CN112053325A - Breast mass image processing and classifying system - Google Patents

Breast mass image processing and classifying system Download PDF

Info

Publication number
CN112053325A
CN112053325A CN202010805337.5A CN202010805337A CN112053325A CN 112053325 A CN112053325 A CN 112053325A CN 202010805337 A CN202010805337 A CN 202010805337A CN 112053325 A CN112053325 A CN 112053325A
Authority
CN
China
Prior art keywords
image
breast
mass
image processing
transformation
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202010805337.5A
Other languages
Chinese (zh)
Inventor
王杉
胡艺莹
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
East China Jiaotong University
Original Assignee
East China Jiaotong University
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by East China Jiaotong University filed Critical East China Jiaotong University
Priority to CN202010805337.5A priority Critical patent/CN112053325A/en
Publication of CN112053325A publication Critical patent/CN112053325A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/0002—Inspection of images, e.g. flaw detection
    • G06T7/0012—Biomedical image inspection
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06T3/18—
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T3/00—Geometric image transformation in the plane of the image
    • G06T3/40—Scaling the whole image or part thereof
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T3/00—Geometric image transformation in the plane of the image
    • G06T3/60—Rotation of a whole image or part thereof
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T5/00—Image enhancement or restoration
    • G06T5/40—Image enhancement or restoration by the use of histogram techniques
    • G06T5/70—
    • G06T5/90—
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/10—Image acquisition modality
    • G06T2207/10116—X-ray image
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20081—Training; Learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20084—Artificial neural networks [ANN]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/30—Subject of image; Context of image processing
    • G06T2207/30004—Biomedical image processing
    • G06T2207/30068—Mammography; Breast
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V2201/00—Indexing scheme relating to image or video recognition or understanding
    • G06V2201/03—Recognition of patterns in medical or anatomical images
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V2201/00—Indexing scheme relating to image or video recognition or understanding
    • G06V2201/07—Target detection

Abstract

The invention relates to a breast lump image processing and classifying system, which comprises: a mammography machine for taking mammographic images; the image preprocessing module is used for preprocessing the mammographic image; a breast region extraction module for extracting a breast region image in the preprocessed mammographic image; a mass detection and classification module for DIoU-YOLOv3 target detection network to perform mass location detection and benign-malignant classification on the breast region images. The breast tumor image processing and classifying system can automatically detect and classify the breast tumor by adopting the target detection network, so that the breast tumor image processing and classifying system is high in speed, efficiency and accuracy compared with manual radiograph reading.

Description

Breast mass image processing and classifying system
Technical Field
The present invention relates to image processing and classification techniques, and more particularly, to a breast mass image processing and classification system.
Background
Relevant data show that breast cancer is leading in the incidence of female malignancies both in developed and developing countries worldwide. Therefore, early detection of breast cancer is critical to reducing mortality in women. Among the current diagnostic methods for breast diseases, digital mammography is the most reliable screening method for early detection of suspicious breast masses and microcalcifications. In diagnosing breast abnormalities, a radiologist classifies a suspicious mass as benign or malignant. However, only a great deal of experience has been accumulated to make a more accurate determination of mammograms. More importantly, the existing diagnosis relies mainly on the subjective judgment of the radiologist, which requires the radiologist to have a high concentration capability. However, long-time, overloaded work inevitably causes visual fatigue and psychological fatigue. In addition, the mammography is projection imaging, dense tissues of glands easily block tumors, and asian women have high gland density ratio and are easy to misdiagnose and leak diagnose.
Therefore, in order to solve the problems of low efficiency, heavy workload of doctors and low accuracy and specificity of breast mass image processing and classification in the prior art, a breast mass image processing and classification system with higher accuracy and high diagnosis speed is needed.
Disclosure of Invention
The technical problem to be solved by the present invention is to provide a breast mass image processing and classifying system with higher accuracy and high diagnosis speed, aiming at the above defects of the prior art.
The technical scheme adopted by the invention for solving the technical problems is as follows: constructing a breast mass image processing and classification system comprising:
a mammography machine for taking mammographic images;
the image preprocessing module is used for preprocessing the mammographic image;
a breast region extraction module for extracting a breast region image in the preprocessed mammographic image;
a mass detection and classification module for DIoU-YOLOv3 target detection network to perform mass location detection and benign-malignant classification on the breast region images.
In the breast mass image processing and classifying system of the present invention, the image preprocessing module includes:
the image contrast enhancement unit is used for converting the mammographic image into a gray-scale image, performing histogram equalization on the gray-scale image, and converting the equalized image into a color RGB image;
and the image data enhancement unit is used for carrying out image data expansion transformation on the color RGB image so as to obtain the preprocessed mammographic image.
In the breast mass image processing and classification system of the present invention, the image data augmentation transformation includes: x-axis translation transformation, deformation transformation, vertical flip transformation, Y-axis miscut transformation, left-right flip transformation, rotation transformation, X-axis miscut transformation, Gaussian blur transformation or scaling transformation.
In the breast mass image processing and classifying system of the present invention, the breast region extracting module further includes:
a detection unit for inputting the preprocessed mammographic image into a YOLOv3 target detection network for breast detection to obtain breast position coordinates;
a cropping unit for cropping the preprocessed mammographic image according to the breast position coordinates to obtain the breast area image.
In the breast mass image processing and classification system of the present invention, the YOLOv3 target detection network is the YOLOv3-tiny network, and the YOLOv3-tiny network only includes two yolo layers with sizes of 13 × 13 and 26 × 26, respectively.
In the breast tumor image processing and classification system of the present invention, the tumor detection and classification module comprises:
a DIoU-YOLOv3 target detection network, wherein the loss function of the DIoU-YOLOv3 target detection network comprises a coordinate regression loss DIoU, a confidence coefficient loss and a category loss, and the confidence coefficient loss and the category loss are binary cross entropies;
a training module for training the DIoU-yollov 3 target detection network based on standard mammographic images, mass location coordinates, and mass benign-malignant labels;
a detection module to classify the breast region images based on the trained DIoU-YOLOv3 target detection network.
In the breast mass image processing and classification system of the present invention, the coordinate regression loss DIoU is expressed as:
Figure BDA0002628742410000031
wherein, bgtCenter points of the prediction frame and the real frame are respectively represented, ρ represents a euclidean distance between the two center points, c represents a diagonal distance of a minimum closure area containing both the prediction frame and the real frame, and IoU represents an intersection ratio.
In the breast mass image processing and classification system of the present invention, the confidence loss is expressed as:
Figure BDA0002628742410000032
s denotes the number of prediction cells into which the image is split,
Figure BDA0002628742410000033
the prediction unit cell representing the ith row and the jth column contains the center of a real object,
Figure BDA0002628742410000034
the cell representing the ith row and the jth column does not contain the center of the real object,
Figure BDA0002628742410000037
indicates the confidence, λnoobjRepresents a weight coefficient, and B represents a bounding box.
The breast lump image processing method of the inventionAnd in the classification system, the weight coefficient lambdanoobjIs 0.5.
In the breast mass image processing and classifying system of the present invention, the class loss is:
Figure BDA0002628742410000035
Pi jrepresenting the probability of classification.
The breast tumor image processing and classifying system can automatically detect and classify the breast tumor by adopting the target detection network, so that the breast tumor image processing and classifying system is high in speed, efficiency and accuracy compared with manual radiograph reading. Further, the detection and the benign and malignant classification of the breast tumor are realized by adopting a target detection network (DIoU-YOLOv3 target detection network) formed by an improved frame regression loss function, and the accuracy of the tumor detection frame in the mammography image is greatly improved.
Drawings
The invention will be further described with reference to the accompanying drawings and examples, in which:
FIG. 1 is a schematic block diagram of a first preferred embodiment of the present invention;
FIG. 2 is a functional block diagram of a second preferred embodiment of the present invention;
FIG. 3 is a schematic diagram of the structure of the YOLOv3-tiny network according to the preferred embodiment shown in FIG. 2;
FIGS. 4A-4B are schematic diagrams of breast detection results obtained from the YOLOv3-tiny network according to the preferred embodiment shown in FIG. 2;
FIG. 5 shows a schematic diagram of the structure of the DIoU-YOLOv3 target detection network according to the preferred embodiment shown in FIG. 2;
FIG. 6 is a schematic diagram of the convolution of the DIoU-YOLOv3 target detection network shown in FIG. 5;
fig. 7A-7B are schematic diagrams of breast detection results obtained by the DIoU-YOLOv3 target detection network according to the preferred embodiment shown in fig. 2.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is described in further detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.
Fig. 1 is a schematic block diagram of a first preferred embodiment of the present invention. As shown in fig. 1, the breast mass image processing and classifying system of the present invention includes: a mammography X-ray machine 100, an image pre-processing module 200, a breast region extraction module 300, and a tumor detection and classification module 400. The mammography X-ray machine 100, also called a molybdenum target X-ray machine, is mainly used for mammography examination of female mammary glands, is a basic mammography examination and diagnosis device in gynecology and specialized hospitals of the present hospitals, can find masses and micro calcifications in mammary tissues in time, can also be used for photography of non-metallic foreign bodies and other soft tissues such as hemangioma, and is used for taking mammography images. The image preprocessing module 200 may receive the mammographic image from the mammography machine 100 through a wired or wireless communication device and preprocess it. Any image pre-processing method known in the art may be used with the present invention, such as contrast enhancement, color change, or image transformation, among others. The breast region extraction module 300 is configured to extract a breast region image in the preprocessed mammographic image. Here, the breast area image in the mammographic image may be selected by any known means, and it is preferable that, for example, the preprocessed mammographic image is input to a YOLOv3 target detection network for breast detection to obtain breast position coordinates, and then the preprocessed mammographic image is clipped according to the breast position coordinates to obtain the breast area image. The mass detection and classification module 400 is used in the DIoU-YOLOv3 target detection network for mass location detection and benign-malignant classification of the breast region images.
Those skilled in the art will appreciate that in the present invention, the image preprocessing module 200, the breast area extraction module 300 and the mass detection and classification module 400 can be implemented using any circuit, module, software known in the art, hardware, software, or a combination of hardware and software, in a centralized fashion in at least one computer system, or in a distributed fashion across different portions of several interconnected computer systems. Any kind of computer system or other apparatus adapted for carrying out the methods of the present invention is suited. A typical combination of hardware and software could be a general purpose computer system with a computer program that, when being loaded and executed, controls the computer system such that it carries out the methods described herein. The computer program in this document refers to: any expression, in any programming language, code or notation, of a set of instructions intended to cause a system having an information processing capability to perform a particular function either directly or after either or both of the following: a) conversion to other languages, codes or symbols; b) reproduced in a different format.
The breast tumor image processing and classifying system can automatically detect and classify the breast tumor by adopting the target detection network, so that the breast tumor image processing and classifying system is high in speed, efficiency and accuracy compared with manual radiograph reading.
Fig. 2 is a schematic block diagram of a second preferred embodiment of the present invention. As shown in fig. 2, the breast mass image processing and classifying system of the present invention includes: a mammography X-ray machine 100, an image pre-processing module 200, a breast region extraction module 300, and a tumor detection and classification module 400. The mammography X-ray machine 100, also called a molybdenum target X-ray machine, is mainly used for mammography examination of female mammary glands, is a basic mammography examination and diagnosis device in gynecology and specialized hospitals of the present hospitals, can find masses and micro calcifications in mammary tissues in time, can also be used for photography of non-metallic foreign bodies and other soft tissues such as hemangioma, and is used for taking mammography images.
The image preprocessing module 200 may receive the mammographic image from the mammography machine 100 through a wired or wireless communication device and preprocess it. In this embodiment, the image preprocessing module 200 includes an image contrast enhancement unit 210 and an image data enhancement unit 220. The image contrast enhancement unit 210 is configured to convert the mammographic image into a gray-scale image, perform histogram equalization on the gray-scale image, increase the contrast between the breast tissue and the surrounding background, and then convert the equalized image into a color RGB image. Histogram equalization is a simple and effective image enhancement technique, which changes the gray scale of each pixel in an image by changing the histogram of the image, and is mainly used for enhancing the contrast of the image with a small dynamic range. The basic principle of histogram equalization is: the gray values with a large number of pixels (namely, the gray values which mainly act on the picture) in the image are widened, and the gray values with a small number of pixels (namely, the gray values which do not mainly act on the picture) are merged, so that the contrast is increased, the image is clear, and the aim of enhancement is fulfilled. The present embodiment thus utilizes histogram equalization to enhance the contrast of breast tissue, which results in a convenient subsequent detection and classification of breast masses.
The image data enhancing unit 220 is configured to perform image data expansion transformation on the color RGB image to obtain the preprocessed mammographic image. In a preferred embodiment of the present invention, the image data expansion transformation comprises: an X-axis translation (translex) transform, a deformation (elastic transformation) transform, a vertical flip (flipd) transform, a Y-axis miscut (ShearY) transform, a left-right flip (Fliplr) transform, a rotation (Rotate) transform, an X-axis miscut (ShearX) transform, a gaussian blur (gaussian blur) transform, or a Scale (Scale) transform. One skilled in the art will recognize that any one or more of the variations described above may be used. It is also preferable that the small sample data set can be expanded to 5 times the original, as shown in table 1, which is the data enhancement strategy of this embodiment, P represents the probability of using the transformation:
transformation operation 1 P Transformation operation 2 P
Sub-strategy 1 Translation of the X axis 0.6 Zoom 0.6
Sub-strategy 2 Deformation of 0.5 Vertically flipped 0.5
Sub-strategy 3 Y-axis miscut 1 Turn over from top to bottom 0.5
Sub-policy 4 Rotate 0.8 Cutting in X axis 1
Sub-strategy 5 Gaussian blur 0.6 Zoom 0.6
The data enhancement method for the breast molybdenum target X-ray photographic image adopted by the embodiment can perform data enhancement on an original small sample data set, improve the generalization performance of subsequent breast mass detection, and reduce the possibility of model overfitting.
Of course, it is understood by those skilled in the art that in a simplified embodiment of the present invention, the image pre-processing module 200 may include only one of the image contrast enhancement unit 210 and the image data enhancement unit 220.
In the present embodiment, the breast area extraction module 300 includes a detection unit 310 and a cropping unit 320. The detection unit 310 is used for inputting the preprocessed mammographic image into a YOLOv3 target detection network for breast detection to obtain breast position coordinates. The cropping unit 320 is configured to crop the preprocessed mammographic image according to the breast position coordinates to obtain the breast region image. Preferably, the YOLOv3 target detection network is a YOLOv3-tiny network, and the YOLOv3-tiny network replaces the YOLOv3 original backbone network Darknet-53 with tiny network, considering that the proportion of the breast to the original image is large and the detection task is simple, the tiny network is adopted to realize breast detection, and under the condition of ensuring the accuracy, the detection speed is also improved, and the detection time is reduced. The tiny network is adopted as a backbone network, the total number of the tiny network is only 24 layers, the tiny network mainly comprises a convolutional layer and a maximum pooling layer, only two yolo layers are arranged, and the sizes of the yolo layers are 13 multiplied by 13 and 26 multiplied by 26 respectively. Table 2 shows the parameter schematic of each layer in the YOLOv3-tiny network of this embodiment.
Figure BDA0002628742410000071
Figure BDA0002628742410000081
FIG. 3 is a schematic diagram of the structure of the YOLOv3-tiny network according to the preferred embodiment. As can be seen in the figure, no residual layer is used, only two different scale yolo output layers are used. Further, when detecting the breast position, the real anchor frame of the breast position in the training set is clustered by adopting a K-means (K-mean) clustering algorithm before training. The results of 6 cluster centers obtained were as follows:
[528,1589 675,1870 797,1989 829,2550 931,2162 1099,2653]
according to the method and the device, irrelevant background can be removed from the original mammary molybdenum target X-ray photographic image, an image which is convenient for subsequent detection of the lump in the breast is obtained, and the processing efficiency can be improved. The breast test results are shown in FIGS. 4A-4B.
Further, in the preferred embodiment shown in fig. 2, the mass detection and classification module 400 includes: DIoU-YOLOv3 target detection network 310, training module 320, and detection module 330. In the preferred embodiment, the YOLOv3 target detection framework is used to detect and classify breast masses, and the YOLOv3 adopts a multi-scale prediction method to enhance the detection capability of small targets, so that the masses occupy a small area relative to the breast area and have different sizes, so that more abundant high-level semantic information can be obtained by using the YOLOv3, and the breast mass detection task is better. Fig. 5 shows a schematic structure diagram of the DIoU-yollov 3 object detection network 310. As shown in fig. 5
As shown in fig. 5, which is a schematic diagram of the target detection network structure of this embodiment, it can be seen that the entire target detection network includes a backbone network Darknet-53 on the left side and three YOLO layers on the right side, where residual is a residual structure, as shown in fig. 6, the structure is obtained by superimposing a feature map generated after convolution with an input, and the superimposed feature map is transmitted to the next layer as a new input. The YOLO main body is composed of a plurality of residual modules, the risk of gradient explosion is reduced, and the learning capacity of the network is enhanced. Further, the backbone network employs a Darknet-53 network, in which 53 convolutional layers are provided, which are composed of a series of 1x1 and 3x3 convolutional layers (each followed by a BN (batch normalization) layer and a LeakyReLU) layer. The sizes of the three YOLO layers are respectively 13 multiplied by 13, 26 multiplied by 26 and 52 multiplied by 52, a feature fusion method is adopted in the YOLO layers, the feature graphs obtained by the last three residual modules in the network are subjected to up-sampling operation, and three-scale feature fusion is carried out. In addition, each YOLO layer corresponds to 3 anchor frames, there are 9 anchor frames in total, the anchor frames are obtained by clustering according to the lump positions of the breast molybdenum target radiographic image only containing breasts by adopting a k-means algorithm, and the 9 clustering centers are as follows:
[100,112 142,141 145,209 177,174 200,224 245,192 276,274 336,381 473,586]
further, this embodiment replaces the original MSE (mean square error) with DIoU (Distance-IoU) as the regression loss function of the frame, and the expression of the DIoU loss is as follows:
Figure BDA0002628742410000091
wherein, bgtRespectively representing the central points of the prediction frame and the real frame, rho represents the Euclidean distance between the two central points, c represents the diagonal distance of the minimum closure area simultaneously containing the prediction frame and the real frame, IoU represents the intersection ratio, and the value of the intersection ratio is 0-1. The purpose of such improvement is: the DIoU is more in line with a target frame regression mechanism than the GIou (Generalized-IoU), the distance, the overlapping rate and the scale between the target and the anchor are taken into consideration, so that the target frame regression becomes more stable, the problems of divergence and the like in the training process like IoU and the GIoU are avoided, the frame regression can be more accurate, and the accurate positioning of the tumor position is facilitated.
In this embodiment, the loss function of the DIoU-YOLOv3 target detection network 310 is divided into three parts: the coordinate regression loss adopts DIoU, the confidence coefficient loss and the category loss both adopt binary cross entropy, wherein the confidence coefficient loss is expressed as:
Figure BDA0002628742410000092
at the time of detection, a picture (a standard picture for training or a detection picture for detection) is detectedDividing into S multiplied by S prediction unit cells, wherein S is a positive integer larger than 0, and i and j are respectively positive integers larger than 0 and smaller than S.
Figure BDA0002628742410000101
The cell representing the ith row and the jth column contains the center of a real object,
Figure BDA0002628742410000102
the cells representing the ith row and the jth column do not contain the center of a real object, and the value is 1 or 0.
Figure BDA0002628742410000103
And representing the confidence coefficient, wherein the value of the confidence coefficient is 0-1, and B represents a frame. The confidence loss function is divided into two terms: the first term indicates the presence of an object and the second term indicates the absence of an object, wherein the absence of an object-missing part also increases the weighting factor λnoobj. The reason for adding weight coefficients is that for an image, in general, most of the content does not contain the object to be detected, which results in that the calculated part without object contributes more than the calculated part with object, which results in that the network tends to predict that the cell does not contain an object. Therefore, we want to reduce the contribution weight of the part without object calculation, such as the weight coefficient λnoobjThe values are as follows: 0.5.
the classification loss is expressed as:
Figure BDA0002628742410000104
Pi jrepresenting the probability of classification.
When training the target detection network, inputting a data format comprising: in standard mammographic images (. JPG format), tumor location coordinates (X)min,Ymin,Xmax,Ymax) And the benign-malignant label of the tumor (0 for benign and 1 for malignant), all breast molybdenum target X-ray images with tumor used in the training set were labeled by professional radiologists.
At the time of detection, the breast area images are classified based on the trained DIoU-yollov 3 target detection network arrangement. In the embodiment, the target detection network formed by the improved frame regression loss function is adopted to realize the detection and benign and malignant classification of the breast tumor, and the accuracy of the tumor detection frame in the X-ray image of the breast molybdenum target gland is greatly improved. The tumor mass test results are shown in FIGS. 7A-7B.
According to the method, the mammary molybdenum target X-ray photographic image is obtained, firstly, the mammary molybdenum target X-ray photographic image is preprocessed, then, the processed image is subjected to image enhancement, the enhanced image is input into a breast detection network to carry out breast extraction, meanwhile, irrelevant backgrounds are removed, an image only containing breasts is obtained, the image is input into a mammary tumor detection and benign and malignant classification network to carry out tumor identification, and the final detection and classification result is obtained.
Accordingly, the present invention can be realized in hardware, software, or a combination of hardware and software. The present invention can be realized in a centralized fashion in at least one computer system, or in a distributed fashion where different elements are spread across several interconnected computer systems. Any kind of computer system or other apparatus adapted for carrying out the methods of the present invention is suited. A typical combination of hardware and software could be a general purpose computer system with a computer program that, when being loaded and executed, controls the computer system such that it carries out the methods described herein. The computer program in this document refers to: any expression, in any programming language, code or notation, of a set of instructions intended to cause a system having an information processing capability to perform a particular function either directly or after either or both of the following: a) conversion to other languages, codes or symbols; b) reproduced in a different format.
While the invention has been described with reference to specific embodiments, it will be understood by those skilled in the art that various changes may be made and equivalents may be substituted without departing from the scope of the invention. In addition, many modifications may be made to adapt a particular situation or material to the teachings of the invention without departing from its scope. Therefore, it is intended that the invention not be limited to the particular embodiment disclosed, but that the invention will include all embodiments falling within the scope of the appended claims.
The above description is only for the purpose of illustrating the preferred embodiments of the present invention and is not to be construed as limiting the invention, and any modifications, equivalents and improvements made within the spirit and principle of the present invention are intended to be included within the scope of the present invention.

Claims (10)

1. A breast mass image processing and classification system, comprising:
a mammography machine for taking mammographic images;
the image preprocessing module is used for preprocessing the mammographic image;
a breast region extraction module for extracting a breast region image in the preprocessed mammographic image;
a mass detection and classification module for DIoU-YOLOv3 target detection network to perform mass location detection and benign-malignant classification on the breast region images.
2. The breast mass image processing and classification system of claim 1 wherein the image pre-processing module comprises:
the image contrast enhancement unit is used for converting the mammographic image into a gray-scale image, performing histogram equalization on the gray-scale image, and converting the equalized image into a color RGB image;
and the image data enhancement unit is used for carrying out image data expansion transformation on the color RGB image so as to obtain the preprocessed mammographic image.
3. The breast mass image processing and classification system of claim 2 wherein the image data augmentation transformation comprises: x-axis translation transformation, deformation transformation, vertical flip transformation, Y-axis miscut transformation, left-right flip transformation, rotation transformation, X-axis miscut transformation, Gaussian blur transformation or scaling transformation.
4. The breast mass image processing and classification system according to any of claims 1-3, wherein the breast region extraction module further comprises:
a detection unit for inputting the preprocessed mammographic image into a YOLOv3 target detection network for breast detection to obtain breast position coordinates;
a cropping unit for cropping the preprocessed mammographic image according to the breast position coordinates to obtain the breast area image.
5. The breast mass image processing and classification system of claim 4, wherein the YOLOv3 target detection network is the YOLOv3-tiny network, the YOLOv3-tiny network comprising only two yolo layers, of sizes 13 x 13 and 26 x 26, respectively.
6. The breast mass image processing and classification system of claim 4 wherein the mass detection and classification module includes:
a DIoU-YOLOv3 target detection network, wherein the loss function of the DIoU-YOLOv3 target detection network comprises a coordinate regression loss DIoU, a confidence coefficient loss and a category loss, and the confidence coefficient loss and the category loss are binary cross entropies;
a training module for training the DIoU-yollov 3 target detection network based on standard mammographic images, mass location coordinates, and mass benign-malignant labels;
a detection module to classify the breast region images based on the trained DIoU-YOLOv3 target detection network.
7. The breast mass image processing and classification system of claim 6 wherein the coordinate regression loss DIoU is expressed as:
Figure FDA0002628742400000021
wherein, bgtCenter points of the prediction frame and the real frame are respectively represented, ρ represents a euclidean distance between the two center points, c represents a diagonal distance of a minimum closure area containing both the prediction frame and the real frame, and IoU represents an intersection ratio.
8. The breast mass image processing and classification system of claim 7 wherein the confidence loss is expressed as:
Figure FDA0002628742400000026
Figure FDA0002628742400000022
s denotes the number of prediction cells into which the image is split,
Figure FDA0002628742400000023
the prediction unit cell representing the ith row and the jth column contains the center of a real object,
Figure FDA0002628742400000024
the cell representing the ith row and the jth column does not contain the center of the real object,
Figure FDA0002628742400000025
indicates the confidence, λnoobjRepresents a weight coefficient, and B represents a bounding box.
9. The breast mass image processing and classification system as claimed in claim 8, wherein the weighting coefficient λnoobjIs 0.5.
10. The breast mass image processing and classification system of claim 8 wherein the class penalty is:
Figure FDA0002628742400000031
Pi jrepresenting the probability of classification.
CN202010805337.5A 2020-08-12 2020-08-12 Breast mass image processing and classifying system Pending CN112053325A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010805337.5A CN112053325A (en) 2020-08-12 2020-08-12 Breast mass image processing and classifying system

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010805337.5A CN112053325A (en) 2020-08-12 2020-08-12 Breast mass image processing and classifying system

Publications (1)

Publication Number Publication Date
CN112053325A true CN112053325A (en) 2020-12-08

Family

ID=73602816

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010805337.5A Pending CN112053325A (en) 2020-08-12 2020-08-12 Breast mass image processing and classifying system

Country Status (1)

Country Link
CN (1) CN112053325A (en)

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112528971A (en) * 2021-02-07 2021-03-19 北京智芯微电子科技有限公司 Power transmission line abnormal target detection method and system based on deep learning
CN112767346A (en) * 2021-01-18 2021-05-07 北京医准智能科技有限公司 Multi-image-based full-convolution single-stage mammary image lesion detection method and device
CN112949614A (en) * 2021-04-29 2021-06-11 成都市威虎科技有限公司 Face detection method and device for automatically allocating candidate areas and electronic equipment
CN114820590A (en) * 2022-06-06 2022-07-29 北京医准智能科技有限公司 Image processing method, image processing apparatus, electronic device, and medium

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108648182A (en) * 2018-04-27 2018-10-12 南京信息工程大学 A kind of breast cancer nuclear magnetic resonance image tumor region dividing method based on molecular isoform
CN110223287A (en) * 2019-06-13 2019-09-10 首都医科大学北京友谊医院 A method of early diagnosing mammary cancer rate can be improved
CN110378426A (en) * 2019-07-23 2019-10-25 安徽磐众信息科技有限公司 A kind of Breast Nodules aided diagnosis method and system based on YOLOV3
WO2019223582A1 (en) * 2018-05-24 2019-11-28 Beijing Didi Infinity Technology And Development Co., Ltd. Target detection method and system
CN111062909A (en) * 2019-11-08 2020-04-24 哈尔滨工业大学（深圳） Method and equipment for judging benign and malignant breast tumor
CN111292355A (en) * 2020-02-12 2020-06-16 江南大学 Nuclear correlation filtering multi-target tracking method fusing motion information

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108648182A (en) * 2018-04-27 2018-10-12 南京信息工程大学 A kind of breast cancer nuclear magnetic resonance image tumor region dividing method based on molecular isoform
WO2019223582A1 (en) * 2018-05-24 2019-11-28 Beijing Didi Infinity Technology And Development Co., Ltd. Target detection method and system
CN110223287A (en) * 2019-06-13 2019-09-10 首都医科大学北京友谊医院 A method of early diagnosing mammary cancer rate can be improved
CN110378426A (en) * 2019-07-23 2019-10-25 安徽磐众信息科技有限公司 A kind of Breast Nodules aided diagnosis method and system based on YOLOV3
CN111062909A (en) * 2019-11-08 2020-04-24 哈尔滨工业大学（深圳） Method and equipment for judging benign and malignant breast tumor
CN111292355A (en) * 2020-02-12 2020-06-16 江南大学 Nuclear correlation filtering multi-target tracking method fusing motion information

Non-Patent Citations (4)

* Cited by examiner, † Cited by third party
Title
PRANAV ADARSH等: "YOLO v3-Tiny: Object Detection and Recognition using one stage improved model", 《2020 6TH INTERNATIONAL CONFERENCE ON ADVANCED COMPUTING AND COMMUNICATION SYSTEMS (ICACCS)》 *
ZHAOHUI ZHENG等: "Distance-IoU Loss: Faster and Better Learning for Bounding Box Regression", 《ARXIV:1911.08287》 *
余路阳: "基于多视图注意力机制的钼靶X光乳腺照影计算机辅助诊断研究", 《中国硕士学位论文全文数据库_医药卫生科技辑》 *
邹承明等: "融合GIoU和Focal loss的YOLOv3目标检测算法", 《计算机工程与应用》 *

Cited By (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112767346A (en) * 2021-01-18 2021-05-07 北京医准智能科技有限公司 Multi-image-based full-convolution single-stage mammary image lesion detection method and device
CN112767346B (en) * 2021-01-18 2021-10-29 北京医准智能科技有限公司 Multi-image-based full-convolution single-stage mammary image lesion detection method and device
CN112528971A (en) * 2021-02-07 2021-03-19 北京智芯微电子科技有限公司 Power transmission line abnormal target detection method and system based on deep learning
CN112528971B (en) * 2021-02-07 2021-06-04 北京智芯微电子科技有限公司 Power transmission line abnormal target detection method and system based on deep learning
CN112949614A (en) * 2021-04-29 2021-06-11 成都市威虎科技有限公司 Face detection method and device for automatically allocating candidate areas and electronic equipment
CN114820590A (en) * 2022-06-06 2022-07-29 北京医准智能科技有限公司 Image processing method, image processing apparatus, electronic device, and medium

Similar Documents

Publication Publication Date Title
CN109447065B (en) Method and device for identifying mammary gland image
Abdelrahman et al. Convolutional neural networks for breast cancer detection in mammography: A survey
EP3432263B1 (en) Semantic segmentation for cancer detection in digital breast tomosynthesis
CN109363698B (en) Method and device for identifying mammary gland image signs
US6185320B1 (en) Method and system for detection of lesions in medical images
CN112053325A (en) Breast mass image processing and classifying system
US7646902B2 (en) Computerized detection of breast cancer on digital tomosynthesis mammograms
CN109363699B (en) Method and device for identifying focus of breast image
CN109363697B (en) Method and device for identifying focus of breast image
US9305204B2 (en) Method and system to detect the microcalcifications in X-ray images using nonlinear energy operator
EP1009283A1 (en) Method and system for automated detection of clustered microcalcifications from digital mammograms
US20030161522A1 (en) Method, and corresponding apparatus, for automatic detection of regions of interest in digital images of biological tissue
CN112700461B (en) System for pulmonary nodule detection and characterization class identification
Jiang et al. Breast cancer detection and classification in mammogram using a three-stage deep learning framework based on PAA algorithm
Hu et al. A multi-instance networks with multiple views for classification of mammograms
Rizzi et al. A supervised method for microcalcification cluster diagnosis
KR20220154100A (en) Automated detection of tumors based on image processing
CN115423806B (en) Breast mass detection method based on multi-scale cross-path feature fusion
Ruchay et al. Segmentation of Breast Masses in Digital Mammography Based on U-Net Deep Convolutional Neural Networks
CN111062909A (en) Method and equipment for judging benign and malignant breast tumor
Loizidou et al. Benign and malignant breast mass detection and classification in digital mammography: the effect of subtracting temporally consecutive mammograms
CN113870194B (en) Breast tumor ultrasonic image processing device with fusion of deep layer characteristics and shallow layer LBP characteristics
CN115880245A (en) Self-supervision-based breast cancer disease classification method
Samulski et al. Matching mammographic regions in mediolateral oblique and cranio caudal views: a probabilistic approach
Pelluet et al. Multi-scale graph neural networks for mammography classification and abnormality detection

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
RJ01 Rejection of invention patent application after publication

Application publication date: 20201208

RJ01 Rejection of invention patent application after publication