CN117541570A - Method, device, computer equipment and medium for processing breast nodule image - Google Patents
Method, device, computer equipment and medium for processing breast nodule image Download PDFInfo
- Publication number
- CN117541570A CN117541570A CN202311603271.1A CN202311603271A CN117541570A CN 117541570 A CN117541570 A CN 117541570A CN 202311603271 A CN202311603271 A CN 202311603271A CN 117541570 A CN117541570 A CN 117541570A
- Authority
- CN
- China
- Prior art keywords
- feature
- convolution
- data set
- segmentation
- classification
- 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
Links
- 206010006272 Breast mass Diseases 0.000 title claims abstract description 62
- 238000012545 processing Methods 0.000 title claims abstract description 53
- 238000000034 method Methods 0.000 title claims abstract description 44
- 230000011218 segmentation Effects 0.000 claims abstract description 77
- 238000000605 extraction Methods 0.000 claims abstract description 51
- 238000012549 training Methods 0.000 claims abstract description 43
- 238000007781 pre-processing Methods 0.000 claims abstract description 38
- 238000004590 computer program Methods 0.000 claims abstract description 22
- 210000000481 breast Anatomy 0.000 claims abstract description 14
- 238000002604 ultrasonography Methods 0.000 claims abstract description 14
- 230000006870 function Effects 0.000 claims description 36
- 230000004927 fusion Effects 0.000 claims description 33
- 210000005075 mammary gland Anatomy 0.000 claims description 28
- 238000010586 diagram Methods 0.000 claims description 26
- 238000013527 convolutional neural network Methods 0.000 claims description 20
- 238000005070 sampling Methods 0.000 claims description 17
- 239000011159 matrix material Substances 0.000 claims description 10
- 238000010606 normalization Methods 0.000 claims description 6
- 230000008569 process Effects 0.000 abstract description 12
- 230000003211 malignant effect Effects 0.000 abstract description 8
- 210000003484 anatomy Anatomy 0.000 abstract description 5
- 238000003672 processing method Methods 0.000 abstract description 5
- 108091006146 Channels Proteins 0.000 description 26
- 238000003745 diagnosis Methods 0.000 description 6
- 206010054107 Nodule Diseases 0.000 description 4
- 238000012216 screening Methods 0.000 description 4
- 201000010099 disease Diseases 0.000 description 3
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 description 3
- 238000004458 analytical method Methods 0.000 description 2
- 230000008859 change Effects 0.000 description 2
- 238000013500 data storage Methods 0.000 description 2
- 230000001419 dependent effect Effects 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 230000003902 lesion Effects 0.000 description 2
- 238000005259 measurement Methods 0.000 description 2
- 230000008447 perception Effects 0.000 description 2
- 230000003068 static effect Effects 0.000 description 2
- 238000012285 ultrasound imaging Methods 0.000 description 2
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 description 1
- 208000004434 Calcinosis Diseases 0.000 description 1
- OKTJSMMVPCPJKN-UHFFFAOYSA-N Carbon Chemical compound [C] OKTJSMMVPCPJKN-UHFFFAOYSA-N 0.000 description 1
- 230000002159 abnormal effect Effects 0.000 description 1
- 230000004913 activation Effects 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 208000030270 breast disease Diseases 0.000 description 1
- 230000002308 calcification Effects 0.000 description 1
- 238000013145 classification model Methods 0.000 description 1
- 230000007797 corrosion Effects 0.000 description 1
- 238000005260 corrosion Methods 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 238000002405 diagnostic procedure Methods 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- PCHJSUWPFVWCPO-UHFFFAOYSA-N gold Chemical compound [Au] PCHJSUWPFVWCPO-UHFFFAOYSA-N 0.000 description 1
- 229910021389 graphene Inorganic materials 0.000 description 1
- 238000003384 imaging method Methods 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 238000012886 linear function Methods 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000000877 morphologic effect Effects 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 230000004044 response Effects 0.000 description 1
- 210000001519 tissue Anatomy 0.000 description 1
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
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/26—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
- G06V10/443—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components by matching or filtering
- G06V10/449—Biologically inspired filters, e.g. difference of Gaussians [DoG] or Gabor filters
- G06V10/451—Biologically inspired filters, e.g. difference of Gaussians [DoG] or Gabor filters with interaction between the filter responses, e.g. cortical complex cells
- G06V10/454—Integrating the filters into a hierarchical structure, e.g. convolutional neural networks [CNN]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/52—Scale-space analysis, e.g. wavelet analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/764—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
- G06V10/80—Fusion, i.e. combining data from various sources at the sensor level, preprocessing level, feature extraction level or classification level
- G06V10/806—Fusion, i.e. combining data from various sources at the sensor level, preprocessing level, feature extraction level or classification level of extracted features
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
-
- 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
- G06T2207/10132—Ultrasound 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
- G06V2201/032—Recognition of patterns in medical or anatomical images of protuberances, polyps nodules, etc.
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- Multimedia (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Health & Medical Sciences (AREA)
- Evolutionary Computation (AREA)
- Health & Medical Sciences (AREA)
- Medical Informatics (AREA)
- Artificial Intelligence (AREA)
- Databases & Information Systems (AREA)
- Software Systems (AREA)
- Computing Systems (AREA)
- Radiology & Medical Imaging (AREA)
- Biodiversity & Conservation Biology (AREA)
- Biomedical Technology (AREA)
- Molecular Biology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Quality & Reliability (AREA)
- Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
- Image Processing (AREA)
Abstract
The application relates to a breast nodule image processing method, a device, computer equipment, a storage medium and a computer program product, on one hand, preprocessing and feature extraction are carried out on an acquired breast ultrasound image data set, and then classification and segmentation of breast nodules are carried out by utilizing a classification subnet and a segmentation subnet, so that clinical experience and anatomical structure knowledge of doctors are not required to be relied on; on the other hand, as the classification sub-network is obtained based on the focal loss function training, the segmentation sub-network is obtained based on the Dice loss function training, and the unbalance problem of benign and malignant nodes is fully considered in the classification sub-network and the segmentation sub-network in the training process; therefore, the whole scheme can realize accurate breast nodule image processing.
Description
Technical Field
The present application relates to the field of medical image processing technology, and in particular, to a method, an apparatus, a computer device, a storage medium, and a computer program product for processing breast nodule images.
Background
Ultrasound imaging has been widely used in screening diagnosis of disease due to its low cost, real-time imaging and non-radiative advantages. In the process of ultrasound diagnosis using ultrasound imaging, a clinician first needs to acquire a scan image of the breast location, identify breast nodules, check if the main anatomy is abnormal, and make analysis and diagnosis.
In current breast ultrasound diagnostic procedures, the identification of breast nodules is largely dependent on the experience of the physician. However, this identification method has some non-negligible drawbacks: first, identification and measurement of breast nodules requires manipulation by the sonographer as it relies on the clinician's clinical experience and anatomical knowledge; secondly, the number of patients who are examined currently is large, the screening work of mammary gland diseases is heavy, and the measurement needs to be manually sketched by doctors, so that the step is time-consuming and labor-consuming, and the patients cannot obtain diagnosis results in real time.
Thus, there is a great need for an accurate breast nodule image processing scheme to provide effective data support for breast nodule diagnosis.
Disclosure of Invention
In view of the foregoing, it is desirable to provide an accurate breast nodule image processing method, apparatus, computer device, storage medium and computer program product.
In a first aspect, the present application provides a method for processing breast nodule images. The method comprises the following steps:
acquiring a mammary gland ultrasonic image data set;
preprocessing the mammary gland ultrasonic image data set to obtain a preprocessed data set;
inputting the preprocessing data set into a feature extraction network to obtain a feature map;
And respectively inputting the feature map into a classification subnet and a segmentation subnet to obtain a classification result and a segmentation result of the breast nodule, wherein the classification subnet is obtained based on a focal loss function training, and the segmentation subnet is obtained based on a Dice loss function training.
In one embodiment, the preprocessing the breast ultrasound image dataset to obtain a preprocessed dataset includes:
performing preset size scaling and normalization processing on the mammary gland ultrasonic image data set to obtain a normalized data set;
and carrying out random enhancement processing on each image in the normalized dataset to obtain a preprocessed dataset.
In one embodiment, the inputting the preprocessed data set into the feature extraction network to obtain the feature map includes:
inputting the preprocessing data set into a backbone network in a trained deep convolutional neural network to perform feature extraction, so as to obtain a feature map;
the step of inputting the feature map into a classification subnet and a segmentation subnet respectively to obtain a classification result and a segmentation result of the breast nodule comprises the following steps:
and respectively inputting the feature map into a classification subnet and a segmentation subnet in the trained deep convolutional neural network to obtain a classification result and a segmentation result of the breast nodule.
In one embodiment, the backbone network includes an input layer and a feature extraction layer;
inputting the preprocessing data set into a backbone network in a trained deep convolutional neural network for feature extraction, and obtaining a feature map comprises the following steps:
inputting the preprocessed data set to the input layer;
processing the preprocessing data through the input layer to obtain a pixel matrix;
and performing four-layer feature extraction on the pixel matrix through the feature extraction layer to respectively obtain a feature map C1, a feature map C2, a feature map C3 and a feature map C4.
In one embodiment, inputting the feature map into the segmentation subnet to obtain the segmentation result includes:
up-sampling the characteristic diagram C4 to the same size as the characteristic diagram C3, and convolving the characteristic diagram C3 by 1*1 to obtain a characteristic F3 with the same channel number as C4;
the up-sampled feature map C4 and the feature F3 are input into a multi-scale feature attention fusion module MF together to obtain a feature M1;
up-sampling the feature M1, passing through a convolution layer, and convolving the feature diagram C2 with the feature F2 obtained by 1*1;
the up-sampled and convolved feature M1 and the feature F2 are input into a multi-scale feature attention fusion module MF together to obtain a feature M2;
Up-sampling the feature M2, passing through a convolution layer, and convolving the feature image C1 with the feature F1 obtained by 1*1;
the feature M2 and the feature F1 after up-sampling convolution are input into a multi-scale feature attention fusion module MF together to obtain a feature M3;
the feature M3 is input into the segmentation head after passing through the convolution layer, and a segmented mask diagram is obtained.
In one embodiment, inputting the feature map into the classification subnet to obtain the classification result includes:
the first convolution characteristic is obtained after the feature map C1 passes through a 1*1 convolution layer and a 3*3 convolution layer, and the second convolution characteristic is obtained after the feature map C2 passes through a 1*1 convolution layer;
fusing the first convolution feature and the second convolution feature according to the channel to obtain a first fused feature;
the first fused feature is subjected to a 1*1 convolution layer and a 3*3 convolution layer to obtain a third convolution feature, and a feature map C3 is subjected to a 1*1 convolution layer to obtain a fourth convolution feature;
fusing the third convolution feature and the fourth convolution feature according to the channel to obtain a second fused feature;
the second fused feature is subjected to a 1*1 convolution layer and a 3*3 convolution layer to obtain a fifth convolution feature, and a feature map C4 is subjected to a 1*1 convolution layer to obtain a sixth convolution feature;
fusing the fifth convolution feature and the sixth convolution feature according to the channel to obtain a third fused feature;
The third fused feature is subjected to a 1*1 convolution layer and a 3*3 convolution layer to obtain a seventh convolution feature;
the feature M2 in the split branch is subjected to 1*1 convolution, 3*3 convolution and 1*1 convolution to obtain an eighth convolution feature;
the seventh convolution feature and the eighth convolution feature are input into a multi-scale feature attention fusion module together to obtain a feature M4;
and the feature M4 passes through a convolution layer and a classification head to obtain a classification result.
In one embodiment, the local attention path in the multi-scale feature attention fusion module is a point-by-point convolution-ReLU-point-by-point convolution; the global attention path in the multi-scale feature attention fusion module is global pooling-point-by-point convolution-ReLU-point-by-point convolution.
In a second aspect, the present application further provides a device for processing a breast nodule image. The device comprises:
the data acquisition module is used for acquiring a mammary gland ultrasonic image data set;
the preprocessing module is used for preprocessing the mammary gland ultrasonic image data set to obtain a preprocessed data set;
the feature extraction module is used for inputting the preprocessing data set into a feature extraction network to obtain a feature map;
the processing module is used for inputting the feature map into a classification subnet and a segmentation subnet respectively to obtain a classification result and a segmentation result of the breast nodule, the classification subnet is obtained based on a local loss function training, and the segmentation subnet is obtained based on a Dice loss function training.
In a third aspect, the present application also provides a computer device. The computer device comprises a memory storing a computer program and a processor which when executing the computer program performs the steps of:
acquiring a mammary gland ultrasonic image data set;
preprocessing the mammary gland ultrasonic image data set to obtain a preprocessed data set;
inputting the preprocessing data set into a feature extraction network to obtain a feature map;
and respectively inputting the feature map into a classification subnet and a segmentation subnet to obtain a classification result and a segmentation result of the breast nodule, wherein the classification subnet is obtained based on a focal loss function training, and the segmentation subnet is obtained based on a Dice loss function training.
In a fourth aspect, the present application also provides a computer-readable storage medium. The computer readable storage medium having stored thereon a computer program which when executed by a processor performs the steps of:
acquiring a mammary gland ultrasonic image data set;
preprocessing the mammary gland ultrasonic image data set to obtain a preprocessed data set;
inputting the preprocessing data set into a feature extraction network to obtain a feature map;
And respectively inputting the feature map into a classification subnet and a segmentation subnet to obtain a classification result and a segmentation result of the breast nodule, wherein the classification subnet is obtained based on a focal loss function training, and the segmentation subnet is obtained based on a Dice loss function training.
In a fifth aspect, the present application also provides a computer program product. The computer program product comprises a computer program which, when executed by a processor, implements the steps of:
acquiring a mammary gland ultrasonic image data set;
preprocessing the mammary gland ultrasonic image data set to obtain a preprocessed data set;
inputting the preprocessing data set into a feature extraction network to obtain a feature map;
and respectively inputting the feature map into a classification subnet and a segmentation subnet to obtain a classification result and a segmentation result of the breast nodule, wherein the classification subnet is obtained based on a focal loss function training, and the segmentation subnet is obtained based on a Dice loss function training.
On one hand, preprocessing and feature extraction are carried out on an acquired mammary gland ultrasonic image data set, and then classification and segmentation of mammary gland nodules are carried out by utilizing a classification subnet and a segmentation subnet, so that clinical experience and anatomical structure knowledge of doctors are not needed to be relied on; on the other hand, as the classification sub-network is obtained based on the focal loss function training, the segmentation sub-network is obtained based on the Dice loss function training, and the unbalance problem of benign and malignant nodes is fully considered in the classification sub-network and the segmentation sub-network in the training process; therefore, the whole scheme can realize accurate breast nodule image processing.
Drawings
FIG. 1 is an application environment diagram of a method of processing breast nodule images in one embodiment;
FIG. 2 is a flow chart of a method of processing breast nodule images in one embodiment;
FIG. 3 is a flow chart of a method of processing breast nodule images in another embodiment;
FIG. 4 is a schematic diagram of an architecture of a deep convolutional neural network in one embodiment;
FIG. 5 is a schematic diagram of an architecture of a multi-scale feature attention fusion module MF in one embodiment;
FIG. 6 is a block diagram of an apparatus for processing breast nodule images in one embodiment;
fig. 7 is an internal structural diagram of a computer device in one embodiment.
Detailed Description
In order to make the objects, technical solutions and advantages of the present application more apparent, the present application will be further described in detail with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are for purposes of illustration only and are not intended to limit the present application.
The method for processing the breast nodule image can be applied to an application environment shown in fig. 1. Wherein the terminal 102 communicates with the server 104 via a network. The data storage system may store data that the server 104 needs to process. The data storage system may be integrated on the server 104 or may be located on a cloud or other network server. The terminal 102 acquires a mammary gland ultrasonic image, generates a mammary gland ultrasonic image processing request, sends the mammary gland ultrasonic image processing request to the server 104, and the server 104 receives the request to acquire a mammary gland ultrasonic image data set; preprocessing a mammary gland ultrasonic image data set to obtain a preprocessed data set; inputting the preprocessed data set into a feature extraction network to obtain a feature map; and respectively inputting the feature map into a classification subnet and a segmentation subnet to obtain a classification result and a segmentation result of the breast nodule. Further, the server 104 may feed back the classification result and the segmentation result of the breast nodule to the terminal 102. The terminal 102 may be, but not limited to, various personal computers, notebook computers, smart phones, tablet computers, internet of things devices, and portable wearable devices, where the internet of things devices may be smart speakers, smart televisions, smart air conditioners, smart vehicle devices, and the like. The portable wearable device may be a smart watch, smart bracelet, headset, or the like. The server 104 may be implemented as a stand-alone server or as a server cluster of multiple servers.
In one embodiment, as shown in fig. 2, a method for processing a breast nodule image is provided, and the method is applied to the server 104 in fig. 1 for illustration, and includes the following steps:
s200: a breast ultrasound image dataset is acquired.
The breast ultrasound image set refers to a data set formed by a plurality of breast ultrasound images, which may specifically be a data set formed by acquisition of the same object with a medical ultrasound image device.
S400: preprocessing the mammary gland ultrasonic image data set to obtain a preprocessed data set.
Preprocessing is carried out on the mammary gland ultrasonic image so as to improve the efficiency and accuracy of subsequent data processing. In particular, preprocessing may include filtering, normalization, enhancement, etc., which may be followed by standard preprocessing datasets.
S600: and inputting the preprocessed data set into a feature extraction network to obtain a feature map.
The preprocessed data set is input into a feature extraction network for feature extraction. Specifically, feature extraction of different stage layers can be performed on the preprocessed data set, so as to obtain feature graphs corresponding to the different stage layers. In practical application, feature extraction of different stage layers can be performed respectively for different breast ultrasound images, and finally feature graphs corresponding to the different stage layers are formed. Further, the feature extraction network may be a Resnet-50, which may include multiple layers, for example, a conv1.X layer, a conv2.X layer, a conv3.X layer, and a conv4.X layer, and output matrices of these layers may be used as the extracted feature maps C1, C2, C3, and C4.
S800: the feature map is respectively input into a classification subnet and a segmentation subnet, so that a classification result and a segmentation result of the breast nodule are obtained, the classification subnet is obtained based on the focal loss function training, and the segmentation subnet is obtained based on the Dice loss function training.
The feature map is input into the classifying sub-network and the dividing sub-network respectively, the classifying sub-network performs breast nodule classifying treatment based on the feature map, and the dividing sub-network performs breast nodule dividing treatment based on the feature to obtain a classifying result and a dividing result respectively. Specifically, the classification sub-network and the segmentation sub-network may be trained in advance based on sample data. In the training process, because of the unbalanced problem of samples of benign and malignant nodules, the classifying sub-network is obtained based on the focal loss function training, and the dividing sub-network is obtained based on the Dice loss function training. More specifically, during conventional segmentation, classification model training, convolution operations may wipe out some details of the target boundaries. In fact, for some local boundaries, it is also difficult for even experienced sonographers to determine pixels at the boundary, so in order to boost the model's focus of attention to the lesion boundary, the Dice Loss function training is employed for the segmented subnetworks, and the classification subnetworks are trained based on the focal Loss function to apply boundary perception constraints, applying higher weights to difficult samples at the boundary to enhance the model's boundary recognition capability.
According to the breast nodule image processing method, on one hand, preprocessing and feature extraction are carried out on an acquired breast ultrasound image data set, and then classification and segmentation of breast nodules are carried out by utilizing a classification subnet and a segmentation subnet, so that clinical experience and anatomical structure knowledge of doctors are not required to be relied on; on the other hand, as the classification sub-network is obtained based on the focal loss function training, the segmentation sub-network is obtained based on the Dice loss function training, and the unbalance problem of benign and malignant nodes is fully considered in the classification sub-network and the segmentation sub-network in the training process; therefore, the whole scheme can realize accurate breast nodule image processing.
As shown in fig. 3, in one embodiment, S400 includes:
s420: performing preset size scaling and normalization processing on the mammary gland ultrasonic image data set to obtain a normalized data set;
s440: and carrying out random enhancement processing on each image in the normalized data set to obtain a preprocessed data set.
The preset size is a preset size, which can be set according to the actual situation, for example, 800×600. The breast ultrasound image data set is subjected to scaling processing of the same size so as to unify the size of input data for training, and the sizes of images scanned by different machines are different, so that the training efficiency can be improved and the video memory space can be saved if the images are scaled to the uniform size. When normalization processing is performed, a linear function can be applied to normalize the scaled image, and a normalized image is obtained. And carrying out random enhancement operation on each normalized image to obtain randomly enhanced images.
In one embodiment, inputting the preprocessed data set into the feature extraction network to obtain the feature map includes: inputting the preprocessed data set into a backbone network in a trained deep convolutional neural network to perform feature extraction, so as to obtain a feature map;
the feature map is respectively input into a classification subnet and a segmentation subnet, and the obtaining of the classification result and the segmentation result of the breast nodule comprises the following steps: and respectively inputting the feature map into a classification subnet and a segmentation subnet in the trained deep convolutional neural network to obtain a classification result and a segmentation result of the breast nodule.
The trained deep convolutional neural network is a pre-trained good network, which may include, in particular, a backbone network, a classification sub-network, and a segmentation sub-network of sequential connection numbers. The backbone network may specifically be ResNet-50. The whole network extraction result can comprise an input layer and a characteristic extraction layer. Specifically, the first layer is an input layer, which is used for processing an input preprocessing data set to obtain a pixel matrix with the same size, and then inputting the pixel matrix into the feature extraction layer; the second layer is a feature extraction layer, which can perform multi-layer feature extraction on the pixel matrix, and specifically can perform four-layer feature extraction on the pixel matrix to obtain a feature map C1, a feature map C2, a feature map C3, and a feature map C4, respectively. Still further, the feature extraction layer may employ a feature extraction network Resnet-50, and take output matrices of four layers of conv1.X layer, conv2.X layer, conv3.X layer and conv4.X layer in the feature extraction network Resnet-50 as the extracted feature maps C1, C2, C3, and C4. In practical application, the above 4 features refer to features of the first four stage layers output from five stages of the resnet50, the feature map C1 represents a feature map size of 300×400×64, the feature map C2 represents 150×200×256, the feature map C3 represents 75×100×512, and the feature map C4 represents 38×50×1024.
In one embodiment, inputting the feature map into the segmentation subnet to obtain the segmentation result includes:
up-sampling the characteristic diagram C4 to the same size as the characteristic diagram C3, and convolving the characteristic diagram C3 by 1*1 to obtain a characteristic F3 with the same channel number as C4;
the up-sampled feature map C4 and the feature F3 are input into a multi-scale feature attention fusion module MF together to obtain a feature M1;
up-sampling the feature M1, passing through a convolution layer, and convolving the feature diagram C2 with the feature F2 obtained by 1*1;
the up-sampled and convolved feature M1 and the feature F2 are input into a multi-scale feature attention fusion module MF together to obtain a feature M2;
up-sampling the feature M2, passing through a convolution layer, and convolving the feature image C1 with the feature F1 obtained by 1*1;
the feature M2 and the feature F1 after up-sampling convolution are input into a multi-scale feature attention fusion module MF together to obtain a feature M3;
the feature M3 is input into the segmentation head after passing through the convolution layer, and a segmented mask diagram is obtained.
The input of the split sub-network is the extracted feature map. As shown in fig. 4, for the split sub-network, the feature C4 extracted by the last layer of convolution layer of the backbone network is up-sampled to the same size as the feature C3, and the feature F3 with the same channel number as the feature C4 obtained by convolving the feature C3 extracted by the third layer of convolution layer with a 1*1 is input to the multi-scale feature attention fusion module MF together to obtain a feature M1; then up-sampling the feature M1, passing through a convolution layer, and inputting the feature F2 obtained by convolving the feature M1 with the feature C2 extracted by the second convolution layer together through a 1*1 convolution into a multi-scale feature attention fusion module MF to obtain a feature M2; similarly, the M2 is up-sampled and passed through a convolution layer, and the feature F1 obtained by convolving the convolution layer with the feature C1 extracted from the first layer is input into a multi-scale feature attention fusion module MF together to obtain M3, and finally the M3 is input into a segmentation head after passing through a convolution layer to obtain a segmented mask map
In one embodiment, inputting the feature map into the classification subnet to obtain the classification result includes:
the first convolution characteristic is obtained after the feature map C1 passes through a 1*1 convolution layer and a 3*3 convolution layer, and the second convolution characteristic is obtained after the feature map C2 passes through a 1*1 convolution layer;
fusing the first convolution feature and the second convolution feature according to the channel to obtain a first fused feature;
the first fused feature is subjected to a 1*1 convolution layer and a 3*3 convolution layer to obtain a third convolution feature, and a feature map C3 is subjected to a 1*1 convolution layer to obtain a fourth convolution feature;
fusing the third convolution feature and the fourth convolution feature according to the channel to obtain a second fused feature;
the second fused feature is subjected to a 1*1 convolution layer and a 3*3 convolution layer to obtain a fifth convolution feature, and a feature map C4 is subjected to a 1*1 convolution layer to obtain a sixth convolution feature;
fusing the fifth convolution feature and the sixth convolution feature according to the channel to obtain a third fused feature;
the third fused feature is subjected to a 1*1 convolution layer and a 3*3 convolution layer to obtain a seventh convolution feature;
the feature M2 in the split branch is subjected to 1*1 convolution, 3*3 convolution and 1*1 convolution to obtain an eighth convolution feature;
the seventh convolution feature and the eighth convolution feature are input into a multi-scale feature attention fusion module together to obtain a feature M4;
And the feature M4 passes through a convolution layer and a classification head to obtain a classification result.
For the classifying sub-network, as shown in fig. 4, features C1 and C2 are fused according to channels by passing through a 1*1 convolution layer and a 8238 convolution layer, then the fused features are fused according to channels by passing through a 1*1 convolution layer and a 3*3 convolution layer, then the fused features are fused according to channels by passing through a 1*1 convolution layer and a 1*1 convolution layer, then the fused features are fused according to channels by passing through a 1*1 convolution layer and a 3*3 convolution layer, then the fused features are fused according to channels by passing through a 1*1 convolution layer and a 3*3 convolution layer, then the fused features are input into a multi-scale feature attention fusion module together with the features of M2 in the dividing branch, which are convolved by passing through a 1*1 convolution layer, 3*3 convolution layer and a 1*1 convolution layer, and finally the classifying result is obtained. Notably, morphological irregularities, aspect ratios greater than 1, edge blurring (angulation, tiny lobes, "crab foot-like protrusions"), inhomogeneous hypoechos, post-echogenic changes, structural disturbances invading surrounding tissues, calcifications, etc. can all be considered malignant markers. The classifying branches integrate the characteristic that the dividing branches are close to the output result, so that the positions of focuses can be indicated for the classifying branches to a certain extent, the classifying branches can pay more attention to the characteristics of focus positions in the images, and the classifying branches are more beneficial to identifying benign and malignant nodules. These features can guide the branching of the classification, both in forward and reverse propagation.
In one embodiment, the local attention path in the multi-scale feature attention fusion module is a point-by-point convolution-ReLU-point-by-point convolution; the global attention path in the multi-scale feature attention fusion module is global pooling-point-by-point convolution-ReLU-point-by-point convolution.
The multi-scale feature attention fusion module MF is shown in fig. 5, features X and Y are added, and then added features are respectively input into a local attention path of 'point-by-point convolution-ReLU-point-by-point convolution' and a global attention path of 'global pooling-point-by-point convolution-ReLU-point-by-point convolution', and global context and local context of the features after feature X and Y are fused are respectively extracted. Notably, the present project selects point-by-point convolution on the attention path as a channel context aggregator that can achieve channel attention on multiple scales using only point-to-channel fusion for each spatial location. Then adding the global context and the global context, and multiplying w by a feature X through a channel attention weight w obtained after the function is activated by Sigmoid; and meanwhile, taking 1-w as the attention weight of another channel, multiplying the attention weight by the feature Y, and carrying out addition fusion on the obtained feature and the feature weighted by the attention weight w of X to obtain output O. Therefore, the local context information and the global context information of the features can be obtained simultaneously, the model can pay attention to the scale problem of the channel, and the channel interaction can be calculated according to the point by point for each spatial position through point convolution. Specifically, the Sigmoid activation function limits the resulting output of the addition of the previous global context feature and the local context feature to between 0 and 1 to represent the weights of each channel that are used to adjust the channel feature response.
In order to describe the technical scheme and the technical effect of the application in detail, the training process of the trained deep convolutional neural network will be described in detail below.
In one embodiment, the deep convolutional neural network is trained by:
(a1) And acquiring a data set, sending the data set to a radiologist, and acquiring the data set marked by the radiologist.
(a2) And performing preprocessing such as random clipping on the marked data set to obtain a preprocessed data set.
(a3) Inputting a batch of data in a training set part in the preprocessed data set obtained in the step (a 2) into a deep convolutional neural network to obtain an inference output, and inputting the inference output and the data set marked by the radio expert in the step (a 1) into a loss function L of the deep convolutional neural network all To obtain a loss value.
(a4) A loss function L of the depth convolution neural network according to the Adam algorithm and using the loss value obtained in step (a 3) all Optimizing; briefly, here we apply to the loss function L all After the optimization, the deep convolutional neural network is optimized, and the optimized deep convolutional neural network is trained again during the training of the next round of training data.
(a5) And (3) repeating the steps (a 3) and (a 4) for the rest batch data in the training set part in the preprocessing data set obtained in the step (a 2) until the iteration times are reached, thereby obtaining the trained deep convolutional neural network.
Preferably, the loss value used in the deep convolutional neural network is obtained by the following loss function L all And (3) calculating:
L all =L cls +L mask
wherein L is cls Is a classification loss, L mask Is the segmentation mask penalty.
In training classification branches, because of imbalance problems with samples of benign and malignant nodules, focalloss is used in this application to train classification branches of models, and the specific Focalloss loss function is as follows:
although the multi-scale feature attention fusion module can improve the ability of the model to extract detailed features of the image, the convolution operation will wipe some details of the target boundary. In fact, for some local boundaries, it is also difficult for even an experienced sonographer to determine the pixels at the boundary, so to boost the model's focus on the lesion boundary, the present application intends to apply higher weights to difficult samples at the boundary based on the Dice Loss function using boundary perception constraints to enhance the model's boundary recognition capability. Specific boundary-aware constraints are as follows:
Wherein x, y represents the pixel of the (x, y) coordinate point in the image, g x,y Then represents the gold standard of the image annotation, p x,y The predicted value of the model is represented. In addition, M x,y Represented are spatial weights calculated as follows:
M=Guass(k·(D(g)-E(g))+1
where D and E represent an expansion operation and a corrosion operation, respectively, gauss is to give some attention to pixels that are close to the boundary but not located on the boundary, here the range for Gauss is set to 5*5; here, 1 indicates whether the pixel 224×224 is far from the boundary, and if so, 1, or 0, on the contrary.
In practical application, the breast nodule image processing method at least has the following remarkable technical effects:
(1) Because the data sets used in the learning process are all selected and accurately marked by the ultrasonic doctors according to clinical experience, the knowledge of the ultrasonic doctors can be obtained through machine learning in the method, so that the doctor can program and automate the whole process of checking patients, the problem that the existing method for automatically dividing and classifying breast nodules is excessively dependent on the clinical experience and anatomical knowledge of the doctor can be solved, and the ultrasonic doctors can be assisted to realize the screening and diagnosis of breast diseases.
(2) Because the utility model is fully automatic and programmed, the ultrasonic doctor does not need to stop to measure in the detection process, and does not need to check the accuracy of breast nodules, and classification and segmentation are all real-time, the technical problem that the existing ultrasonic doctor has overlong checking time in the aspect of disease screening can be solved.
It should be understood that, although the steps in the flowcharts related to the above embodiments are sequentially shown as indicated by arrows, these steps are not necessarily sequentially performed in the order indicated by the arrows. The steps are not strictly limited to the order of execution unless explicitly recited herein, and the steps may be executed in other orders. Moreover, at least some of the steps in the flowcharts described in the above embodiments may include a plurality of steps or a plurality of stages, which are not necessarily performed at the same time, but may be performed at different times, and the order of the steps or stages is not necessarily performed sequentially, but may be performed alternately or alternately with at least some of the other steps or stages.
Based on the same inventive concept, the embodiment of the application also provides a processing device for the breast nodule image for realizing the processing method of the breast nodule image. The implementation of the solution provided by the device is similar to the implementation described in the above method, so the specific limitation in the embodiments of the device for processing a breast nodule image or images provided below may be referred to as the limitation of the method for processing a breast nodule image hereinabove, and will not be described herein.
In one embodiment, as shown in fig. 6, there is provided a processing apparatus for breast nodule images, comprising:
a data acquisition module 200 for acquiring a breast ultrasound image dataset;
the preprocessing module 400 is used for preprocessing the mammary gland ultrasonic image data set to obtain a preprocessed data set;
the feature extraction module 600 is configured to input the preprocessed data set to a feature extraction network to obtain a feature map;
the processing module 800 is configured to input the feature map to a classification subnet and a segmentation subnet, respectively, to obtain a classification result and a segmentation result of the breast nodule, the classification subnet is obtained based on the focal loss function training, and the segmentation subnet is obtained based on the Dice loss function training.
On one hand, the processing device of the breast nodule image carries out pretreatment and feature extraction on the acquired breast ultrasonic image data set, and then uses the classification subnet and the segmentation subnet to classify and segment the breast nodule, so that the clinical experience and anatomical structure knowledge of doctors are not needed to be relied on; on the other hand, as the classification sub-network is obtained based on the focal loss function training, the segmentation sub-network is obtained based on the Dice loss function training, and the unbalance problem of benign and malignant nodes is fully considered in the classification sub-network and the segmentation sub-network in the training process; therefore, the whole scheme can realize accurate breast nodule image processing.
In one embodiment, the preprocessing module 400 is further configured to perform a preset size scaling and normalization process on the breast ultrasound image dataset to obtain a normalized dataset; and carrying out random enhancement processing on each image in the normalized data set to obtain a preprocessed data set.
In one embodiment, the feature extraction module 600 is further configured to input the preprocessed data set into a backbone network of the trained deep convolutional neural network for feature extraction, so as to obtain a feature map; the processing module 800 is further configured to input the feature map to a classification subnet and a segmentation subnet in the trained deep convolutional neural network, respectively, to obtain a classification result and a segmentation result of the breast nodule.
In one embodiment, the backbone network includes an input layer and a feature extraction layer; the feature extraction module 600 is further configured to input the preprocessed data set to the input layer; processing the preprocessed data through the input layer to obtain a pixel matrix; four layers of feature extraction are carried out on the pixel matrix through the feature extraction layer, and a feature map C1, a feature map C2, a feature map C3 and a feature map C4 are respectively obtained.
In one embodiment, the processing module 800 inputs the feature map to the segmentation subnet to obtain the segmentation result includes:
up-sampling the characteristic diagram C4 to the same size as the characteristic diagram C3, and convolving the characteristic diagram C3 by 1*1 to obtain a characteristic F3 with the same channel number as C4; the up-sampled feature map C4 and the feature F3 are input into a multi-scale feature attention fusion module MF together to obtain a feature M1; up-sampling the feature M1, passing through a convolution layer, and convolving the feature diagram C2 with the feature F2 obtained by 1*1; the up-sampled and convolved feature M1 and the feature F2 are input into a multi-scale feature attention fusion module MF together to obtain a feature M2; up-sampling the feature M2, passing through a convolution layer, and convolving the feature image C1 with the feature F1 obtained by 1*1; the feature M2 and the feature F1 after up-sampling convolution are input into a multi-scale feature attention fusion module MF together to obtain a feature M3; the feature M3 is input into the segmentation head after passing through the convolution layer, and a segmented mask diagram is obtained.
In one embodiment, the processing module 800 inputs the feature map to the classification subnet to obtain the classification result includes:
the first convolution characteristic is obtained after the feature map C1 passes through a 1*1 convolution layer and a 3*3 convolution layer, and the second convolution characteristic is obtained after the feature map C2 passes through a 1*1 convolution layer; fusing the first convolution feature and the second convolution feature according to the channel to obtain a first fused feature; the first fused feature is subjected to a 1*1 convolution layer and a 3*3 convolution layer to obtain a third convolution feature, and a feature map C3 is subjected to a 1*1 convolution layer to obtain a fourth convolution feature; fusing the third convolution feature and the fourth convolution feature according to the channel to obtain a second fused feature; the second fused feature is subjected to a 1*1 convolution layer and a 3*3 convolution layer to obtain a fifth convolution feature, and a feature map C4 is subjected to a 1*1 convolution layer to obtain a sixth convolution feature; fusing the fifth convolution feature and the sixth convolution feature according to the channel to obtain a third fused feature; the third fused feature is subjected to a 1*1 convolution layer and a 3*3 convolution layer to obtain a seventh convolution feature; the feature M2 in the split branch is subjected to 1*1 convolution, 3*3 convolution and 1*1 convolution to obtain an eighth convolution feature; the seventh convolution feature and the eighth convolution feature are input into a multi-scale feature attention fusion module together to obtain a feature M4; and the feature M4 passes through a convolution layer and a classification head to obtain a classification result.
In one embodiment, the local attention path in the multi-scale feature attention fusion module is a point-by-point convolution-ReLU-point-by-point convolution; the global attention path in the multi-scale feature attention fusion module is global pooling-point-by-point convolution-ReLU-point-by-point convolution.
The respective modules in the processing device for breast nodule images may be implemented in whole or in part by software, hardware, and combinations thereof. The above modules may be embedded in hardware or may be independent of a processor in the computer device, or may be stored in software in a memory in the computer device, so that the processor may call and execute operations corresponding to the above modules.
In one embodiment, a computer device is provided, which may be a server, the internal structure of which may be as shown in fig. 7. The computer device includes a processor, a memory, and a network interface connected by a system bus. Wherein the processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The database of the computer device is used for storing data such as trained network models. The network interface of the computer device is used for communicating with an external terminal through a network connection. The computer program is executed by a processor to implement a method of processing breast nodule images.
It will be appreciated by those skilled in the art that the structure shown in fig. 7 is merely a block diagram of some of the structures associated with the present application and is not limiting of the computer device to which the present application may be applied, and that a particular computer device may include more or fewer components than shown, or may combine certain components, or have a different arrangement of components.
In an embodiment, a computer device is provided, comprising a memory and a processor, the memory storing a computer program, the processor implementing the method of processing breast nodule images of any of the embodiments described above when executing the computer program.
In one embodiment, a computer readable storage medium is provided, on which a computer program is stored, which when executed by a processor, implements the method of processing breast nodule images of any of the embodiments described above.
In an embodiment, a computer program product is provided, comprising a computer program which, when executed by a processor, implements the method of processing breast nodule images of any of the embodiments described above.
It should be noted that, user information (including but not limited to user equipment information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, presented data, etc.) referred to in the present application are information and data authorized by the user or sufficiently authorized by each party.
Those skilled in the art will appreciate that implementing all or part of the above-described methods may be accomplished by way of a computer program stored on a non-transitory computer readable storage medium, which when executed may comprise the steps of the embodiments of the methods described above. Any reference to memory, database, or other medium used in the various embodiments provided herein may include at least one of non-volatile and volatile memory. The nonvolatile Memory may include Read-Only Memory (ROM), magnetic tape, floppy disk, flash Memory, optical Memory, high density embedded nonvolatile Memory, resistive random access Memory (ReRAM), magnetic random access Memory (Magnetoresistive Random Access Memory, MRAM), ferroelectric Memory (Ferroelectric Random Access Memory, FRAM), phase change Memory (Phase Change Memory, PCM), graphene Memory, and the like. Volatile memory can include random access memory (Random Access Memory, RAM) or external cache memory, and the like. By way of illustration, and not limitation, RAM can be in the form of a variety of forms, such as static random access memory (Static Random Access Memory, SRAM) or dynamic random access memory (Dynamic Random Access Memory, DRAM), and the like. The databases referred to in the various embodiments provided herein may include at least one of relational databases and non-relational databases. The non-relational database may include, but is not limited to, a blockchain-based distributed database, and the like. The processors referred to in the embodiments provided herein may be general purpose processors, central processing units, graphics processors, digital signal processors, programmable logic units, quantum computing-based data processing logic units, etc., without being limited thereto.
The technical features of the above embodiments may be arbitrarily combined, and all possible combinations of the technical features in the above embodiments are not described for brevity of description, however, as long as there is no contradiction between the combinations of the technical features, they should be considered as the scope of the description.
The foregoing examples represent only a few embodiments of the present application, which are described in more detail and are not thereby to be construed as limiting the scope of the present application. It should be noted that it would be apparent to those skilled in the art that various modifications and improvements could be made without departing from the spirit of the present application, which would be within the scope of the present application. Accordingly, the scope of protection of the present application shall be subject to the appended claims.
Claims (10)
1. A method of processing breast nodule images, the method comprising:
acquiring a mammary gland ultrasonic image data set;
preprocessing the mammary gland ultrasonic image data set to obtain a preprocessed data set;
inputting the preprocessing data set into a feature extraction network to obtain a feature map;
and respectively inputting the feature map into a classification subnet and a segmentation subnet to obtain a classification result and a segmentation result of the breast nodule, wherein the classification subnet is obtained based on a focal loss function training, and the segmentation subnet is obtained based on a Diceloss loss function training.
2. The method of claim 1, wherein preprocessing the breast ultrasound image dataset to obtain a preprocessed dataset comprises:
performing preset size scaling and normalization processing on the mammary gland ultrasonic image data set to obtain a normalized data set;
and carrying out random enhancement processing on each image in the normalized dataset to obtain a preprocessed dataset.
3. The method of claim 1, wherein inputting the preprocessed data set into a feature extraction network to obtain a feature map comprises:
inputting the preprocessing data set into a backbone network in a trained deep convolutional neural network to perform feature extraction, so as to obtain a feature map;
the step of inputting the feature map into a classification subnet and a segmentation subnet respectively to obtain a classification result and a segmentation result of the breast nodule comprises the following steps:
and respectively inputting the feature map into a classification subnet and a segmentation subnet in the trained deep convolutional neural network to obtain a classification result and a segmentation result of the breast nodule.
4. A method according to claim 3, wherein the backbone network comprises an input layer and a feature extraction layer;
Inputting the preprocessing data set into a backbone network in a trained deep convolutional neural network for feature extraction, and obtaining a feature map comprises the following steps:
inputting the preprocessed data set to the input layer;
processing the preprocessing data through the input layer to obtain a pixel matrix;
and performing four-layer feature extraction on the pixel matrix through the feature extraction layer to respectively obtain a feature map C1, a feature map C2, a feature map C3 and a feature map C4.
5. The method of claim 4, wherein inputting the feature map to a segmentation subnet to obtain a segmentation result comprises:
up-sampling the characteristic diagram C4 to the same size as the characteristic diagram C3, and convolving the characteristic diagram C3 by 1*1 to obtain a characteristic F3 with the same channel number as C4;
the up-sampled feature map C4 and the feature F3 are input into a multi-scale feature attention fusion module MF together to obtain a feature M1;
up-sampling the feature M1, passing through a convolution layer, and convolving the feature diagram C2 with the feature F2 obtained by 1*1;
the up-sampled and convolved feature M1 and the feature F2 are input into a multi-scale feature attention fusion module MF together to obtain a feature M2;
up-sampling the feature M2, passing through a convolution layer, and convolving the feature image C1 with the feature F1 obtained by 1*1;
The feature M2 and the feature F1 after up-sampling convolution are input into a multi-scale feature attention fusion module MF together to obtain a feature M3;
the feature M3 is input into the segmentation head after passing through the convolution layer, and a segmented mask diagram is obtained.
6. The method of claim 5, wherein inputting the feature map to the classification subnet to obtain the classification result comprises:
the first convolution characteristic is obtained after the feature map C1 passes through a 1*1 convolution layer and a 3*3 convolution layer, and the second convolution characteristic is obtained after the feature map C2 passes through a 1*1 convolution layer;
fusing the first convolution feature and the second convolution feature according to the channel to obtain a first fused feature;
the first fused feature is subjected to a 1*1 convolution layer and a 3*3 convolution layer to obtain a third convolution feature, and a feature map C3 is subjected to a 1*1 convolution layer to obtain a fourth convolution feature;
fusing the third convolution feature and the fourth convolution feature according to the channel to obtain a second fused feature;
the second fused feature is subjected to a 1*1 convolution layer and a 3*3 convolution layer to obtain a fifth convolution feature, and a feature map C4 is subjected to a 1*1 convolution layer to obtain a sixth convolution feature;
fusing the fifth convolution feature and the sixth convolution feature according to the channel to obtain a third fused feature;
The third fused feature is subjected to a 1*1 convolution layer and a 3*3 convolution layer to obtain a seventh convolution feature;
the feature M2 in the split branch is subjected to 1*1 convolution, 3*3 convolution and 1*1 convolution to obtain an eighth convolution feature;
the seventh convolution feature and the eighth convolution feature are input into a multi-scale feature attention fusion module together to obtain a feature M4;
and the feature M4 passes through a convolution layer and a classification head to obtain a classification result.
7. The method according to claim 5 or 6, wherein the local attention path in the multi-scale feature attention fusion module is a point-by-point convolution-ReLU-point-by-point convolution; the global attention path in the multi-scale feature attention fusion module is global pooling-point-by-point convolution-ReLU-point-by-point convolution.
8. A device for processing breast nodule images, the device comprising:
the data acquisition module is used for acquiring a mammary gland ultrasonic image data set;
the preprocessing module is used for preprocessing the mammary gland ultrasonic image data set to obtain a preprocessed data set;
the feature extraction module is used for inputting the preprocessing data set into a feature extraction network to obtain a feature map;
the processing module is used for inputting the feature map into a classification subnet and a segmentation subnet respectively to obtain a classification result and a segmentation result of the breast nodule, the classification subnet is obtained based on a focal loss function training, and the segmentation subnet is obtained based on a Diceloss loss function training.
9. A computer device comprising a memory and a processor, the memory storing a computer program, characterized in that the processor implements the steps of the method of any of claims 1 to 7 when the computer program is executed.
10. A computer readable storage medium, on which a computer program is stored, characterized in that the computer program, when being executed by a processor, implements the steps of the method of any of claims 1 to 7.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202311603271.1A CN117541570A (en) | 2023-11-28 | 2023-11-28 | Method, device, computer equipment and medium for processing breast nodule image |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202311603271.1A CN117541570A (en) | 2023-11-28 | 2023-11-28 | Method, device, computer equipment and medium for processing breast nodule image |
Publications (1)
Publication Number | Publication Date |
---|---|
CN117541570A true CN117541570A (en) | 2024-02-09 |
Family
ID=89782139
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202311603271.1A Pending CN117541570A (en) | 2023-11-28 | 2023-11-28 | Method, device, computer equipment and medium for processing breast nodule image |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN117541570A (en) |
-
2023
- 2023-11-28 CN CN202311603271.1A patent/CN117541570A/en active Pending
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110599476B (en) | Disease grading method, device, equipment and medium based on machine learning | |
Oskal et al. | A U-net based approach to epidermal tissue segmentation in whole slide histopathological images | |
Roth et al. | A new 2.5 D representation for lymph node detection using random sets of deep convolutional neural network observations | |
Shaziya et al. | Automatic lung segmentation on thoracic CT scans using U-net convolutional network | |
CN111161275B (en) | Method and device for segmenting target object in medical image and electronic equipment | |
WO2020038240A1 (en) | Image processing method and apparatus, computer-readable storage medium and computer device | |
CN112041912A (en) | Systems and methods for diagnosing gastrointestinal tumors | |
CN110599528A (en) | Unsupervised three-dimensional medical image registration method and system based on neural network | |
Li et al. | Automated measurement network for accurate segmentation and parameter modification in fetal head ultrasound images | |
CN112381164B (en) | Ultrasound image classification method and device based on multi-branch attention mechanism | |
CN111862044A (en) | Ultrasonic image processing method and device, computer equipment and storage medium | |
US20220101034A1 (en) | Method and system for segmenting interventional device in image | |
Qi et al. | Automatic lacunae localization in placental ultrasound images via layer aggregation | |
CN114332132A (en) | Image segmentation method and device and computer equipment | |
CN114445356A (en) | Multi-resolution-based full-field pathological section image tumor rapid positioning method | |
Mridha et al. | A comprehensive survey on the progress, process, and challenges of lung cancer detection and classification | |
CN110992310A (en) | Method and device for determining partition where mediastinal lymph node is located | |
Gilani et al. | Skin lesion analysis using generative adversarial networks: a review | |
CN116228753A (en) | Tumor prognosis evaluation method, device, computer equipment and storage medium | |
CN114360695B (en) | Auxiliary system, medium and equipment for breast ultrasonic scanning and analyzing | |
CN116129184A (en) | Multi-phase focus classification method, device, equipment and readable storage medium | |
CN116128895A (en) | Medical image segmentation method, apparatus and computer readable storage medium | |
CN113379770B (en) | Construction method of nasopharyngeal carcinoma MR image segmentation network, image segmentation method and device | |
CN117541570A (en) | Method, device, computer equipment and medium for processing breast nodule image | |
Mansour et al. | Kidney segmentations using cnn models |
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 |