Breast ultrasonic image automatic analysis system based on deep learning

Abstract

The invention discloses an automatic breast ultrasound image analysis system based on deep learning. The automatic breast ultrasound image analysis system comprises an ultrasound image input module, an ultrasound image preprocessing module, an image enhancement module, a classification diagnosis module, a diagnosis verification module, a diagnosis optimization module and a diagnosis adjustment module, according to the breast ultrasound image automatic analysis system based on deep learning provided by the invention, the attribute component and the illumination component of the preprocessed ultrasound image of the patient are decomposed based on the image light intensity to obtain the inherent characteristics of the image, and the image enhancement processing is carried out based on the attribute component by using the adversarial network algorithm. A high-quality patient ultrasonic image is obtained, an intelligent diagnosis model is constructed through a deep clustering algorithm, training of the intelligent diagnosis model is carried out by utilizing the ultrasonic images of a large number of patients, and intelligent diagnosis of the patient ultrasonic image is achieved.

Description

Technical Field

[0001] The present invention relates to the field of intelligent medical treatment, and in particular to an automatic breast ultrasound image analysis system based on deep learning. Background Art

[0002] With the rapid development of medical imaging technology, especially the application of artificial intelligence and deep learning technology, the field of medical image analysis has made significant progress. Driven by these technologies, traditional medical image analysis methods have gradually been replaced by automated and intelligent analysis systems, and the application of related technologies has provided strong auxiliary support for clinicians.

[0003] Breast cancer is one of the most common malignant tumors in women worldwide. Its early diagnosis is of vital importance to improve the cure rate and reduce the mortality rate. Breast ultrasound examination is a non-invasive, radiation-free, low-cost and easy-to-operate imaging examination method, which is widely used in the screening and diagnosis of breast diseases. However, the diagnosis and analysis of breast ultrasound images usually rely on the experience of imaging doctors, and the ultrasound images themselves may be affected by many factors, such as equipment differences and patient body shape, which makes the results of breast ultrasound examination easily interfered by subjective factors, thus affecting the accuracy and consistency of diagnosis. With the development of ultrasound image acquisition technology, image quality and resolution have been significantly improved, but these advances have not fundamentally solved the problems that still exist in the image analysis process, such as noise in ultrasound images, low image contrast, unclear mass boundaries, etc., especially in the case of small early lesions and atypical manifestations, which makes diagnosis more difficult. In order to solve these problems, many researchers have tried to introduce computer-aided diagnosis systems and use automated algorithms to process breast ultrasound images to improve diagnostic accuracy and efficiency. However, the existing computer-aided diagnosis system still has many limitations in practical applications. The characteristics of breast ultrasound images are large image noise, low contrast, and the morphology and boundaries of the mass may not be obvious, which makes it difficult for traditional deep learning models to effectively extract effective features, resulting in insufficient classification accuracy. In order to make up for this deficiency, some researchers have proposed a solution to improve image quality through image preprocessing and enhancement. For example, using denoising, contrast enhancement and other technologies to improve the visibility and clarity of the image, so that the deep learning algorithm can better identify and classify the lesion area and improve the accuracy of breast ultrasound image analysis. In response to these challenges, the present invention proposes a breast ultrasound image automatic analysis system based on deep learning, which aims to solve the deficiencies in the prior art through the collaborative work of multiple modules, and make innovations in improving diagnostic accuracy, optimizing image processing effects, and improving system robustness. Through the combination of image preprocessing, enhancement, intelligent classification diagnosis, diagnostic verification and feedback mechanism, the system of the present invention can provide more accurate and reliable diagnostic results in breast ultrasound image analysis, which can not only improve the early screening effect of breast diseases, but also reduce the workload of doctors and improve diagnostic efficiency. It has important clinical application value and broad market prospects. Summary of the invention

[0004] The purpose of the present invention is to provide a deep learning-based automatic analysis system for breast ultrasound images, aiming to solve the problems mentioned in the above background.

[0005] In order to achieve the above-mentioned purpose, the present invention proposes a breast ultrasound image automatic analysis system based on deep learning, including an ultrasound image input module, an ultrasound image preprocessing module, an image enhancement module, a classification diagnosis module, a diagnosis verification module, a diagnosis optimization module, and a diagnosis adjustment module; the ultrasound image input module inputs the patient's ultrasound image by selecting the patient's ultrasound image; the ultrasound image preprocessing module preprocesses the received patient ultrasound image to obtain ultrasound image data; the image enhancement module proposes an adversarial enhancement algorithm to perform image enhancement operations on the patient's ultrasound image; the classification diagnosis module uses a deep clustering algorithm to construct an intelligent diagnosis model to perform intelligent diagnosis classification of the patient's ultrasound image; in the diagnosis verification module, the imaging doctor verifies the intelligent diagnosis result through the ultrasound image data and the intelligent diagnosis result; the diagnosis optimization module outputs the final diagnosis result through the diagnosis verification result of the imaging doctor; the diagnosis adjustment module optimizes the intelligent diagnosis model based on the comparative analysis of the intelligent diagnosis result and the final diagnosis result.

[0006] Furthermore, an ultrasound image input module is used to input the ultrasound image of the patient and transmit it to the ultrasound image preprocessing module.

[0007] Furthermore, the ultrasound image preprocessing module performs an image normalization operation on the received ultrasound image of the patient, eliminates the noise of the ultrasound image by using a filter, and corrects the ultrasound image pixels by using a correction curve.

[0008] Furthermore, the adversarial enhancement algorithm performs image enhancement processing by decomposing the pre-processed ultrasound image of the patient into attribute components and illumination components based on light intensity.

[0009] Further, the image enhancement process is as follows:

[0010] For the pre-processed ultrasound image of the patient, the image intensity is used for image representation, which is expressed as follows:

[0011]

[0012] Among them, S ori Represented as the original image, I 1,1 ,I 1,n ,I i,j ,I m,1 ,I m,nThey respectively represent the light intensity of the pixel in the 1st row and 1st column of the original image, the light intensity of the pixel in the 1st row and nth column of the original image, the light intensity of the pixel in the ith row and jth column of the original image, the light intensity of the pixel in the mth row and 1st column of the original image, and the light intensity of the pixel in the mth row and nth column of the original image. The original image consists of two parts: attribute component and illumination component. The attribute component represents the inherent characteristics of the image, and the illumination component represents the dynamic range of the image. First, the image is decomposed based on the attribute component and the illumination component, and the scale function is defined. The scale function is expressed as follows:

[0013]

[0014] Among them, x i,j ,y i,j They represent the horizontal and vertical coordinates of the coordinate values ​​of the pixel point in the i-th row and j-th column in the original image, g(·) represents the scale function, α i,j represents the decomposition coefficient, I mean Represents the mean value of light intensity. Through the scaling function, the original image is decomposed based on attribute components and illumination components. The formula is as follows:

[0015] R i,j =ln(I i,j )-ln(g(x i,j ,y i,j )×|I i,j -I mean |)

[0016] Among them, R i,j Represents the value of the attribute component of the pixel point in the i-th row and j-th column of the original image, eliminates the influence of some illumination on the image, and obtains the attribute component that reflects the basic attributes of the image. Based on the attribute component, image enhancement processing is performed, and the generation function and discriminant function are constructed. The image attribute component is represented as R, which constitutes the attribute component data set, and the distribution function is F R (r), through the generating function, the generated data of the image attribute component is generated, and the generated data is input into the discriminant function, and the discriminant function discriminates the generated data. The discriminant function is expressed as follows:

[0017]

[0018] Where D(r) represents the discriminant function, G R (r) represents the distribution function of the generated data set, β1 represents the first discriminant coefficient, β2 represents the second discriminant coefficient, and then the adversarial target optimization function is constructed. The function is expressed as follows:

[0019] L=E[ln(G R (r))]+E[ln(1-D(r))]

[0020] Wherein, L represents the adversarial target optimization function, and E[·] represents the expectation. By minimizing the adversarial target optimization function, the generated data is optimized to achieve image enhancement processing. The adversarial enhancement algorithm proposed in the present invention decomposes the patient's ultrasound image into attribute classification that expresses inherent characteristics and illumination components that express dynamic range based on light intensity using a scale function, and then constructs an attribute component data set based on the attribute components, generates image data using a generating function, discriminates data using a discriminant function, constructs an adversarial target optimization function, optimizes the generated data, and realizes that the discriminant function cannot determine whether the data comes from the generating function or the attribute component data set, thereby completing image enhancement. The adversarial enhancement algorithm proposed in the present invention performs image enhancement based on the inherent attributes of the image, which can reduce the impact of image quality and improve image quality.

[0021] Furthermore, the intelligent diagnosis model divides the patient's ultrasound image into grids to form basic areas, and based on the separation of the basic areas, uses a deep clustering algorithm to perform clustering to perform intelligent diagnostic classification of the patient's ultrasound image.

[0022] Further, the intelligent diagnosis model, the detailed process is as follows:

[0023] For the enhanced ultrasound image of the patient, the enhanced ultrasound image of the patient is divided into regions of the same pixel size by grid division, which are defined as basic regions. Each basic region has the same number of pixels, as shown below:

[0024]

[0025] Among them, T k represents the kth basic region, They are respectively represented by the grayscale value of the pixel in the 1st row and 1st column of the kth basic area, the grayscale value of the pixel in the 1st row and nth column of the kth basic area, the grayscale value of the pixel in the ith row and jth column of the kth basic area, the grayscale value of the pixel in the nth row and 1st column of the kth basic area, and the grayscale value of the pixel in the nth row and nth column of the kth basic area. Based on the grayscale values ​​of the pixels in the basic area, the separation degree of the basic area is calculated, and the calculation formula is as follows:

[0026]

[0027] Among them, U k represents the separation degree of the kth basis region, Represents the gray value of the pixel in the ath row and the bth column of the kth basic area. Based on the separation degree of the basic area, the basic area is clustered. First, a cluster center is selected in each basic area, and the first clustering iteration is performed to calculate the cluster components of the basic area and the cluster center. The cluster components are used to cluster the basic area. The cluster component calculation formula is as follows:

[0028]

[0029] Among them, C k,cen,i represents the clustering component of the kth basic region and the i-th cluster center, P k,cen,i represents the distance between the kth base region and the i-th cluster center, U cen,i Represents the separation degree of the i-th cluster center. Based on the clustering components of the basic area and the cluster center, the basic area is clustered to obtain different clusters. Then, clustering is performed again. In different clusters, the center points in different clusters are selected as new cluster centers for clustering. The clustering components of the basic area and the basic center are calculated again. Clustering is performed again based on the clustering components. It is iterated until the cluster center is no longer updated and clustering convergence is achieved. At this time, the clustering division of the basic area is completed. Based on the clustering division of the basic area in the image, the ultrasound image is classified. The classified ultrasound image is manually reviewed and the classification result is optimized. The ultrasound image of the patient's lesion is used to form a detection set. The ultrasound image of the patient's lesion in the detection set is input into the intelligent diagnosis model to test the model classification result. The intelligent diagnosis model that achieves the accuracy target completes the model training. The intelligent diagnosis model constructed by the present invention utilizes a deep clustering algorithm and grid division to divide the ultrasound image of the patient after enhancement processing into basic areas of the same pixel size, calculates the separation degree of the basic area based on the grayscale value of the pixel points in the basic area, clusters the basic area by selecting cluster centers, continuously iterates the clustering operation to achieve clustering convergence, completes the clustering division of the basic area, classifies the ultrasound image, and then uses the ultrasound image of the patient's lesion to form a detection set to test the classification results of the intelligent diagnosis model, optimize the intelligent diagnosis model, and complete the training of the intelligent diagnosis model. The intelligent diagnosis model proposed by the present invention performs clustering based on the basic area of ​​the image, enhances the image boundary, and improves the accuracy of model classification.

[0030] Furthermore, in the diagnosis verification module, the imaging doctor makes a verification diagnosis result through the patient's ultrasound image data and the diagnosis classification result output by the intelligent diagnosis model, and verifies the diagnosis classification result of the intelligent diagnosis model.

[0031] Furthermore, the diagnosis optimization module combines the intelligent diagnosis model with the diagnosis results of the imaging doctor, optimizes the results of the intelligent diagnosis model, outputs the final diagnosis results, and issues a final diagnosis report.

[0032] Furthermore, the diagnosis adjustment module analyzes and compares the final diagnosis result and the diagnosis classification result of the intelligent diagnosis model, optimizes the parameters within the intelligent diagnosis model, further clusters the image pixels, performs deep clustering on the basic image area, and optimizes the intelligent diagnosis model.

[0033] Beneficial Effects

[0034] 1. The adversarial enhancement algorithm proposed in the present invention decomposes the patient's ultrasound image into attribute classification that represents inherent characteristics and illumination components that represent dynamic range based on light intensity and using a scale function, and then constructs an attribute component data set based on the attribute components, generates image data using a generating function, and discriminates the data using a discriminant function, and constructs an adversarial target optimization function to optimize the generated data, so that the discriminant function cannot determine whether the data comes from the generating function or the attribute component data set, thereby completing image enhancement. The adversarial enhancement algorithm proposed in the present invention performs image enhancement based on the inherent attributes of the image, can reduce the impact of image quality, and improve image quality.

[0035] 2. The intelligent diagnosis model constructed by the present invention utilizes a deep clustering algorithm and grid division to divide the enhanced ultrasound image of the patient into basic areas of the same pixel size, calculates the separation of the basic areas based on the grayscale values ​​of the pixels in the basic areas, selects cluster centers to cluster the basic areas, continuously iterates clustering operations to achieve clustering convergence, completes clustering division of the basic areas, classifies the ultrasound images, and then uses the ultrasound images of the patient's lesions to form a detection set to test the classification results of the intelligent diagnosis model, optimize the intelligent diagnosis model, and complete the training of the intelligent diagnosis model. The intelligent diagnosis model proposed by the present invention performs clustering based on the basic areas of the image, enhances the image boundaries, and improves the accuracy of model classification. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the prior art descriptions. Obviously, the drawings in the following description are only some embodiments of the present invention, and for ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0037] Figure 1 is a system schematic diagram of the present invention; DETAILED DESCRIPTION

[0038] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0039] In order to achieve the above-mentioned purpose, the present invention proposes a breast ultrasound image automatic analysis system based on deep learning, including an ultrasound image input module, an ultrasound image preprocessing module, an image enhancement module, a classification diagnosis module, a diagnosis verification module, a diagnosis optimization module, and a diagnosis adjustment module; the ultrasound image input module inputs the patient's ultrasound image by selecting the patient's ultrasound image; the ultrasound image preprocessing module preprocesses the received patient ultrasound image to obtain ultrasound image data; the image enhancement module proposes an adversarial enhancement algorithm to perform image enhancement operations on the patient's ultrasound image; the classification diagnosis module uses a deep clustering algorithm to construct an intelligent diagnosis model to perform intelligent diagnosis classification of the patient's ultrasound image; in the diagnosis verification module, the imaging doctor verifies the intelligent diagnosis result through the ultrasound image data and the intelligent diagnosis result; the diagnosis optimization module outputs the final diagnosis result through the diagnosis verification result of the imaging doctor; the diagnosis adjustment module optimizes the intelligent diagnosis model based on the comparative analysis of the intelligent diagnosis result and the final diagnosis result.

[0040] Specifically, the ultrasound image input module, the patient's ultrasound image is input through the ultrasound image input module and transmitted to the ultrasound image preprocessing module.

[0041] Specifically, the ultrasound image preprocessing module performs an image normalization operation on the received ultrasound image of the patient, eliminates the noise of the ultrasound image by using a filter, and corrects the ultrasound image pixels by using a correction curve.

[0042] Specifically, the adversarial enhancement algorithm performs image enhancement processing by decomposing the pre-processed ultrasound image of the patient into attribute components and illumination components based on light intensity. The detailed process is as follows:

[0043] For the pre-processed ultrasound image of the patient, the image intensity is used for image representation, which is expressed as follows:

[0044]

[0045] Among them, S ori Represented as the original image, I 1,1 ,I 1,n ,I i,j ,I m,1 ,I m,n They respectively represent the light intensity of the pixel in the 1st row and 1st column of the original image, the light intensity of the pixel in the 1st row and nth column of the original image, the light intensity of the pixel in the ith row and jth column of the original image, the light intensity of the pixel in the mth row and 1st column of the original image, and the light intensity of the pixel in the mth row and nth column of the original image. The original image consists of two parts: attribute component and illumination component. The attribute component represents the inherent characteristics of the image, and the illumination component represents the dynamic range of the image. First, the image is decomposed based on the attribute component and the illumination component, and the scale function is defined. The scale function is expressed as follows:

[0046]

[0047] Among them, x i,j ,y i,j They represent the horizontal and vertical coordinates of the coordinate values ​​of the pixel point in the i-th row and j-th column in the original image, g(·) represents the scale function, α i,j represents the decomposition coefficient, I mean Represents the mean value of light intensity. Through the scaling function, the original image is decomposed based on attribute components and illumination components. The formula is as follows:

[0048] R i,j =ln(I i,j )-ln(g(x i,j ,y i,j )×|I i,j -I mean |)

[0049] Among them, R i,j Represents the value of the attribute component of the pixel point in the i-th row and j-th column of the original image, eliminates the influence of some illumination on the image, and obtains the attribute component that reflects the basic attributes of the image. Based on the attribute component, the adversarial network model is used to perform image enhancement processing, construct the generation function and the discriminant function, and the image attribute component is represented as R, which constitutes the attribute component data set, and the distribution function is F R (r), through the generating function, the generated data of the image attribute component is generated, the generated data is input into the discriminant function, the generated data is input into the discriminant function, and the discriminant function discriminates the generated data. The discriminant function is expressed as follows:

[0050]

[0051] Where D(r) represents the discriminant function, G R (r) represents the distribution function of the generated data set, β1 represents the first discriminant coefficient, β2 represents the second discriminant coefficient, and then the adversarial target optimization function is constructed. The function is expressed as follows:

[0052] L=E[ln(G R (r))]+E[ln(1-D(r))]

[0053] Where L represents the adversarial objective optimization function, and E[·] represents the expectation. By minimizing the adversarial objective optimization function, the generated data is optimized to achieve image enhancement processing.

[0054] Specifically, the intelligent diagnosis model divides the patient's ultrasound image into grids to form basic areas. Based on the separation of the basic areas, the deep clustering algorithm is used for clustering to perform intelligent diagnosis and classification of the patient's ultrasound images. The detailed process is as follows:

[0055] For the enhanced ultrasound image of the patient, the enhanced ultrasound image of the patient is divided into regions of the same pixel size by grid division, which are defined as basic regions. Each basic region has the same number of pixels, as shown below:

[0056]

[0057] Among them, T k represents the kth basic region, They are respectively represented by the grayscale value of the pixel in the 1st row and 1st column of the kth basic area, the grayscale value of the pixel in the 1st row and nth column of the kth basic area, the grayscale value of the pixel in the ith row and jth column of the kth basic area, the grayscale value of the pixel in the nth row and 1st column of the kth basic area, and the grayscale value of the pixel in the nth row and nth column of the kth basic area. Based on the grayscale values ​​of the pixels in the basic area, the separation degree of the basic area is calculated, and the calculation formula is as follows:

[0058]

[0059] Among them, U k represents the separation degree of the kth basis region, Represents the gray value of the pixel in the ath row and the bth column of the kth basic area. Based on the separation degree of the basic area, the basic area is clustered. First, a cluster center is selected in each basic area, and the first clustering iteration is performed to calculate the cluster components of the basic area and the cluster center. The cluster components are used to cluster the basic area. The cluster component calculation formula is as follows:

[0060]

[0061] Among them, C k,cen,i represents the clustering component of the kth basic region and the i-th cluster center, P k,cen,i represents the distance between the kth base region and the i-th cluster center, U cen,iRepresents the separation degree of the i-th cluster center. Based on the clustering components of the basic area and the cluster center, the basic area is clustered to obtain different clusters. Then, clustering is performed again. In different clusters, the center points in different clusters are selected as new cluster centers for clustering. The clustering components of the basic area and the basic center are calculated again. Clustering is performed again based on the clustering components. It is iterated until the cluster center is no longer updated and clustering convergence is achieved. At this time, the clustering division of the basic area is completed. Based on the clustering division of the basic area in the image, the ultrasound image is classified. The classified ultrasound image is manually reviewed and the classification result is optimized. The ultrasound image of the patient's lesion is used to form a detection set. The ultrasound image of the patient's lesion in the detection set is input into the intelligent diagnosis model to test the model classification result. The intelligent diagnosis model that achieves the accuracy target completes the model training.

[0062] Specifically, in the diagnosis verification module, the imaging doctor makes a verification diagnosis result through the patient's ultrasound image data and the diagnosis classification result output by the intelligent diagnosis model, and verifies the diagnosis classification result of the intelligent diagnosis model.

[0063] Specifically, the diagnosis optimization module combines the intelligent diagnosis model and the diagnosis results of the imaging doctor, optimizes the results of the intelligent diagnosis model, outputs the final diagnosis results, and issues a final diagnosis report.

[0064] Specifically, the diagnosis adjustment module analyzes and compares the final diagnosis result and the diagnosis classification result of the intelligent diagnosis model, optimizes the parameters within the intelligent diagnosis model, further clusters the image pixels, performs deep clustering on the basic image areas, and optimizes the intelligent diagnosis model.

[0065] In a specific embodiment, the user inputs the patient's ultrasound image through the ultrasound image input module, transmits it to the ultrasound image preprocessing module, performs image normalization operation on the patient's ultrasound image, uses a filter to eliminate ultrasound image noise, and uses a correction curve to correct ultrasound image pixels; then, using an adversarial enhancement algorithm, the preprocessed patient's ultrasound image is decomposed into attribute components and illumination components based on light intensity, and image enhancement processing is performed to obtain a high-quality patient ultrasound image; then, using an intelligent diagnosis model, the patient's ultrasound image after image enhancement processing is intelligently diagnosed; on this basis, the imaging doctor makes a verification diagnosis result through the patient's ultrasound image data and the diagnosis classification result output by the intelligent diagnosis model, and verifies the diagnosis classification result of the intelligent diagnosis model; then, combining the intelligent diagnosis model and the imaging doctor's diagnosis result, the result of the intelligent diagnosis model is optimized, the final diagnosis result is output, and a final diagnosis report is issued; finally, the final diagnosis result and the diagnosis classification result of the intelligent diagnosis model are analyzed and compared to optimize the intelligent diagnosis model.

[0066] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0067] Each embodiment in this specification is described in a related manner, and the same or similar parts between the embodiments can be referred to each other, and 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 parts can be referred to the partial description of the method embodiment.

[0068] The above description is only a preferred embodiment of the present invention and is not intended to limit the protection scope of the present invention. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention are included in the protection scope of the present invention.

Claims

1. A deep learning-based breast ultrasound image automatic analysis system, comprising an ultrasound image input module, an ultrasound image preprocessing module, an image enhancement module, a classification diagnosis module, a diagnosis verification module, a diagnosis optimization module, and a diagnosis adjustment module; the ultrasound image input module selects the ultrasound image of the patient and inputs the ultrasound image of the patient; the ultrasound image preprocessing module preprocesses the received ultrasound image of the patient to obtain ultrasound image data; the image enhancement module proposes an adversarial enhancement algorithm to perform image enhancement operations on the ultrasound image of the patient; The classification diagnosis module uses a deep clustering algorithm to build an intelligent diagnosis model to perform intelligent diagnosis classification of patient ultrasound images; in the diagnosis verification module, the imaging doctor verifies the intelligent diagnosis results through ultrasound image data and intelligent diagnosis results; the diagnosis optimization module outputs the final diagnosis result based on the diagnostic verification results of the imaging doctor; The diagnosis adjustment module optimizes the intelligent diagnosis model based on the comparative analysis of the intelligent diagnosis results and the final diagnosis results.

2. The deep learning-based breast ultrasound image automatic analysis system according to claim 1, characterized in that: The ultrasonic image input module is used to input the ultrasonic image of the patient and transmit it to the ultrasonic image preprocessing module.

3. The deep learning-based breast ultrasound image automatic analysis system according to claim 1, characterized in that: The ultrasound image preprocessing module performs an image normalization operation on the received ultrasound image of the patient, eliminates the noise of the ultrasound image by using a filter, and corrects the ultrasound image pixels by using a correction curve.

4. The deep learning-based breast ultrasound image automatic analysis system according to claim 1, characterized in that: The adversarial enhancement algorithm performs image enhancement processing by decomposing the pre-processed ultrasound image of the patient into attribute components and illumination components based on light intensity.

5. The deep learning-based breast ultrasound image automatic analysis system according to claim 4, characterized in that: The image enhancement process is as follows: For the pre-processed ultrasound image of the patient, the image intensity is used for image representation, which is expressed as follows: Among them, Sori represents the original image, I1,1, I1,n, Ii,j, Im,1, Im,n represent the light intensity of the pixel in the 1st row and 1st column of the original image, the light intensity of the pixel in the 1st row and nth column of the original image, the light intensity of the pixel in the ith row and jth column of the original image, the light intensity of the pixel in the mth row and 1st column of the original image, and the light intensity of the pixel in the mth row and nth column of the original image, respectively. The original image consists of two parts: attribute component and illumination component. The attribute component represents the inherent characteristics of the image, and the illumination component represents the dynamic range of the image. First, the image is decomposed based on the attribute component and the illumination component, and the scale function is defined. The scale function is expressed as follows: Among them, xi,j and yi,j represent the horizontal and vertical coordinates of the coordinate values ​​of the pixel point in the i-th row and j-th column in the original image, respectively, g(·) represents the scaling function, αi,j represents the decomposition coefficient, and Imean represents the mean value of the light intensity. The scaling function is used to decompose the original image based on the attribute component and the illumination component. The formula is as follows: Ri,j=ln(Ii,j)-ln(g(xi,j,yi,j)×Ii,j-Imean|) Among them, Ri,j represents the value of the attribute component of the pixel point in the i-th row and j-th column of the original image. The influence of some illumination on the image is eliminated to obtain the attribute component reflecting the basic attributes of the image. Based on the attribute component, image enhancement processing is performed to construct the generating function and the discriminant function. The image attribute component is represented by R, which constitutes the attribute component data set. The distribution function is FR(r). The generating function generates the generated data of the image attribute component. The generated data is input into the discriminant function, and the discriminant function discriminates the generated data. The discriminant function is expressed as follows: Among them, D(r) represents the discriminant function, GR(r) represents the distribution function of the generated data set, β1 represents the first discriminant coefficient, β2 represents the second discriminant coefficient, and then constructs the adversarial target optimization function, which is expressed as follows: L=E[ln(GR(r))]+E[ln(1-D(r))] Where L represents the adversarial objective optimization function, and E[·] represents the expectation. By minimizing the adversarial objective optimization function, the generated data is optimized to achieve image enhancement processing.

6. The deep learning-based breast ultrasound image automatic analysis system according to claim 1, characterized in that: The intelligent diagnosis model divides the patient's ultrasound image into grids to form basic areas, and uses a deep clustering algorithm to perform clustering based on the separation of the basic areas to perform intelligent diagnosis classification of the patient's ultrasound image.

7. The deep learning-based breast ultrasound image automatic analysis system according to claim 6, characterized in that: The detailed process of the intelligent diagnosis model is as follows: For the enhanced ultrasound image of the patient, the enhanced ultrasound image of the patient is divided into regions of the same pixel size by grid division, which are defined as basic regions. Each basic region has the same number of pixels, as shown below: Among them, Tk represents the kth basic area, They are respectively represented by the grayscale value of the pixel in the 1st row and 1st column of the kth basic area, the grayscale value of the pixel in the 1st row and nth column of the kth basic area, the grayscale value of the pixel in the ith row and jth column of the kth basic area, the grayscale value of the pixel in the nth row and 1st column of the kth basic area, and the grayscale value of the pixel in the nth row and nth column of the kth basic area. Based on the grayscale values ​​of the pixels in the basic area, the separation degree of the basic area is calculated, and the calculation formula is as follows: Among them, Uk represents the separation degree of the kth basic region, Represents the gray value of the pixel in the ath row and the bth column of the kth basic area. Based on the separation degree of the basic area, the basic area is clustered. First, a cluster center is selected in each basic area, and the first clustering iteration is performed to calculate the cluster components of the basic area and the cluster center. The cluster components are used to cluster the basic area. The cluster component calculation formula is as follows: Among them, Ck,cen,i represents the clustering component of the kth basic area and the ith cluster center, Pk,cen,i represents the distance between the kth basic area and the ith cluster center, and Ucen,i represents the separation degree of the ith cluster center. Based on the clustering components of the basic area and the cluster center, the basic area is clustered to obtain different clusters, and then clustering is performed again. In different clusters, the center points in different clusters are selected as new cluster centers for clustering. The clustering components of the basic area and the basic center are calculated again, and clustering is performed again based on the clustering components. It is iterated until the cluster center is no longer updated and clustering convergence is achieved. At this time, the clustering division of the basic area is completed. Based on the clustering division of the basic area in the image, the ultrasound image is classified. The classified ultrasound image is manually reviewed and the classification result is optimized. Then, the ultrasound image of the patient's lesion is used to form a detection set. The ultrasound image of the patient's lesion in the detection set is input into the intelligent diagnosis model to test the model classification result. The intelligent diagnosis model that achieves the accuracy target completes the model training.

8. The deep learning-based breast ultrasound image automatic analysis system according to claim 1, characterized in that: In the diagnosis verification module, the imaging doctor makes a verification diagnosis result based on the patient's ultrasound image data and the diagnosis classification result output by the intelligent diagnosis model, and verifies the diagnosis classification result of the intelligent diagnosis model.

9. The deep learning-based breast ultrasound image automatic analysis system according to claim 1, characterized in that: The diagnosis optimization module combines the intelligent diagnosis model and the diagnosis results of the imaging doctor, optimizes the results of the intelligent diagnosis model, outputs the final diagnosis results, and issues a final diagnosis report.

10. The deep learning-based breast ultrasound image automatic analysis system according to claim 1, characterized in that: The diagnosis adjustment module analyzes and compares the final diagnosis result and the diagnosis classification result of the intelligent diagnosis model, optimizes the parameters in the intelligent diagnosis model, further clusters the image pixels, performs deep clustering on the basic area of ​​the image, and optimizes the intelligent diagnosis model.

Images