CN110232396A - X-ray breast image deep learning classification method - Google Patents
X-ray breast image deep learning classification method Download PDFInfo
- Publication number
- CN110232396A CN110232396A CN201910278910.9A CN201910278910A CN110232396A CN 110232396 A CN110232396 A CN 110232396A CN 201910278910 A CN201910278910 A CN 201910278910A CN 110232396 A CN110232396 A CN 110232396A
- Authority
- CN
- China
- Prior art keywords
- membership
- degree
- classification
- sample
- convolution
- 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.)
- Granted
Links
Classifications
-
- 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
- G06F18/217—Validation; Performance evaluation; Active pattern learning techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- 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
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10072—Tomographic images
- G06T2207/10081—Computed x-ray tomography [CT]
-
- 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
Abstract
The invention discloses a kind of X-ray breast lump image automatic classification methods.The present invention devises from image processing point and carries out automatic sorter network to X-ray breast lump image, the network carries out convolution using different size of convolution kernel using two calculating paths to X-ray breast lump image first and down-sampling operates, extract the convolution characteristic pattern of different scale type, the characteristic pattern that two calculate path input is overlapped fusion, obtains double calculating fused characteristic informations in path.Feature extraction is carried out using full convolutional network to fusion feature again, Softmax classification layer is finally sent into and classifies to feature, obtain breast lump image classification result.Model is trained using the objective function based on degree of membership for being suitable for X-ray breast lump image classification, new objective function promotes classification accuracy by increasing the degree of membership of breast lump sample and generic, reducing and realize enhancing model generalization ability with the degree of membership of non-belonging classification.
Description
Technical field
The present invention relates to technical field of image processing, especially X-ray breast image deep learning classification method.
Background technique
It is counted according to GLOBOCAN, the whole world increases about 12,000,000 cancer patients, cancer death's number about 8,200,000 newly every year.
Cancer causes greatly to bear to society, and especially in developing country, cancer influences personal lifestyle and life quality more tight
Weight.Breast cancer is to threaten one of most important malignant tumour of women life and health.Nipple correction inspection, because of its cost
It is low, to patient injure small, simple and easy to do and high resolution, meet medical requirement, become current breast cancer mainstream inspection method.X
Ray breast image provides very rich and intuitive lesion information for breast cancer diagnosis, and doctor is facilitated to check patient
Patient symptom is determined with analysis, is most widely used in breast cancer diagnosis.
The similitudes such as lump texture and other tissues such as muscle, body of gland are high in breast image;Breast lump edge mould
Paste is low with background discrimination;The color change of breast image concentrates in a relatively narrow region, is extracted using convolution operation
Characteristic similarity is high, it is difficult to distinguish, these factors both increase the difficulty of breast lump image feature extraction.Using CNN to cream
Gland image carries out feature extraction, since the variations such as the profile, texture, shape of breast lump are smaller, if using breast image
More convolution operations will cause the overdraft of image feature, reduce the quality of image feature.
Summary of the invention
Technical problem to be solved by the invention is to provide a kind of X-ray breast image deep learning classification method, its energy
The accurate breast lump differentiated in X-ray breast lump image is benign or pernicious.
X-ray breast image deep learning classification method, including 1) the multiple dimensioned convolution spy of breast lump image is extracted in design
Double calculating paths sub-network structure of sign;2) full convolution X-ray breast lump feature extraction and categorization module;3) it is based on degree of membership
Objective function optimization method.
Specific step is as follows:
1) double calculating paths sub-network, using two calculating paths, respectively using different size of convolution kernel to input
Breast lump image carries out the corresponding down-sampling operation of convolution sum, obtains two kinds of breast lump image convolution feature, then
Fusion Features are carried out by the method for superposition, obtain fused multiple dimensioned convolution feature;
2) the fused multiple dimensioned convolution feature for generating step 1) carries out feature extraction using full convolutional network;Tool
Body is fused multiple dimensioned convolution feature to be carried out convolution using the convolution kernel of two layers of each 1000 1*1 size, then input
Classify to Softmax layers, exports the benign pernicious classification information of breast lump;
3) classification degree of membership measure is introduced, new objective function is designed, the degree of membership of sample and cluster centre is believed
Breath is introduced into the design of objective function, by increasing the degree of membership of sample and generic, reducing the person in servitude with non-belonging classification
Category degree, the promotion of implementation model accuracy rate.
Double calculating Path Methods include 9 neural net layers, are four convolutional layers respectively, maximize under four
Down-sampling layer and a characteristic pattern fused layer;The X ray breast lump image of 200*200 size is separately input to two calculating
Convolution feature extraction is carried out in path, and two calculating path characteristic patterns generated are finally overlapped fusion, are merged
Feature afterwards.
The concrete operations that two calculating path characteristic patterns generated are overlapped fusion are to use two respectively
The convolution kernel of group 7*7 size and the convolution kernel of 5*5 size carry out convolutional calculation to the breast lump image of input, and in each volume
Lamination is followed by a 2*2 and maximizes down-sampling layer;After second maximization down-sampling layer, two are counted using the method for superposition
The breast lump characteristic pattern for calculating coordinates measurement is merged, and new fused feature is generated.
Described introduces degree of membership information objective function design, and new objective function is implemented as follows:
In objective function, degree of membership calculation formula between sample and classification:
uijRefer to the degree of membership between j-th of sample and ith cluster center;
Refer to the square value of the Euclidean distance at j-th of sample and ith cluster center;
C refers to classification number;
The intensity of membership that degree of membership measures sample to all classification is subordinate to when sample is closer from cluster centre
Intensity is bigger, otherwise it is smaller to be subordinate to intensity;Degree of membership value range is [0,1], and single sample is to the degree of membership of all classification
And be 1, the property formula of degree of membership is as follows:
C refers to classification number;
Degree of membership penalty term:
N refers to that training data concentrates sample size total number;
C refers to classification number;
η refers to the degree of membership penalty term weight of sample with other classifications in addition to clustering belonging to it, for controlling degree of membership
Contribution of the penalty term to training error;
hθ(xi) refer to the sample x under parameter θiPredicted value and generic degree of membership, i indicate i-th of sample;
It is the sample x under parameter θiPredicted value and j-th of non-belonging classification degree of membership, i indicate i-th of sample
This.
First item in formula is the penalty term of the degree of membership of predicted value and affiliated cluster;As shown in figure (2), y=-ln
(x) function monotone decreasing, for objective function, it is 1 that this, which inputs ideal value,.Pass through increasing it can be seen from functional digraph
The degree of membership of large sample and affiliated classification can reduce degree of membership error, make model into the class for reducing sample and affiliated classification away from
From direction evolution, the promotion of implementation model accuracy rate.
Section 2 in formula is the degree of membership penalty term between predicted value and other clusters in addition to affiliated cluster;Such as figure (2)
Shown, y=-ln (1-x) function monotonic increase, for objective function, it is 0 that this, which inputs ideal value,.It can by functional digraph
To find out, the degree of membership by reducing sample and non-belonging classification can reduce degree of membership error, make model to increase sample with
The direction evolution of the between class distance of non-belonging classification, the generalization ability of lift scheme.
Cross entropy degree of membership objective function:
J (θ)=CrossEntropy+ λ S (θ) (4)
CrossEntropy is to intersect entropy function, the difference for predictive metrics value and sample label;
λ is degree of membership punishment term coefficient, for adjusting the size of degree of membership penalty term contribution.
By adopting the above-described technical solution, compared with prior art, the present invention is devised from image processing point and is penetrated to X
Line breast lump image carries out automatic sorter network, which uses two calculating paths to X-ray breast lump image first
Convolution is carried out using different size of convolution kernel and down-sampling operates, and the convolution characteristic pattern of different scale type is extracted, by two
The characteristic pattern for calculating path input is overlapped fusion, obtains double calculating fused characteristic informations in path.Again to fusion feature
Feature extraction is carried out using full convolutional network, Softmax classification layer is finally sent into and classifies to feature, obtain breast lump
Image classification result.During model training, the mesh based on degree of membership for being suitable for X-ray breast lump image classification is used
Scalar functions are trained model, and new objective function is by increasing the degree of membership of breast lump sample and generic, reducing
Enhancing model generalization ability is realized with the degree of membership of non-belonging classification, promotes classification accuracy.
Detailed description of the invention
Fig. 1 is the multiple dimensioned convolution feature extraction network structure of breast image of the invention;
Fig. 2 is breast image depth of assortment learning network structure of the invention;
Fig. 3 is error in classification mapping function.
Specific embodiment
The embodiment of the present invention: X-ray breast image deep learning classification method, including 1) breast lump is extracted in design
Double calculating paths sub-network structure of the multiple dimensioned convolution feature of image;2) full convolution X-ray breast lump feature extraction and classification
Module;3) the objective function optimization method based on degree of membership.
Specific step is as follows:
1) double calculating paths sub-network, using two calculating paths, respectively using different size of convolution kernel to input
Breast lump image carries out the corresponding down-sampling operation of convolution sum, obtains two kinds of breast lump image convolution feature, then
Fusion Features are carried out by the method for superposition, obtain fused multiple dimensioned convolution feature;
2) the fused multiple dimensioned convolution feature for generating step 1) carries out feature extraction using full convolutional network;Tool
Body is fused multiple dimensioned convolution feature to be carried out convolution using the convolution kernel of two layers of each 1000 1*1 size, then input
Classify to Softmax layers, exports the benign pernicious classification information of breast lump;
3) classification degree of membership measure is introduced, new objective function is designed, the degree of membership of sample and cluster centre is believed
Breath is introduced into the design of objective function, by increasing the degree of membership of sample and generic, reducing the person in servitude with non-belonging classification
Category degree, the promotion of implementation model accuracy rate.
Double calculating Path Methods include 9 neural net layers, are four convolutional layers respectively, maximize under four
Down-sampling layer and a characteristic pattern fused layer;The X ray breast lump image of 200*200 size is separately input to two calculating
Convolution feature extraction is carried out in path, and two calculating path characteristic patterns generated are finally overlapped fusion, are merged
Feature afterwards.
The concrete operations that two calculating path characteristic patterns generated are overlapped fusion are to use two respectively
(two groups of 7*7, two groups of 5*5, two groups altogether, wherein one group only uses 7*7 size convolution kernel, another group only uses the big rouleau of 5*5 to group
Product core) convolution kernel of 7*7 size and the convolution kernel of 5*5 size carry out convolutional calculation to the breast lump image of input, and every
A convolutional layer is followed by a 2*2 and maximizes down-sampling layer;After second maximization down-sampling layer, using the method for superposition by two
The breast lump characteristic pattern that item calculates coordinates measurement is merged, and new fused feature is generated.
Described introduces degree of membership information objective function design, and new objective function is implemented as follows:
In objective function, degree of membership calculation formula between sample and classification:
uijRefer to the degree of membership between j-th of sample and ith cluster center;
Refer to the square value of the Euclidean distance at j-th of sample and ith cluster center;
C refers to classification number;
The intensity of membership that degree of membership measures sample to all classification is subordinate to when sample is closer from cluster centre
Intensity is bigger, otherwise it is smaller to be subordinate to intensity;Degree of membership value range is [0,1], and single sample is to the degree of membership of all classification
And be 1, the property formula of degree of membership is as follows:
C refers to classification number;
Degree of membership penalty term:
N refers to that training data concentrates sample size total number;
C refers to classification number;
η refers to the degree of membership penalty term weight of sample with other classifications in addition to clustering belonging to it, for controlling degree of membership
Contribution of the penalty term to training error;
hθ(xi) refer to the sample x under parameter θiPredicted value and generic degree of membership, i indicate i-th of sample;
It is the sample x under parameter θiPredicted value and j-th of non-belonging classification degree of membership, i indicate i-th of sample
This.
First item in formula is the penalty term of the degree of membership of predicted value and affiliated cluster;As shown in figure (2), y=-ln
(x) function monotone decreasing, for objective function, it is 1 that this, which inputs ideal value,.Pass through increasing it can be seen from functional digraph
The degree of membership of large sample and affiliated classification can reduce degree of membership error, make model into the class for reducing sample and affiliated classification away from
From direction evolution, the promotion of implementation model accuracy rate.
Section 2 in formula is the degree of membership penalty term between predicted value and other clusters in addition to affiliated cluster;Such as figure (2)
Shown, y=-ln (1-x) function monotonic increase, for objective function, it is 0 that this, which inputs ideal value,.It can by functional digraph
To find out, the degree of membership by reducing sample and non-belonging classification can reduce degree of membership error, make model to increase sample with
The direction evolution of the between class distance of non-belonging classification, the generalization ability of lift scheme.
Cross entropy degree of membership objective function:
J (θ)=CrossEntropy+ λ S (θ) (4)
CrossEntropy is to intersect entropy function, the difference for predictive metrics value and sample label;
λ is degree of membership punishment term coefficient, for adjusting the size of degree of membership penalty term contribution.
Claims (5)
1. a kind of X-ray breast image deep learning classification method, it is characterised in that: extract breast lump image including 1) design
Double calculating paths sub-network structure of multiple dimensioned convolution feature;2) full convolution X-ray breast lump feature extraction and categorization module;
3) the objective function optimization method based on degree of membership.
2. X-ray breast image deep learning classification method according to claim 1, which is characterized in that specific steps are such as
Under:
1) double calculating paths sub-network, using two calculating paths, respectively using different size of convolution kernel to the mammary gland of input
Lump image carries out the corresponding down-sampling operation of convolution sum, obtains two kinds of breast lump image convolution feature, then pass through
The method of superposition carries out Fusion Features, obtains fused multiple dimensioned convolution feature;
2) the fused multiple dimensioned convolution feature for generating step 1) carries out feature extraction using full convolutional network;Specifically,
Fused multiple dimensioned convolution feature is subjected to convolution using the convolution kernel of two layers of each 1000 1*1 size, then is input to
Softmax layers are classified, and the benign pernicious classification information of breast lump is exported;
3) classification degree of membership measure is introduced, new objective function is designed, the degree of membership information of sample and cluster centre is drawn
Enter into the design of objective function, by increasing the degree of membership of sample and generic, reducing the degree of membership with non-belonging classification,
The promotion of implementation model accuracy rate.
3. X-ray breast image deep learning classification method according to claim 2, which is characterized in that it is characterized by:
Double calculating Path Methods include 9 neural net layers, are four convolutional layers respectively, maximize down-sampling layer under four
With a characteristic pattern fused layer;The X-ray breast lump image of 200*200 size is separately input to carry out in two calculating paths
Two calculating path characteristic patterns generated are finally overlapped fusion, obtain fused feature by convolution feature extraction.
4. X-ray breast image deep learning classification method according to claim 3, which is characterized in that it is characterized by:
The concrete operations that two calculating path characteristic patterns generated are overlapped fusion are, big using two groups of 7*7 respectively
The convolution kernel of small convolution kernel and 5*5 size carries out convolutional calculation to the breast lump image of input, and after each convolutional layer
It meets a 2*2 and maximizes down-sampling layer;After second maximization down-sampling layer, using the method for superposition by two calculating paths
The breast lump characteristic pattern of generation is merged, and new fused feature is generated.
5. X-ray breast image deep learning classification method according to claim 2, which is characterized in that it is characterized by:
Described introduces degree of membership information objective function design, and new objective function is implemented as follows:
In objective function, degree of membership calculation formula between sample and classification:
uijRefer to the degree of membership between j-th of sample and ith cluster center;
Refer to the square value of the Euclidean distance at j-th of sample and ith cluster center;
C refers to classification number;
The intensity of membership that degree of membership measures sample to all classification is subordinate to intensity when sample is closer from cluster centre
It is bigger, on the contrary it is smaller to be subordinate to intensity;Degree of membership value range is [0,1], and single sample to all classification degree of membership sum
It is 1, the property formula of degree of membership is as follows:
C refers to classification number;
Degree of membership penalty term:
N refers to that training data concentrates sample size total number;
C refers to classification number;
η refers to the degree of membership penalty term weight of sample with other classifications in addition to clustering belonging to it, for controlling degree of membership punishment
Contribution of the item to training error;
hθ(xi) refer to the sample x under parameter θiPredicted value and generic degree of membership, i indicate i-th of sample;
It is the sample x under parameter θiPredicted value and j-th of non-belonging classification degree of membership, i indicate i-th of sample.
First item in formula is the penalty term of the degree of membership of predicted value and affiliated cluster;As shown in figure (2), y=-ln (x) function
Monotone decreasing, for objective function, it is 1 that this, which inputs ideal value,.It can be seen from functional digraph by increase sample with
The degree of membership of affiliated classification can reduce degree of membership error, make model to the direction for the inter- object distance for reducing sample and affiliated classification
Evolution, the promotion of implementation model accuracy rate.
Section 2 in formula is the degree of membership penalty term between predicted value and other clusters in addition to affiliated cluster;Such as figure (2) institute
Show, y=-ln (1-x) function monotonic increase, for objective function, it is 0 that this, which inputs ideal value,.It can be with by functional digraph
Find out, the degree of membership by reducing sample and non-belonging classification can reduce degree of membership error, make model to increase sample with it is non-
The direction evolution of the between class distance of affiliated classification, the generalization ability of lift scheme.
Cross entropy degree of membership objective function:
J (θ)=CrossEntropy+ λ S (θ) (4)
CrossEntropy is to intersect entropy function, the difference for predictive metrics value and sample label;
λ is degree of membership punishment term coefficient, for adjusting the size of degree of membership penalty term contribution.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910278910.9A CN110232396B (en) | 2019-04-09 | 2019-04-09 | X-ray mammary gland image deep learning classification method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910278910.9A CN110232396B (en) | 2019-04-09 | 2019-04-09 | X-ray mammary gland image deep learning classification method |
Publications (2)
Publication Number | Publication Date |
---|---|
CN110232396A true CN110232396A (en) | 2019-09-13 |
CN110232396B CN110232396B (en) | 2022-07-01 |
Family
ID=67860664
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910278910.9A Active CN110232396B (en) | 2019-04-09 | 2019-04-09 | X-ray mammary gland image deep learning classification method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110232396B (en) |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110782444A (en) * | 2019-10-25 | 2020-02-11 | 深圳技术大学 | Holographic microwave breast lump identification method and identification system |
CN113139931A (en) * | 2021-03-17 | 2021-07-20 | 杭州迪英加科技有限公司 | Thyroid slice image classification model training method and device |
CN113177559A (en) * | 2021-04-22 | 2021-07-27 | 重庆兆光科技股份有限公司 | Image recognition method, system, device and medium combining breadth and dense convolutional neural network |
CN115423806A (en) * | 2022-11-03 | 2022-12-02 | 南京信息工程大学 | Breast mass detection method based on multi-scale cross-path feature fusion |
Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20080234578A1 (en) * | 2007-03-19 | 2008-09-25 | General Electric Company | Multi-modality mammography reconstruction method and system |
CN103425986A (en) * | 2013-08-31 | 2013-12-04 | 西安电子科技大学 | Breast lump image feature extraction method based on edge neighborhood weighing |
CN104008386A (en) * | 2014-05-13 | 2014-08-27 | 中国科学院深圳先进技术研究院 | Method and system for identifying type of tumor |
US20160066872A1 (en) * | 2008-12-08 | 2016-03-10 | Hologic, Inc. | Displaying Computer-Aided Detection Information With Associated Breast Tomosynthesis Image Information |
CN108464840A (en) * | 2017-12-26 | 2018-08-31 | 安徽科大讯飞医疗信息技术有限公司 | A kind of breast lump automatic testing method and system |
CN108765387A (en) * | 2018-05-17 | 2018-11-06 | 杭州电子科技大学 | Based on Faster RCNN mammary gland DBT image lump automatic testing methods |
CN108830282A (en) * | 2018-05-29 | 2018-11-16 | 电子科技大学 | A kind of the breast lump information extraction and classification method of breast X-ray image |
CN109409413A (en) * | 2018-09-28 | 2019-03-01 | 贵州大学 | X-ray breast lump image automatic classification method |
-
2019
- 2019-04-09 CN CN201910278910.9A patent/CN110232396B/en active Active
Patent Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20080234578A1 (en) * | 2007-03-19 | 2008-09-25 | General Electric Company | Multi-modality mammography reconstruction method and system |
US20160066872A1 (en) * | 2008-12-08 | 2016-03-10 | Hologic, Inc. | Displaying Computer-Aided Detection Information With Associated Breast Tomosynthesis Image Information |
CN103425986A (en) * | 2013-08-31 | 2013-12-04 | 西安电子科技大学 | Breast lump image feature extraction method based on edge neighborhood weighing |
CN104008386A (en) * | 2014-05-13 | 2014-08-27 | 中国科学院深圳先进技术研究院 | Method and system for identifying type of tumor |
CN108464840A (en) * | 2017-12-26 | 2018-08-31 | 安徽科大讯飞医疗信息技术有限公司 | A kind of breast lump automatic testing method and system |
CN108765387A (en) * | 2018-05-17 | 2018-11-06 | 杭州电子科技大学 | Based on Faster RCNN mammary gland DBT image lump automatic testing methods |
CN108830282A (en) * | 2018-05-29 | 2018-11-16 | 电子科技大学 | A kind of the breast lump information extraction and classification method of breast X-ray image |
CN109409413A (en) * | 2018-09-28 | 2019-03-01 | 贵州大学 | X-ray breast lump image automatic classification method |
Non-Patent Citations (5)
Title |
---|
SEUNG YEON SHIN 等: "Joint Weakly and Semi-Supervised Deep Learning for Localization and Classification of Masses in Breast Ultrasound Images", 《IEEE TRANSACTIONS ON MEDICAL IMAGING》 * |
XIAOYONG ZHANG 等: "Classification of mammographic masses by deep learning", 《2017 56TH ANNUAL CONFERENCE OF THE SOCIETY OF INSTRUMENT AND CONTROL ENGINEERS OF JAPAN (SICE)》 * |
侯丽: "基于AdaBoost和深度学习的红外乳腺癌检测方法研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
孙利雷 等: "基于深度学习的乳腺X射线影像分类方法研究", 《计算机工程与应用》 * |
韩哲: "基于卷积神经网络的乳腺肿瘤良恶性鉴定技术研究", 《中国优秀硕士学位论文全文数据库 医药卫生科技辑》 * |
Cited By (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110782444A (en) * | 2019-10-25 | 2020-02-11 | 深圳技术大学 | Holographic microwave breast lump identification method and identification system |
CN113139931A (en) * | 2021-03-17 | 2021-07-20 | 杭州迪英加科技有限公司 | Thyroid slice image classification model training method and device |
CN113139931B (en) * | 2021-03-17 | 2022-06-03 | 杭州迪英加科技有限公司 | Thyroid section image classification model training method and device |
CN113177559A (en) * | 2021-04-22 | 2021-07-27 | 重庆兆光科技股份有限公司 | Image recognition method, system, device and medium combining breadth and dense convolutional neural network |
CN115423806A (en) * | 2022-11-03 | 2022-12-02 | 南京信息工程大学 | Breast mass detection method based on multi-scale cross-path feature fusion |
CN115423806B (en) * | 2022-11-03 | 2023-03-24 | 南京信息工程大学 | Breast mass detection method based on multi-scale cross-path feature fusion |
Also Published As
Publication number | Publication date |
---|---|
CN110232396B (en) | 2022-07-01 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN108364006B (en) | Medical image classification device based on multi-mode deep learning and construction method thereof | |
Xue et al. | An application of transfer learning and ensemble learning techniques for cervical histopathology image classification | |
CN110232396A (en) | X-ray breast image deep learning classification method | |
Shah et al. | A robust approach for brain tumor detection in magnetic resonance images using finetuned efficientnet | |
CN109034045A (en) | A kind of leucocyte automatic identifying method based on convolutional neural networks | |
CN111192245A (en) | Brain tumor segmentation network and method based on U-Net network | |
Shen et al. | Simultaneous segmentation and classification of mass region from mammograms using a mixed-supervision guided deep model | |
CN107748900A (en) | Tumor of breast sorting technique and device based on distinction convolutional neural networks | |
CN106683081A (en) | Brain glioma molecular marker nondestructive prediction method and prediction system based on radiomics | |
CN110265095A (en) | For HCC recurrence and construction method and the application of the prediction model and nomogram of RFS | |
CN102737379A (en) | Captive test (CT) image partitioning method based on adaptive learning | |
CN104834943A (en) | Brain tumor classification method based on deep learning | |
CN106096654A (en) | A kind of cell atypia automatic grading method tactful based on degree of depth study and combination | |
CN102096804A (en) | Method for recognizing image of carcinoma bone metastasis in bone scan | |
Jaszcz et al. | Lung x-ray image segmentation using heuristic red fox optimization algorithm | |
CN108492877A (en) | A kind of cardiovascular disease auxiliary prediction technique based on DS evidence theories | |
CN110379509A (en) | A kind of Breast Nodules aided diagnosis method and system based on DSSD | |
CN110097921A (en) | Allelic heterogeneity visualization quantitative approach and system in glioma based on image group | |
Luo et al. | Classification of tumor in one single ultrasound image via a novel multi-view learning strategy | |
CN110444294A (en) | A kind of prostate cancer aided analysis method and equipment based on layered perception neural networks | |
Biradar et al. | Lung Cancer Detection and Classification using 2D Convolutional Neural Network | |
CN109671060A (en) | Area of computer aided breast lump detection method based on selective search and CNN | |
CN104463885B (en) | A kind of Multiple Sclerosis lesions region segmentation method | |
CN114398979A (en) | Ultrasonic image thyroid nodule classification method based on feature decoupling | |
CN108765411A (en) | A kind of tumor classification method based on image group |
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 | ||
GR01 | Patent grant | ||
GR01 | Patent grant |