ultrasound imaging multi-mode statistical feature fusion module 330 is configured to input the breast cancer ultrasound imaging LBP feature vector and the breast cancer ultrasound imaging HOG feature vector into a feature vector dynamic interaction fusion module based on gating response to obtain a breast cancer ultrasound imaging multi-mode gating interaction fusion vector. Considering that the breast cancer ultrasound imaging LBP feature vector and the breast cancer ultrasound imaging HOG feature vector respectively contain gradient directions, intensities and local binary patterns in the breast cancer ultrasound imaging image, the statistical feature information reflects local texture and edge feature information in the image, and the information carried by the two different types of image statistical features can be comprehensively utilized by fusing the two different types of image statistical features, so that the expression capability of the image features and the analysis performance of the subsequent transfer states are improved. Based on the above, in the technical scheme of the application, the breast cancer ultrasound imaging LBP feature vector and the breast cancer ultrasound imaging HOG feature vector are further input into a feature vector dynamic interaction fusion module based on gating response to obtain a breast cancer ultrasound imaging multi-mode gating interaction fusion vector. Particularly, through the processing of the characteristic vector dynamic interaction fusion module based on the gating response, the correlation relation and interaction influence between the breast cancer ultrasound imaging LBP characteristic vector and the breast cancer ultrasound imaging HOG characteristic vector can be learned and captured, so that the implicit correlation semantics between the two ultrasound imaging image statistical characteristic information are utilized to carry out interaction supplement and fusion, the contribution degree of different types of image statistical characteristics to the subsequent transfer state analysis task is identified, the fusion degree of the characteristics is dynamically adjusted according to the contribution degree of different characteristics, the characteristics of different breast cancer ultrasound imaging characteristics are more flexibly adapted, the characteristic information of the breast cancer ultrasound imaging is better expressed, and the judgment on whether the breast cancer axillary lymph node is transferred is facilitated.

In an embodiment of the present application, inputting the breast cancer ultrasound contrast LBP feature vector and the breast cancer ultrasound contrast HOG feature vector into a feature vector dynamic interaction fusion module based on gating response to obtain a breast cancer ultrasound contrast multi-mode gating interaction fusion vector, including: inputting the breast cancer ultrasound radiography LBP feature vector and the breast cancer ultrasound radiography HOG feature vector into a feature combination module for cascade processing to obtain a breast cancer ultrasound radiography LBP feature-HOG feature combination feature vector; after calculating the matrix multiplication of the LBP characteristic-HOG characteristic combined characteristic vector and the parameter matrix of the breast cancer ultrasonic radiography, the obtained characteristic vector and the offset vector are added according to positions to obtain the linear transformation breast cancer ultrasonic radiography LBP characteristic-HOG characteristic combined characteristic vector; usingActivating the linear transformation breast cancer ultrasound radiography LBP characteristic-HOG characteristic combined characteristic vector by a function to obtain a breast cancer ultrasound radiography LBP characteristic-HOG characteristic dynamic information fusion response gating value; the breast cancer ultrasound radiography LBP characteristic weight modulation calculates the position-based product between the breast cancer ultrasound radiography LBP characteristic vector and the breast cancer ultrasound radiography LBP characteristic-HOG characteristic dynamic information fusion response gating value to obtain a weight modulation breast cancer ultrasound radiography LBP characteristic vector; calculating a response gating value of subtracting the dynamic information fusion response of the breast cancer ultrasonic imaging LBP characteristic-HOG characteristic, and multiplying the obtained weight value with the breast cancer ultrasonic imaging HOG characteristic vector according to the position to obtain a weight-modulated breast cancer ultrasonic imaging HOG characteristic vector; and carrying out position point-based on the weight-modulated breast cancer ultrasound radiography LBP feature vector and the weight-modulated breast cancer ultrasound radiography HOG feature vector to obtain the breast cancer ultrasound radiography multi-mode gating interaction fusion vector.

To sum up, in the above embodiment, inputting the breast cancer ultrasound contrast LBP feature vector and the breast cancer ultrasound contrast HOG feature vector into a feature vector dynamic interaction fusion module based on gating response to obtain a breast cancer ultrasound contrast multi-mode gating interaction fusion vector, including: inputting the breast cancer ultrasound radiography LBP feature vector and the breast cancer ultrasound radiography HOG feature vector into the feature vector dynamic interaction fusion module based on gating response, and processing the feature vector dynamic interaction fusion module according to the following dynamic interaction fusion formula to obtain the breast cancer ultrasound radiography multi-mode gating interaction fusion vector; the dynamic interaction fusion formula is as follows:

；

Wherein, AndRespectively the breast cancer ultrasound contrast LBP characteristic vector and the breast cancer ultrasound contrast HOG characteristic vector,A vector concatenation operation is represented and is performed,Is a matrix of parameters that are selected from the group consisting of,Is the offset vector of the reference signal,Is a sigmoid function of the number of bits,Is the dynamic information fusion response gating value of the breast cancer ultrasonic radiography LBP characteristic-HOG characteristic,Is the breast cancer ultrasound contrast multi-mode gating interaction fusion vector.

In particular, the ultrasound contrast image implicit feature extraction optimizing module 340 is configured to perform image feature extraction on the ultrasound contrast image of breast cancer, and then input the image feature extraction to the compression-suppression optimizing module based on feature attention selection to obtain an optimized ultrasound contrast feature map of breast cancer. In particular, in one specific example of the present application, as shown in fig. 4, the ultrasound contrast image implicit feature extraction optimization module 340 includes: the breast cancer ultrasound radiography implicit feature extraction unit 341 is configured to input the breast cancer ultrasound radiography image into an image feature extractor based on AlexNet model to obtain a breast cancer ultrasound radiography feature map; an ultrasound contrast feature attention selection optimization unit 342 for inputting the breast cancer ultrasound contrast feature map into a compression-suppression optimization module based on feature attention selection to obtain the optimized breast cancer ultrasound contrast feature map.

Specifically, the breast cancer ultrasound imaging implicit feature extraction unit 341 is configured to input the breast cancer ultrasound imaging image into an image feature extractor based on AlexNet models to obtain a breast cancer ultrasound imaging feature map. That is, after the statistical features of the ultrasound contrast image are subjected to multi-modal fusion, in order to learn and identify the implicit feature distribution information contained in the ultrasound contrast image of the breast cancer, so as to improve the understanding and expression capability of the system on the image content, in the technical scheme of the application, the ultrasound contrast image of the breast cancer is further input into the image feature extractor based on the AlexNet model so as to obtain an ultrasound contrast feature map of the breast cancer. The image feature extractor based on AlexNet model can extract complex implicit features and deep semantics in the breast cancer ultrasound contrast image, thereby better capturing abstract information and features in the image, and being beneficial to improving the understanding and expression capability of hidden and complex feature contents such as blood flow distribution, microvascular perfusion condition and the like of the axillary lymph nodes contained in the ultrasound contrast image.

Specifically, the ultrasound contrast feature attention selection optimizing unit 342 is configured to input the ultrasound contrast feature map of breast cancer into a compression-suppression optimizing module based on feature attention selection to obtain the optimized ultrasound contrast feature map of breast cancer. It should be understood that, in the process of performing the analysis of the metastasis state of the axillary lymph node of the breast cancer, since the ultrasound imaging of the breast cancer may contain a lot of background interference noise related to other tissue structures and redundant information, in order to be able to identify the most important feature areas in the ultrasound imaging of the breast cancer, so as to optimize the feature representation, to help the model concentrate on those most valuable information for the task of the metastasis state analysis, in the technical solution of the present application, the ultrasound imaging of the breast cancer is further input to a compression-suppression optimization module selected based on the feature concentration, so as to obtain an optimized ultrasound imaging feature of the breast cancer. Specifically, the compression-suppression optimization module based on feature attention selection compresses local feature information of the breast cancer ultrasonic radiography feature map along each feature matrix of the channel dimension by utilizing global average pooling, captures association information between the compressed information by one-dimensional convolution coding, performs feature extraction by a multi-layer perceptron consisting of two full-connection layers and SiLU activation functions, normalizes the extracted features to obtain weight values of different channels, performs feature amplification and suppression operation on the breast cancer ultrasonic radiography feature map by combining the weight values, and improves the recognition and selection capability of the network on important area features of the breast cancer ultrasonic radiography feature map so as to emphasize important parts in the breast cancer ultrasonic radiography feature map. For example, the model may be aided in focusing on the status semantic features of breast cancer axillary lymph nodes in the ultrasound contrast image of breast cancer, including peripheral vignetting, blood flow classification, perfusion order, peripheral convergence, maximum cortical thickness, etc., which are critical for subsequent metastatic status analysis. In particular, siLU activation functions, compared with the original ReLU activation functions, have the characteristics of no upper bound, low bound, smoothness and non-monotonic, and can improve the deep network training effect and performance. Based on the method, redundant or unimportant image features can be effectively removed through the processing of the compression-inhibition optimization module based on feature attention selection, and only breast cancer ultrasound contrast semantic features with the most representation and discrimination are reserved, so that the characterization capability of the features is optimized, and the selective feature extraction is beneficial to reducing noise and interference of irrelevant information and is beneficial to subsequent analysis and judgment of breast cancer axillary lymph node metastasis state.

In an embodiment of the present application, inputting the breast cancer ultrasound contrast profile into a compression-suppression optimization module selected based on feature attention to obtain the optimized breast cancer ultrasound contrast profile comprises: calculating the global average value of each feature matrix of the breast cancer ultrasonic radiography feature map along the channel dimension to obtain a breast cancer ultrasonic radiography feature compression information expression vector; carrying out one-dimensional convolution coding on the breast cancer ultrasonic contrast characteristic compressed information representation vector to obtain a correlation representation characteristic vector between breast cancer ultrasonic contrast compressed information; cascading the breast cancer ultrasound contrast characteristic compressed information expression vector and the breast cancer ultrasound contrast compressed information association expression characteristic vector to obtain a breast cancer ultrasound contrast compressed information multiscale expression vector; inputting the compressed information multiscale expression vector of the breast cancer ultrasonic imaging to a compressed information feature extraction module to obtain a compressed information multiscale associated feature vector of the breast cancer ultrasonic imaging, wherein the compressed information feature extraction module is a multi-layer perceptron comprising two full-connection layers and SiLU activation functions; performing normalization operation on the breast cancer ultrasound contrast compression information multi-scale associated feature vector by using a Sigmoid function to obtain a breast cancer ultrasound contrast weight feature vector; and performing feature amplification and suppression operation on the breast cancer ultrasonic radiography feature map based on the breast cancer ultrasonic radiography weight feature vector to obtain the optimized breast cancer ultrasonic radiography feature map.

The process of inputting the breast cancer ultrasound contrast compressed information multi-scale representation vector into the compressed information feature extraction module to obtain the breast cancer ultrasound contrast compressed information multi-scale associated feature vector comprises the following steps: taking the negative numbers of the characteristic values of each position in the breast cancer ultrasound contrast compression information multi-scale associated characteristic vector as indexes of natural constants to calculate index function values based on the natural constants according to the positions so as to obtain the breast cancer ultrasound contrast compression information multi-scale associated support characteristic vector; and calculating the inverse of the sum of the characteristic value of each position in the breast cancer ultrasonic contrast compression information multi-scale association support characteristic vector and a constant one to obtain the breast cancer ultrasonic contrast weight characteristic vector. More specifically, based on the breast cancer ultrasound contrast weight feature vector, the process of performing feature amplification and suppression operations on the breast cancer ultrasound contrast feature map to obtain the optimized breast cancer ultrasound contrast feature map includes: and multiplying the feature value of each position in the breast cancer ultrasonic radiography weight feature vector with each feature matrix of the breast cancer ultrasonic radiography feature map along the channel dimension according to the position to obtain the optimized breast cancer ultrasonic radiography feature map.

It should be noted that, in other specific examples of the present application, the image feature extraction may be performed on the breast cancer ultrasound contrast image in other manners, and then the image feature extraction is input to a compression-suppression optimization module selected based on feature attention, so as to obtain an optimized breast cancer ultrasound contrast feature map, for example: inputting the breast cancer ultrasound contrast image; feature extraction is carried out on the breast cancer ultrasonic contrast image by using a deep learning model (such as a convolutional neural network); using a feature attention mechanism to select features that are most critical for breast cancer diagnosis; using a compression-suppression optimization module for further optimizing the selected features; and integrating the features subjected to feature extraction, feature selection and compression-inhibition optimization to obtain the optimized breast cancer ultrasonic contrast characteristic map.

In particular, the cross-domain joint encoding module 350 is configured to input the optimized breast cancer ultrasound contrast feature map and the cross-domain joint encoder based on a meta-network to obtain a breast cancer ultrasound contrast modulation feature map under a statistical feature constraint. It should be understood that, because the optimized breast cancer ultrasound contrast feature map and the breast cancer ultrasound contrast multi-mode gating interaction fusion vector respectively include the optimized semantic feature and the multi-mode statistical feature related to the breast cancer ultrasound contrast image, both features can reflect the state semantics related to the blood flow distribution and the microvascular perfusion condition of the axillary lymph node of the breast cancer in the breast cancer ultrasound contrast image, which is beneficial to the analysis of the metastatic state. Based on the above, in order to understand the semantics contained in the breast cancer ultrasound contrast image more deeply and accurately so as to improve the breast cancer ultrasound contrast semantics and the expression capability of the breast cancer axillary lymph node metastasis state, in the technical scheme of the application, the optimized breast cancer ultrasound contrast feature map and the breast cancer ultrasound contrast multi-mode gating interaction fusion vector are further input into a cross-domain joint encoder based on a meta-network so as to obtain a breast cancer ultrasound contrast modulation feature map under the constraint of statistical features. Through the processing of the cross-domain joint encoder based on the meta-network, shared and interactive feature representations can be learned among image features of different modes, and effective fusion among the state features of different breast cancer axillary lymph nodes can be realized. Specifically, the multi-mode fusion characteristic of the breast cancer ultrasonic contrast image is utilized for carrying out channel weighting so as to assist in optimizing the expression of the breast cancer ultrasonic contrast semantic characteristic, and the understanding capability of the breast cancer ultrasonic contrast image and the analysis capability of the breast cancer axillary lymph node metastasis state are improved.

Specifically, inputting the optimized breast cancer ultrasound contrast characteristic map and the breast cancer ultrasound contrast multi-mode gating interaction fusion vector into a cross-domain joint encoder based on a meta-network to obtain a breast cancer ultrasound contrast modulation characteristic map under statistical characteristic constraint, wherein the method comprises the following steps of: inputting the breast cancer ultrasound contrast characteristic map into the compression-inhibition optimization module based on characteristic attention selection, and processing the compression-inhibition optimization module by using the following characteristic attention selection strengthening formula to obtain the optimized breast cancer ultrasound contrast characteristic map; wherein, the characteristic attention selection strengthening formula is:

；

Wherein, Is the ultrasonic contrast characteristic map of the breast cancer,Ultrasound contrast characteristic diagram for representing breast cancerIs the first of (2)Coordinates in the channel areThe characteristic value of the point is set to be,AndThe height and width of the breast cancer ultrasound contrast characteristic map are respectively,Is the breast cancer ultrasound contrast characteristic compressed information representing vector,Is a one-dimensional convolutional code,Is to perform cascade processing on the vector,Is the breast cancer ultrasound contrast compressed information multi-scale representation vector,Is a multi-layer sensing machine, which comprises a main body,Representing the multiplication by location,Is the optimized breast cancer ultrasound contrast characteristic map.

In particular, the transition state analysis module 360 is configured to determine a transition state analysis result based on the ultrasound contrast modulation profile of breast cancer under the statistical feature constraint, where the transition state analysis result is used to indicate whether the probability of occurrence of transition exceeds a preset threshold. In the technical scheme of the application, the ultrasonic contrast modulation characteristic diagram of the breast cancer under the constraint of the statistical characteristic is input into a transfer state analysis result generator based on a classifier to obtain a transfer state analysis result, and the transfer state analysis result is used for indicating whether the probability of occurrence of transfer exceeds a preset threshold value. That is, breast cancer axillary lymph node metastasis state analysis and judgment are performed by using the breast cancer ultrasound radiography semantics under the constraint of the multi-mode statistical fusion characteristics of the breast cancer ultrasound radiography images, so that whether the breast cancer axillary lymph node metastasis occurs is detected. Therefore, the method can automatically learn and capture the slight changes of blood flow distribution, microvascular perfusion condition and the like of the axillary lymph nodes by utilizing the ultrasonic contrast image characteristics, thereby realizing more intelligent analysis of the breast cancer axillary lymph node metastasis state, and assisting in diagnosing breast axillary lymph node metastasis so as to better guide clinical diagnosis and decision.

It should be appreciated that the gated response based feature vector dynamic interaction fusion module, the AlexNet model based image feature extractor, the feature attention selection based compression-suppression optimization module, the meta-network based cross-domain joint encoder, and the classifier based transition state analysis result generator need to be trained prior to the inference using the neural network model described above. That is, the breast cancer axillary lymph node metastasis state analysis system 300 based on ultrasound contrast image according to the present application further comprises a training stage for training the feature vector dynamic interaction fusion module based on gating response, the image feature extractor based on AlexNet model, the compression-suppression optimization module based on feature attention selection, the cross-domain joint encoder based on meta-network, and the classifier based metastasis state analysis result generator.

Fig. 3 is a block diagram of a training phase of a breast cancer axillary lymph node metastasis status analysis system based on ultrasound contrast images in accordance with an embodiment of the present application. As shown in fig. 3, the breast cancer axillary lymph node metastasis status analysis system 300 based on an ultrasound contrast image according to an embodiment of the present application includes: training module 400, comprising: a training data acquisition unit 410, configured to acquire training data, where the training data includes training breast cancer ultrasound contrast images; a training image statistical feature extraction unit 420, configured to extract HOG features and LBP features of the training breast cancer ultrasound contrast image to obtain a training breast cancer ultrasound contrast HOG feature vector and a training breast cancer ultrasound contrast LBP feature vector; the training multi-mode statistical feature fusion unit 430 is configured to input the training breast cancer ultrasound imaging LBP feature vector and the training breast cancer ultrasound imaging HOG feature vector into a feature vector dynamic interaction fusion module based on gating response to obtain a training breast cancer ultrasound imaging multi-mode gating interaction fusion vector; the training image implicit feature extraction optimizing unit 440 is configured to perform image feature extraction on the training breast cancer ultrasound contrast image, and input the image feature extraction to the compression-suppression optimizing module based on feature attention selection to obtain an optimized training breast cancer ultrasound contrast feature map; the training multi-mode feature cross-domain joint coding unit 450 is used for inputting the optimized training breast cancer ultrasound contrast feature map and the training breast cancer ultrasound contrast multi-mode gating interaction fusion vector into a cross-domain joint coder based on a meta-network to obtain a breast cancer ultrasound contrast modulation feature map under the constraint of training statistical features; a classification loss unit 460, configured to obtain a classification loss function value by using the ultrasound contrast modulation feature map of breast cancer under the training statistical feature constraint through a transfer state analysis result generator based on a classifier; the loss calculation unit 470 is configured to calculate an ultrasound contrast modulation loss function value of the breast cancer under the statistical feature constraint based on the ultrasound contrast modulation feature loss term of the breast cancer under the statistical feature constraint; a weighting unit 480 for calculating a weighted sum of the classification loss function value and the breast cancer ultrasound contrast modulation loss function value under the statistical feature constraint to obtain a final loss function value; the training unit 490 is configured to train the feature vector dynamic interaction fusion module based on the gating response, the image feature extractor based on AlexNet models, the compression-suppression optimization module based on feature attention selection, the cross-domain joint encoder based on the meta-network, and the transfer state analysis result generator based on the classifier based on the final loss function value.

In particular, it is preferred to further introduce a new loss function value outside the classification loss function value, wherein the construction of the new loss function value comprises the steps of:

The breast cancer ultrasonic radiography modulation feature map under the training statistical feature constraint is unfolded into a breast cancer ultrasonic radiography modulation feature vector under the training statistical feature constraint;

Calculating a breast cancer ultrasound contrast modulation and matrix under the training statistical feature constraint based on the breast cancer ultrasound contrast modulation feature vector under the training statistical feature constraint and a breast cancer ultrasound contrast modulation difference matrix under the training statistical feature constraint, wherein the breast cancer ultrasound contrast modulation and matrix under the training statistical feature constraint is the first The characteristic value of the position is the first ultrasonic contrast modulation characteristic vector of the breast cancer under the constraint of the training statistical characteristicEigenvalue sum of firstThe mean value of the characteristic values and the first of the breast cancer ultrasonic contrast modulation difference matrixes under the constraint of the training statistical characteristicsThe characteristic value of the position is the first ultrasonic contrast modulation characteristic vector of the breast cancer under the constraint of the training statistical characteristicEigenvalue sum of firstOne half of the absolute value of the difference of the eigenvalues;

Multiplying the breast cancer ultrasound contrast modulation feature vector under the training statistical feature constraint with the breast cancer ultrasound contrast modulation sum matrix under the training statistical feature constraint and the breast cancer ultrasound contrast modulation difference matrix under the training statistical feature constraint respectively to obtain a breast cancer ultrasound contrast modulation query sum vector under the training statistical feature constraint and a breast cancer ultrasound contrast modulation query difference vector under the training statistical feature constraint;

Calculating the vector inner product of the breast cancer ultrasonic radiography modulation inquiry under the training statistical characteristic constraint and the vector and the breast cancer ultrasonic radiography modulation inquiry difference vector under the training statistical characteristic constraint to obtain a breast cancer ultrasonic radiography modulation characteristic loss term under the first statistical characteristic constraint;

multiplying the breast cancer ultrasound contrast modulation sum matrix under the training statistical characteristic constraint and the breast cancer ultrasound contrast modulation difference matrix under the training statistical characteristic constraint by a matrix, and calculating a product matrix The norm is used for obtaining a breast cancer ultrasonic contrast modulation characteristic loss term under the constraint of the second statistical characteristic;

Subtracting the product of the pre-determined weight super-parameter and the breast cancer ultrasound contrast modulation characteristic loss term under the first statistical characteristic constraint from the breast cancer ultrasound contrast modulation characteristic loss term under the second statistical characteristic constraint to obtain a new loss function value.

Wherein, the new loss function value, for example, is called as the breast cancer ultrasonic contrast modulation loss function value under the statistical feature constraint, and specifically expressed as:

；

Wherein, Modulating feature vectors for breast cancer ultrasonic radiography under the training statistical feature constraint,AndRespectively a breast cancer ultrasonic radiography modulation and matrix under the training statistical characteristic constraint and a breast cancer ultrasonic radiography modulation difference matrix under the training statistical characteristic constraint,AndRespectively carrying out ultrasonic contrast modulation and matrix of the breast cancer under the constraint of the training statistical characteristics and carrying out ultrasonic contrast modulation difference matrix of the breast cancer under the constraint of the training statistical characteristicsThe characteristic value of the location is used to determine,AndThe first ultrasonic contrast modulation characteristic vector of the breast cancer under the training statistical characteristic constraintEigenvalue sum of firstThe characteristic value of the characteristic value is calculated,For the matrix multiplication to be performed,The transpose of the vector is represented,To calculate a matrixThe norm of the sample is calculated,For the predetermined weight to exceed the parameters,The loss function value is modulated for the ultrasonic contrast of the breast cancer under the constraint of statistical characteristics.

Here, considering that the optimized training breast cancer ultrasound contrast feature map represents image semantic features of the training breast cancer ultrasound contrast image, and the breast cancer ultrasound contrast multi-mode gating interactive fusion vector represents dynamic interactive fusion features of LBP features and HOG features of the training breast cancer ultrasound contrast image, when the optimized training breast cancer ultrasound contrast feature map and the training breast cancer ultrasound contrast multi-mode gating interactive fusion vector are input into a cross-domain joint encoder based on a meta-network, the breast cancer ultrasound contrast modulation feature map under the constraint of the training statistical features also has classification regression recognition difficulty due to image feature distribution differences under different dimensions, so that accuracy of classification results is affected.

Therefore, the application carries out query composition of detail inner product space in the breast cancer ultrasonic radiography modulation characteristic diagram under the training statistical characteristic constraint through the short-distance trans-scale detail linked structural characteristic representation of the breast cancer ultrasonic radiography modulation characteristic diagram under the training statistical characteristic constraint, approximates the low-rank independent observable composition of the link detail composition provided by structural detail interaction of the breast cancer ultrasonic radiography modulation characteristic diagram under the training statistical characteristic constraint, and carries out training through the distributed detail group of the breast cancer ultrasonic radiography modulation characteristic diagram under the training statistical characteristic constraint on the basis of detail complexity so as to promote classification regression decomposition recognition of the breast cancer ultrasonic radiography modulation characteristic diagram under the training statistical characteristic constraint based on complex characteristic representation and improve the accuracy of classification results. Therefore, the analysis and judgment of the breast cancer axillary lymph node metastasis state can be more accurately carried out, so that whether the breast cancer axillary lymph node metastasis occurs or not is detected, and diagnosis of the breast axillary lymph node metastasis is assisted, so that clinical diagnosis and decision are better guided.

As described above, the breast cancer axillary lymph node metastasis status analysis system 300 based on an ultrasound contrast image according to an embodiment of the present application can be implemented in various wireless terminals, such as a server or the like having a breast cancer axillary lymph node metastasis status analysis algorithm based on an ultrasound contrast image. In one possible implementation, the ultrasound contrast image based breast cancer axillary lymph node metastasis status analysis system 300 according to embodiments of the present application may be integrated into the wireless terminal as a software module and/or hardware module. For example, the ultrasound contrast image based breast cancer axillary lymph node metastasis status analysis system 300 may be a software module in the operating system of the wireless terminal or may be an application developed for the wireless terminal; of course, the ultrasound contrast image based breast cancer axillary lymph node metastasis status analysis system 300 can also be one of many hardware modules of the wireless terminal.

Alternatively, in another example, the ultrasound contrast image based breast cancer axillary lymph node metastasis status analysis system 300 and the wireless terminal may also be separate devices, and the ultrasound contrast image based breast cancer axillary lymph node metastasis status analysis system 300 may be connected to the wireless terminal through a wired and/or wireless network and transmit interactive information in a agreed data format.

The foregoing description of the embodiments of the present disclosure has been presented for purposes of illustration and description, and is not intended to be exhaustive or limited to the embodiments disclosed. Many modifications and variations will be apparent to those of ordinary skill in the art without departing from the scope and spirit of the various embodiments described. The terminology used herein was chosen in order to best explain the principles of the embodiments, the practical application, or the improvement of technology in the marketplace, or to enable others of ordinary skill in the art to understand the embodiments disclosed herein.

Claims

1. An ultrasound contrast image-based breast cancer axillary lymph node metastasis status analysis system, comprising:

The transfer state analysis module is used for determining a transfer state analysis result based on the ultrasonic contrast modulation characteristic diagram of the breast cancer under the constraint of the statistical characteristics, wherein the transfer state analysis result is used for indicating whether the probability of transfer exceeds a preset threshold value;

wherein, the implicit characteristic extraction optimization module of the ultrasonic contrast image comprises:

An ultrasound contrast feature attention selection optimizing unit, configured to input the breast cancer ultrasound contrast feature map into a compression-suppression optimizing module based on feature attention selection to obtain the optimized breast cancer ultrasound contrast feature map;

Wherein the ultrasound contrast feature attention selection optimizing unit comprises:

2. The ultrasound contrast image-based breast cancer axillary lymph node metastasis status analysis system of claim 1, wherein the ultrasound contrast image statistics feature extraction module is configured to:

3. The ultrasound contrast image-based breast cancer axillary lymph node metastasis status analysis system of claim 2, wherein the breast cancer ultrasound contrast multi-modality statistical feature fusion module comprises:

4. The ultrasound contrast image-based breast cancer axillary lymph node metastasis status analysis system of claim 3, wherein the compressed information feature extraction subunit is configured to:

5. The ultrasound contrast image based breast cancer axillary lymph node metastasis status analysis system of claim 4, wherein the optimization subunit is configured to: and multiplying the feature value of each position in the breast cancer ultrasonic radiography weight feature vector with each feature matrix of the breast cancer ultrasonic radiography feature map along the channel dimension according to the position to obtain the optimized breast cancer ultrasonic radiography feature map.

6. The ultrasound contrast image-based breast cancer axillary lymph node metastasis status analysis system of claim 5, wherein said metastasis status analysis module is configured to: inputting the ultrasonic contrast modulation characteristic diagram of the breast cancer under the constraint of the statistical characteristic into a transfer state analysis result generator based on a classifier to obtain a transfer state analysis result, wherein the transfer state analysis result is used for indicating whether the probability of occurrence of transfer exceeds a preset threshold value.

7. The ultrasound contrast image based breast cancer axillary lymph node metastasis state analysis system of claim 6, further comprising a training module for training the gating response based feature vector dynamic interaction fusion module, the AlexNet model based image feature extractor, the feature attention selection based compression-suppression optimization module, the meta-network based cross-domain joint encoder, and the classifier based metastasis state analysis result generator.

8. The ultrasound contrast image based breast cancer axillary lymph node metastasis status analysis system of claim 7, wherein said training module comprises: