Simplified magnetic resonance breast intelligent recognition system based on deep learning

Abstract

The invention relates to the technical field of disease diagnosis, in particular to a simplified magnetic resonance intelligent breast recognition system based on deep learning, which comprises a magnetic resonance imaging unit, an image enhancement unit, a lump positioning unit, a property detection unit, an intelligent display unit, a model matching unit, a human body modeling unit and a dynamic rendering unit, specifically, the magnetic resonance imaging unit is used for acquiring a mammary gland magnetic resonance image of a target patient based on magnetic resonance imaging equipment and transmitting the mammary gland magnetic resonance image to the image enhancement unit. According to the method, positioning and property detection of the masses are combined, specific information of the breast masses is combined with the three-dimensional human body model, so that doctors and patients can visually see the three-dimensional reconstruction effect of the breast area, the doctors are helped to comprehensively know the positions and properties of the masses in the mammary gland by using the dynamic rendering and three-dimensional modeling technology, and the accuracy of the three-dimensional reconstruction effect is improved. And a more accurate treatment scheme can be formulated.

Description

Technical Field

[0001] The present invention relates to the technical field of disease diagnosis, and in particular to a simplified magnetic resonance breast intelligent recognition system based on deep learning. Background Art

[0002] Deep Learning is a subfield of machine learning that performs data analysis and prediction by simulating the structure and function of the human brain's neural network. Deep learning models are typically composed of multiple levels (i.e., "deep") of neural networks. Magnetic Resonance Imaging (MRI) is a medical imaging technology that uses strong magnetic fields and radio waves to generate detailed images of structures inside the body. It does not rely on X-rays, so it is considered a non-invasive and radiation-free imaging method. MRI is ideal for viewing details of soft tissues (such as the brain, spinal cord, muscles, breasts, etc.) and is widely used in the diagnosis of brain diseases, spinal problems, joint injuries, etc.

[0003] Traditional systems rely on manual analysis or relatively simple image processing technology to detect breast masses. Doctors need to manually observe and judge abnormal areas in breast images, which are easily affected by subjective experience, fatigue, vision and other factors, resulting in a high risk of missed diagnosis or misdiagnosis. In traditional systems, the location of masses usually depends on the doctor's experience and visual judgment, and the accuracy of positioning and analysis is difficult to guarantee. Traditional technologies lack automated precise analysis and multi-dimensional integration, and cannot fully identify the nature of the mass (such as benign or malignant), and are prone to miss tiny details of the lesions. Moreover, most traditional systems can only provide two-dimensional images, and doctors can only make judgments based on planar images, lacking a three-dimensional sense and a global perspective. The spatial position, morphology and other information of breast masses are not intuitive enough, affecting the doctor's comprehensive assessment of the mass. Summary of the invention

[0004] The technical problem to be solved by the present invention is to overcome the shortcomings of the above-mentioned prior art and provide a simplified magnetic resonance breast intelligent identification system based on deep learning.

[0005] The technical solution adopted to solve the above technical problems is: a simplified magnetic resonance breast intelligent recognition system based on deep learning, including a magnetic resonance imaging unit, an image enhancement unit, a mass positioning unit, a property detection unit, an intelligent display unit, a model matching unit, a human body modeling unit and a dynamic rendering unit, specifically:

[0006] The magnetic resonance imaging unit is used to obtain a breast magnetic resonance image of a target patient based on a magnetic resonance imaging device, and transmit the breast magnetic resonance image to an image enhancement unit;

[0007] The image enhancement unit is used to receive the breast magnetic resonance image transmitted by the magnetic resonance imaging unit, and acquire the breast magnetic resonance image of the target patient based on the magnetic resonance imaging device, perform image enhancement on the breast magnetic resonance image to obtain a breast magnetic resonance enhanced image, and transmit the breast magnetic resonance enhanced image to the mass localization unit;

[0008] The mass localization unit is used to receive the breast magnetic resonance enhanced image transmitted by the image enhancement unit, and perform regional localization on multiple mass regions in the breast magnetic resonance enhanced image based on a trained regional coarse localization model to obtain a mass region image, determine whether the target patient has a mass in the breast based on the mass region image, transmit the mass region image to the property detection unit and the model matching unit, and transmit the result of whether the target patient has a mass in the breast to the intelligent display unit.

[0009] Preferably, the property detection unit is used to receive the mass area image transmitted by the mass locating unit, and perform property detection on the mass in the mass area image based on a trained mass detection model to obtain the mass property in the mass area image, determine the mass property in the breast of the target patient based on the mass property in the mass area image, and transmit the mass property in the breast of the target patient to the intelligent display unit and the model matching unit.

[0010] Preferably, the human body modeling unit is used to construct a three-dimensional human body model based on three-dimensional modeling technology, perform region segmentation on the three-dimensional human body model to obtain a three-dimensional model of the breast region, and transmit the three-dimensional model of the breast region to the model matching unit.

[0011] Preferably, the model matching unit is used to receive the mass area image transmitted by the mass localization unit, the mass properties in the target patient's breast transmitted by the property detection unit, and the three-dimensional model of the breast area transmitted by the human body modeling unit, and construct a plane coordinate system in the breast magnetic resonance enhanced image and a spatial coordinate system of the three-dimensional model of the breast area, map the mass area and mass properties in the breast magnetic resonance enhanced image to corresponding positions in the three-dimensional model of the breast area based on a preset coordinate relationship table between the plane coordinate system and the spatial coordinate system, and transmit the corresponding positions in the three-dimensional model of the breast area to the dynamic rendering unit.

[0012] Preferably, the dynamic rendering unit is used to receive the corresponding position in the three-dimensional model of the breast area transmitted by the model matching unit, and dynamically render the corresponding position in the three-dimensional model of the breast area through a dynamic rendering engine to obtain a dynamic human body three-dimensional model, and transmit the dynamic human body three-dimensional model to the intelligent display unit.

[0013] Preferably, the intelligent display unit is used to receive the result of whether the target patient has a breast lump transmitted by the mass locating unit, the property of the lump in the target patient's breast transmitted by the property detection unit, and the dynamic human body three-dimensional model transmitted by the dynamic rendering unit, and display the dynamic human body three-dimensional model, the result of whether the target patient has a breast lump, and the property of the lump in the target patient's breast on a display terminal.

[0014] Preferably, the breast magnetic resonance image is enhanced to obtain a breast magnetic resonance enhanced image, comprising:

[0015] Acquire a rough illumination of the breast magnetic resonance image based on bilateral filtering, estimate an illumination component of the breast magnetic resonance image based on the rough illumination, and perform gamma correction on the illumination component of the breast magnetic resonance image to obtain a corrected illumination component of the breast magnetic resonance image;

[0016] Acquire a detail portion of the breast magnetic resonance image based on a difference between the breast magnetic resonance image and the rough illumination, and fuse the corrected illumination component and the detail portion of the breast magnetic resonance image based on an Alpha fusion algorithm to obtain a fused illumination component of the breast magnetic resonance image;

[0017] The breast magnetic resonance image is converted from the RGB color space to the HSV color space to obtain a breast magnetic resonance conversion image, the brightness channel in the breast magnetic resonance conversion image is replaced based on the fused illumination component of the breast magnetic resonance image to obtain a breast magnetic resonance conversion enhanced image, and the breast magnetic resonance conversion enhanced image is converted from the HSV color space to the RGB color space to obtain a breast magnetic resonance enhanced image.

[0018] Preferably, the regional coarse positioning model includes a feature extraction module, a feature fusion module and a detection module. The feature extraction module is used to extract a feature map from the breast magnetic resonance enhancement image, and transfer the feature map to the feature fusion module through continuous downsampling. The feature fusion module is used to fuse feature maps of different scales to obtain a fused feature map. The detection module is used to detect the fused feature map to obtain the location and category prediction results of the target mass.

[0019] Preferably, the mass detection model includes an image feature extraction module, a feature screening module and a feature classification module, the image feature extraction module is used to locate the shape of the mass in the mass area image based on SIFT, characterize the texture characteristics of the mass in the mass area image based on Gist, capture the edge information of the mass in the mass area image based on HOG, and characterize the texture change of the mass in the mass area image based on LBP, the feature screening module is used to perform feature screening on image features to obtain valid image features, and classify the valid image features based on support vector counting to obtain the properties of the mass in the mass area image, wherein the mass properties include normal properties and abnormal properties.

[0020] Preferably, the image features are screened, including:

[0021] Calculate the effective interval of the image feature on all categories of samples, where the calculation formula of the effective interval is as follows:

[0022]

[0023] Among them, R ij Indicates the effective interval, and Represents the image feature f i In category c j Upper effective interval R ij The lower and upper bounds of p i Represents category c in the dataset j The sample ratio, γ represents the preset ratio adjustment parameter, μ ij Represents the image feature f i In category c j The mean value on ij Represents the image feature f i In category c j The standard deviation on ;

[0024] The overlapping degree of the image features on all categories of samples is calculated, wherein the calculation formula of the overlapping degree is as follows:

[0025]

[0026] Among them, OAR i Represents the image feature f i The degree of overlap on all category samples, l represents the total number of all categories;

[0027] If the overlap degree of the image feature on all categories of samples is greater than a preset overlap threshold, the image feature is regarded as a valid image feature.

[0028] Preferably, the method for constructing the coordinate relationship table is as follows:

[0029] A transformation matrix between the plane coordinate system and the space coordinate system is established based on known calibration points, and a coordinate relationship table is obtained based on the transformation matrix, wherein the calculation formula of the transformation matrix is ​​as follows:

[0030]

[0031] Among them, (X i ,Y i ,Z i ) represents the plane coordinates of the known calibration points in the plane coordinate system, (x i ,y i ) represents the spatial coordinates of the known calibration points in the spatial coordinate system, and H represents the transformation matrix.

[0032] The beneficial effects of the present invention are as follows: (1) The present invention uses a deep learning model to perform multi-step processing such as enhancement, mass location, and property detection on breast magnetic resonance images, which can effectively improve the detection rate of breast masses and reduce missed diagnosis and misdiagnosis. In particular, through a trained mass detection model, the properties of the mass can be accurately located and analyzed, thereby providing doctors with a more reliable diagnosis basis; (2) The present invention combines the specific information of the breast mass with a three-dimensional human body model by combining the location and property detection of the mass, so that doctors and patients can intuitively see the three-dimensional reconstruction effect of the breast area. The use of dynamic rendering and three-dimensional modeling technology helps doctors fully understand the position and properties of the mass in the breast, which helps to formulate a more accurate treatment plan; (3) The present invention cooperates with an intelligent display unit and a dynamic rendering unit, so that the patient's breast health condition can be displayed in a real-time three-dimensional dynamic model, providing a more interactive and visual diagnostic tool, which is convenient for communication and exchange between doctors and patients. This visualization result helps patients understand their own condition more intuitively and also improves the diagnosis and treatment efficiency of doctors. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 A schematic diagram of the system architecture of an overall system in an embodiment of the present invention.

[0034] Figure numerals: 1. Magnetic resonance imaging unit; 2. Image enhancement unit; 3. Tumor localization unit; 4. Property detection unit; 5. Intelligent display unit; 6. Model matching unit; 7. Human body modeling unit; 8. Dynamic rendering unit. DETAILED DESCRIPTION

[0035] Embodiment 1, as Figure 1As shown, a simplified magnetic resonance breast intelligent recognition system based on deep learning proposed by the present invention includes a magnetic resonance imaging unit 1, an image enhancement unit 2, a mass positioning unit 3, a property detection unit 4, an intelligent display unit 5, a model matching unit 6, a human body modeling unit 7 and a dynamic rendering unit 8. Specifically:

[0036] The magnetic resonance imaging unit 1 is used to obtain a breast magnetic resonance image of a target patient based on a magnetic resonance imaging device, and transmit the breast magnetic resonance image to the image enhancement unit 2;

[0037] The image enhancement unit 2 is used to receive the breast magnetic resonance image transmitted by the magnetic resonance imaging unit 1, and acquire the breast magnetic resonance image of the target patient based on the magnetic resonance imaging device, perform image enhancement on the breast magnetic resonance image to obtain a breast magnetic resonance enhanced image, and transmit the breast magnetic resonance enhanced image to the mass localization unit 3;

[0038] The mass localization unit 3 is used to receive the breast magnetic resonance enhanced image transmitted by the image enhancement unit 2, and to perform regional localization on multiple mass areas in the breast magnetic resonance enhanced image based on the trained regional coarse localization model to obtain a mass area image, and to determine whether the target patient has a mass in the breast based on the mass area image, and to transmit the mass area image to the property detection unit 4 and the model matching unit 6, and to transmit the result of whether the target patient has a mass in the breast to the intelligent display unit 5.

[0039] In the present invention, magnetic resonance imaging (MRI) is a technology that uses strong magnetic fields and radio frequency waves for medical imaging. MRI can provide high-resolution soft tissue images and is widely used for imaging of the breast, brain, spine and other parts; mass localization refers to identifying areas where masses may exist in breast magnetic resonance images; property detection refers to further analysis of the located mass area to determine the nature of the mass.

[0040] In an optional embodiment, the property detection unit 4 is used to receive the mass area image transmitted by the mass localization unit 3, and perform property detection on the mass in the mass area image based on a trained mass detection model to obtain the mass property in the mass area image, determine the mass property in the target patient's breast based on the mass property in the mass area image, and transmit the mass property in the target patient's breast to the intelligent display unit 5 and the model matching unit 6.

[0041] It should be noted that the mass area image refers to the local image area containing the mass that has been extracted from the overall breast image; mass property detection refers to analyzing the mass characteristics in this mass area image to determine its nature, which includes benign and malignant. Benign means that the mass is usually cystic or fibrous and does not have the characteristics of a malignant tumor. Malignant means that the mass may be invasive, with characteristics such as irregular shape, blurred boundaries, and rapid growth.

[0042] In an optional embodiment, the human body modeling unit 7 is used to construct a three-dimensional model of the human body based on three-dimensional modeling technology, perform regional segmentation on the three-dimensional model of the human body to obtain a three-dimensional model of the breast region, and transmit the three-dimensional model of the breast region to the model matching unit 6.

[0043] It should be noted that 3D modeling technology is a technology that creates 3D object models through computer programs. The human body 3D model refers to a 3D model constructed based on the external surface contour of the human body (such as the skin layer) through 3D modeling technology. This type of model focuses on the external form of the human body, such as surface features such as the face, chest, and limbs.

[0044] In an optional embodiment, the model matching unit 6 is used to receive the mass area image transmitted by the mass localization unit 3, the mass properties in the target patient's breast transmitted by the property detection unit 4, and the three-dimensional model of the breast area transmitted by the human body modeling unit 7, and construct a plane coordinate system in the breast magnetic resonance enhanced image and a spatial coordinate system of the three-dimensional model of the breast area, and based on a preset coordinate relationship table between the plane coordinate system and the spatial coordinate system, map the mass area and mass properties in the breast magnetic resonance enhanced image to corresponding positions in the three-dimensional model of the breast area, and transmit the corresponding positions in the three-dimensional model of the breast area to the dynamic rendering unit 8.

[0045] In an optional embodiment, the dynamic rendering unit 8 is used to receive the corresponding position in the three-dimensional model of the breast area transmitted by the model matching unit 6, and dynamically render the corresponding position in the three-dimensional model of the breast area through the dynamic rendering engine to obtain a dynamic human body three-dimensional model, and transmit the dynamic human body three-dimensional model to the intelligent display unit 5.

[0046] It should be noted that a rendering engine is a computer program or algorithm used to generate images, especially in 3D graphics, animation or virtual reality. It converts this information into the final 2D image or animation based on the data of the 3D model and environmental parameters (such as lighting, material, camera angle, etc.).

[0047] In an optional embodiment, the intelligent display unit 5 is used to receive the result of whether the target patient has a lump in the breast transmitted by the lump localization unit 3, the property of the lump in the target patient's breast transmitted by the property detection unit 4, and the dynamic human body three-dimensional model transmitted by the dynamic rendering unit 8, and display the dynamic human body three-dimensional model, the result of whether the target patient has a lump in the breast, and the property of the lump in the target patient's breast in a display terminal.

[0048] It should be noted that the display terminal refers to a device or platform used to display images or data. The display terminal is a computer, medical display screen or other professional equipment used by doctors or medical personnel to display the three-dimensional model of the target patient's breast, the location of the lump and other diagnostic information.

[0049] Embodiment 2, a simplified magnetic resonance breast intelligent recognition system based on deep learning proposed by the present invention, compared with embodiment 1, this embodiment also includes: performing image enhancement on a breast magnetic resonance image to obtain a breast magnetic resonance enhanced image, including:

[0050] Acquiring a rough illumination of a breast magnetic resonance image based on bilateral filtering, estimating an illumination component of the breast magnetic resonance image based on the rough illumination, and performing gamma correction on the illumination component of the breast magnetic resonance image to obtain a corrected illumination component of the breast magnetic resonance image;

[0051] The detail part of the breast magnetic resonance image is obtained based on the difference between the breast magnetic resonance image and the rough illumination, and the corrected illumination component and the detail part of the breast magnetic resonance image are fused based on the Alpha fusion algorithm to obtain the fused illumination component of the breast magnetic resonance image;

[0052] The breast magnetic resonance image is converted from the RGB color space to the HSV color space to obtain a breast magnetic resonance conversion image, the brightness channel in the breast magnetic resonance conversion image is replaced based on the fused illumination component of the breast magnetic resonance image to obtain a breast magnetic resonance conversion enhanced image, and the breast magnetic resonance conversion enhanced image is converted from the HSV color space to the RGB color space to obtain a breast magnetic resonance enhanced image.

[0053] In this embodiment, bilateral filtering is an image processing technology used to remove noise while maintaining image edge details. It smoothes the image by considering the spatial distance of pixels and the difference in pixel values, thereby reducing noise while retaining image edge information; coarse illumination refers to low-frequency brightness information in an image, which usually indicates the overall illumination and background brightness of the image, but does not include details (such as edges, textures, etc.). When processing an image, coarse illumination helps to estimate the illumination distribution of the image; illumination component is the brightness part of the image, which indicates the illumination conditions of the image at a certain moment. The illumination component is usually low-frequency information in the image, which includes overall brightness changes but does not include details; gamma correction is an image adjustment technology used to correct the brightness of an image. In digital image processing, gamma correction can adjust the image brightness. The alpha fusion algorithm is an image fusion technology that generates a new image by combining different information of two images. It is used to fuse the corrected illumination component and detail part of breast magnetic resonance images. The RGB color space is a common color model that represents the color of an image based on a combination of three colors: red, green, and blue. The HSV color space is a color model based on hue, saturation, and value. In the HSV color space, the brightness channel (V) represents the overall brightness (or intensity) of the image, which is distinguished from hue (H) and saturation (S).

[0054] In an optional embodiment, the regional coarse positioning model includes a feature extraction module, a feature fusion module and a detection module. The feature extraction module is used to extract a feature map from the breast magnetic resonance enhancement image, and pass the feature map to the feature fusion module through continuous downsampling. The feature fusion module is used to fuse feature maps of different scales to obtain a fused feature map. The detection module is used to detect the fused feature map to obtain the location and category prediction results of the target mass.

[0055] In an optional embodiment, the mass detection model includes an image feature extraction module, a feature screening module and a feature classification module. The image feature extraction module is used to locate the shape of the mass in the mass area image based on SIFT, to characterize the texture characteristics of the mass in the mass area image based on Gist, to capture the edge information of the mass in the mass area image based on HOG, and to characterize the texture change of the mass in the mass area image based on LBP. The feature screening module is used to perform feature screening on image features to obtain valid image features, and to classify the valid image features based on support vector counting to obtain the properties of the mass in the mass area image, wherein the mass properties include normal properties and abnormal properties.

[0056] It should be noted that SIFT is a commonly used image feature extraction method, which aims to extract local feature points in the image. These feature points are invariant to changes in image scaling, rotation, translation, etc. The features extracted by SIFT can help locate the mass area; Gist is a feature description method based on global information of the image, which can capture the overall texture characteristics of the image; HOG is an algorithm that describes the local shape characteristics of the image, which is usually used for object detection, especially edge detection; LBP is a texture descriptor, which is often used to describe the texture changes of the image. It generates a binary value by comparing each pixel of the image with the pixels in its neighborhood, and then uses these binary values ​​as the texture features of the image; Support Vector Machine (SVM) is a supervised learning model, which is often used in classification and regression analysis. SVM achieves efficient classification by constructing a hyperplane to maximize the boundary between categories.

[0057] In an optional embodiment, performing feature screening on image features includes:

[0058] Calculate the effective interval of the image features on all categories of samples, where the calculation formula of the effective interval is as follows:

[0059]

[0060] Among them, R ij Indicates the effective interval, and Represents the image feature f i In category c j Upper effective interval R ij The lower and upper bounds of p i Represents category c in the dataset j The sample ratio, γ represents the preset ratio adjustment parameter, μ ij Represents the image feature f i In category c j The mean value on ij Represents the image feature f i In category c j The standard deviation on ;

[0061] Calculate the degree of overlap of image features on all category samples, where the calculation formula for the degree of overlap is as follows:

[0062]

[0063] Among them, OAR i Represents the image feature f i The degree of overlap on all category samples, l represents the total number of all categories;

[0064] If the overlap degree of image features on all category samples is greater than the preset overlap threshold, the image features are regarded as valid image features.

[0065] In an optional embodiment, the method for constructing the coordinate relationship table is as follows:

[0066] Based on the known calibration points, a transformation matrix between the plane coordinate system and the space coordinate system is established, and a coordinate relationship table is obtained based on the transformation matrix. The calculation formula of the transformation matrix is ​​as follows:

[0067]

[0068] Among them, (X i ,Y i ,Z i ) represents the plane coordinates of the known calibration points in the plane coordinate system, (x i ,y i ) represents the spatial coordinates of the known calibration points in the spatial coordinate system, and H represents the transformation matrix.

[0069] The embodiments of the present invention are described in detail above with reference to the accompanying drawings, but the present invention is not limited thereto, and various changes can be made within the knowledge scope of technicians in the relevant technical field without departing from the purpose of the present invention.

Claims

1. A simplified magnetic resonance breast intelligent recognition system based on deep learning, characterized in that: The system comprises a magnetic resonance imaging unit (1), an image enhancement unit (2), a tumor localization unit (3), a property detection unit (4), an intelligent display unit (5), a model matching unit (6), a human body modeling unit (7) and a dynamic rendering unit (8), specifically: The magnetic resonance imaging unit (1) is used to obtain a breast magnetic resonance image of a target patient based on a magnetic resonance imaging device, and transmit the breast magnetic resonance image to an image enhancement unit (2); The image enhancement unit (2) is used to receive the breast magnetic resonance image transmitted by the magnetic resonance imaging unit (1), and acquire the breast magnetic resonance image of the target patient based on the magnetic resonance imaging device, perform image enhancement on the breast magnetic resonance image to obtain a breast magnetic resonance enhanced image, and transmit the breast magnetic resonance enhanced image to the mass localization unit (3); The mass localization unit (3) is used to receive the breast magnetic resonance enhanced image transmitted by the image enhancement unit (2), and to perform regional localization on multiple mass regions in the breast magnetic resonance enhanced image based on a trained regional coarse localization model to obtain a mass region image, and to determine whether the target patient has a mass in the breast based on the mass region image, and to transmit the mass region image to the property detection unit (4) and the model matching unit (6), and to transmit the result of whether the target patient has a mass in the breast to the intelligent display unit (5).

2. According to the simplified magnetic resonance breast intelligent recognition system based on deep learning in claim 1, it is characterized in that: The property detection unit (4) is used to receive the mass area image transmitted by the mass positioning unit (3), and perform property detection on the mass in the mass area image based on the trained mass detection model to obtain the mass property in the mass area image, determine the mass property in the breast of the target patient based on the mass property in the mass area image, and transmit the mass property in the breast of the target patient to the intelligent display unit (5) and the model matching unit (6); The human body modeling unit (7) is used to construct a three-dimensional human body model based on three-dimensional modeling technology, perform regional segmentation on the three-dimensional human body model to obtain a three-dimensional model of the breast region, and transmit the three-dimensional model of the breast region to the model matching unit (6).

3. The simplified magnetic resonance breast intelligent recognition system based on deep learning according to claim 2 is characterized in that: The model matching unit (6) is used to receive the mass area image transmitted by the mass positioning unit (3), the mass properties in the target patient's breast transmitted by the property detection unit (4), and the three-dimensional model of the breast area transmitted by the human body modeling unit (7), and to construct a plane coordinate system in the breast magnetic resonance enhanced image and a space coordinate system of the three-dimensional model of the breast area, and based on a preset coordinate relationship table between the plane coordinate system and the space coordinate system, map the mass area and mass properties in the breast magnetic resonance enhanced image to corresponding positions in the three-dimensional model of the breast area, and transmit the corresponding positions in the three-dimensional model of the breast area to the dynamic rendering unit (8).

4. The simplified magnetic resonance breast intelligent recognition system based on deep learning according to claim 3 is characterized in that: The dynamic rendering unit (8) is used to receive the corresponding position in the three-dimensional model of the breast region transmitted by the model matching unit (6), and dynamically render the corresponding position in the three-dimensional model of the breast region through a dynamic rendering engine to obtain a dynamic three-dimensional human body model, and transmit the dynamic three-dimensional human body model to the intelligent display unit (5); The intelligent display unit (5) is used to receive the result of whether the target patient has a breast lump transmitted by the lump locating unit (3), the property of the lump in the target patient's breast transmitted by the property detecting unit (4), and the dynamic human body three-dimensional model transmitted by the dynamic rendering unit (8), and to display the dynamic human body three-dimensional model, the result of whether the target patient has a breast lump, and the property of the lump in the target patient's breast on a display terminal.

5. The simplified magnetic resonance breast intelligent recognition system based on deep learning according to claim 1 is characterized in that: The breast magnetic resonance image is enhanced to obtain a breast magnetic resonance enhanced image, comprising: Acquire a rough illumination of the breast magnetic resonance image based on bilateral filtering, estimate an illumination component of the breast magnetic resonance image based on the rough illumination, and perform gamma correction on the illumination component of the breast magnetic resonance image to obtain a corrected illumination component of the breast magnetic resonance image; Acquire a detail portion of the breast magnetic resonance image based on a difference between the breast magnetic resonance image and the rough illumination, and fuse the corrected illumination component and the detail portion of the breast magnetic resonance image based on an Alpha fusion algorithm to obtain a fused illumination component of the breast magnetic resonance image; The breast magnetic resonance image is converted from the RGB color space to the HSV color space to obtain a breast magnetic resonance conversion image, the brightness channel in the breast magnetic resonance conversion image is replaced based on the fused illumination component of the breast magnetic resonance image to obtain a breast magnetic resonance conversion enhanced image, and the breast magnetic resonance conversion enhanced image is converted from the HSV color space to the RGB color space to obtain a breast magnetic resonance enhanced image.

6. The simplified magnetic resonance breast intelligent recognition system based on deep learning according to claim 1, characterized in that: The regional coarse positioning model includes a feature extraction module, a feature fusion module and a detection module. The feature extraction module is used to extract a feature map from the breast magnetic resonance enhancement image, and transfer the feature map to the feature fusion module through continuous downsampling. The feature fusion module is used to fuse feature maps of different scales to obtain a fused feature map. The detection module is used to detect the fused feature map to obtain the location and category prediction results of the target mass.

7. The simplified magnetic resonance breast intelligent recognition system based on deep learning according to claim 2, characterized in that: The mass detection model includes an image feature extraction module, a feature screening module and a feature classification module. The image feature extraction module is used to locate the shape of the mass in the mass area image based on SIFT, to characterize the texture characteristics of the mass in the mass area image based on Gist, to capture the edge information of the mass in the mass area image based on HOG, and to characterize the texture change of the mass in the mass area image based on LBP. The feature screening module is used to perform feature screening on image features to obtain valid image features, and to classify the valid image features based on support vector counting to obtain the properties of the mass in the mass area image, wherein the mass properties include normal properties and abnormal properties.

8. The simplified magnetic resonance breast intelligent recognition system based on deep learning according to claim 7, characterized in that: Perform feature screening on image features, including: Calculate the effective interval of the image feature on all categories of samples, where the calculation formula of the effective interval is as follows: Among them, R ij Indicates the effective interval, and Represents the image feature f i In category c j Upper effective interval R ij The lower and upper bounds of p i Represents category c in the dataset j The sample ratio, γ represents the preset ratio adjustment parameter, μ ij Represents the image feature f i In category c j The mean value on ij Represents the image feature f i In category c j The standard deviation on ; The overlapping degree of the image features on all categories of samples is calculated, wherein the calculation formula of the overlapping degree is as follows: Among them, OAR i Represents the image feature f i The degree of overlap on all category samples, l represents the total number of all categories; If the overlap degree of the image feature on all categories of samples is greater than a preset overlap threshold, the image feature is regarded as a valid image feature.

9. The simplified magnetic resonance breast intelligent recognition system based on deep learning according to claim 3, characterized in that: The method for constructing the coordinate relationship table is as follows: A transformation matrix between the plane coordinate system and the space coordinate system is established based on known calibration points, and a coordinate relationship table is obtained based on the transformation matrix, wherein the calculation formula of the transformation matrix is ​​as follows: Among them, (X i ,Y i ,Z i ) represents the plane coordinates of the known calibration points in the plane coordinate system, (x i ,y i ) represents the spatial coordinates of the known calibration points in the spatial coordinate system, and H represents the transformation matrix.

Images