CN110175979A - A kind of Lung neoplasm classification method based on collaboration attention mechanism - Google Patents
A kind of Lung neoplasm classification method based on collaboration attention mechanism Download PDFInfo
- Publication number
- CN110175979A CN110175979A CN201910276013.4A CN201910276013A CN110175979A CN 110175979 A CN110175979 A CN 110175979A CN 201910276013 A CN201910276013 A CN 201910276013A CN 110175979 A CN110175979 A CN 110175979A
- Authority
- CN
- China
- Prior art keywords
- data
- network
- attention mechanism
- attention
- lung
- 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
- G06F18/24—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
- G06T7/0014—Biomedical image inspection using an image reference approach
-
- 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
- G06T2207/30061—Lung
- G06T2207/30064—Lung nodule
Abstract
The invention discloses a kind of Lung neoplasm classification method based on attention mechanism, main target is classified to the benign malicious nodes of Lung neoplasm.The present invention the following steps are included: 1, to data set carry out data prediction;2, it designs the Lung neoplasm sorter network structure 3 based on attention mechanism, integrated classification is carried out to final result.The present invention proposes a kind of model for Lung neoplasm classification, the new method proposed based on attention mechanism.And obtain the extraordinary effect given up in field in Lung neoplasm at present.The present invention schemes the CT of low dosage, carries out the classification of the benign malicious nodes of Lung neoplasm by merging attention mechanism using deep learning network.For computer-aided diagnosis (CAD) system, which is embedded into CAD, can play auxiliary and enhance the effect of the work of radiologist.
Description
Technical field
The present invention relates to a kind of based on Lung neoplasm classification (the Pulmonary Nodules for cooperateing with attention mechanism
Classification, PNC) deep neural network, greatly improve the accuracy rate of current Lung neoplasm classification.
Background technique
In recent years, scientist achieves very big breakthrough in depth learning technology, for example, the appearance of AlexNet is very big
Improve the ability of computer identification picture.Compared with general machine learning, deep learning can automatically extract richer from data
Rich, useful information, thus have higher accuracy.Meanwhile the rapid promotion and the increase of data available of computer performance,
Make it possible the training of deep learning network.Therefore, depth learning technology is just gradually applied in cancer detection.
Traditionally, researcher is explored hand-made feature and is predicted the classification of tubercle using classifier.Nowadays, deep
Degree learning art, especially convolutional neural networks (CNNs) achieve huge success in various high-level vision understanding tasks,
Many researchs also turn out, more more accurate than general machine learning method, either intestinal cancer, breast cancer, lung cancer or cancer of pancreas
Detection.Deep learning is used for medical imaging analysis, such as X-ray, CT image, or the number for analyzing molecules level
According to, such as gene mutation, gene expression data.Currently, depth learning technology can't be applied in all types of cancers, because
This existing research is generally by common cancers such as lung cancer, breast cancer, as detection target.
Currently, lung cancer has become the threat got worse to human lives.Due to the increasing of air pollution and smoking population
Add, the disease incidence of lung cancer is growing day by day, and the death toll as caused by lung cancer is increasing.Therefore, the detection and classification of Lung neoplasm
It is of crucial importance for the diagnosing and treating of the early stage of lung cancer.The early stage of lung cancer, can not be with the resolution ratio of rabat (x-ray) due to small volume
It was found that.So rabat is limited to the value of screening lung cancer, mainly or to help to examine by computed tomography (CT)
It is disconnected.The screening lung cancer means recommended in the world are mainly low-dose CT (LDCT).As its name suggests, exactly in traditional CT examination
On the basis of reduce roentgen dose X.Since lung's gassiness is more, low density feature, so that can still be obtained after adjustment dosage
Satisfied lung's imaging, compared with rabat, the resolution ratio of LDCT and sensitivity are greatly improved, it can be found that diameter is less than 5mm's
Minimal disease.
In conjunction with deep learning in the development of medicine and the harmfulness of lung cancer, patient's treating as early as possible can be allowed, this hair
It is bright to put forward a solution, a method of the Lung neoplasm classification based on attention mechanism.To the CT figure of low dosage, pass through
It merges attention mechanism and carries out the classification of the benign malicious nodes of Lung neoplasm using deep learning network.For area of computer aided
(CAD) system of diagnosis, the algorithm are embedded into CAD, can play auxiliary and enhance the effect of the work of radiologist.
Summary of the invention
In view of the deficiencies of the prior art, it is an object of the present invention to provide one kind to carry out Lung neoplasm classification based on attention mechanism
Method.We are classified experiment using LUNA16 data set and LIDC-IDRI data set.
Step (1), data processing
Data processing is divided into two parts:
First part pre-processes raw data set.
Second part carries out data augmentation to the data handled well.
Data prediction: being mapped to LIDC-IDRI for the data in LUNA16, and data set is pre-processed, that is, go unless
CT figure is simultaneously normalized to rgb space by normal data, is then stacked into 3D rendering;
Data augmentation: map function is carried out to the data handled well, data is expanded;
Step (2), Lung neoplasm sorter network structure of the design based on attention mechanism
As shown in Figure 1, method proposes a kind of novel deep learnings using attention mechanism to go to carry out Lung neoplasm
Method.This method can input the lung tumors node for needing to classify, and export the tumour and be benign or pernicious.
The data for marking and having handled well with tumor region are input in network, finally obtain one two points
As a result, the i.e. tumour is benign or pernicious;
Step (3), node-classification
As a result, be weighted summation, final Fusion Model result will be obtained by different attention mechanism.
Data processing described in step (1), specific as follows:
LUNA16 is that the slice thickness CT scan inconsistent less than 2.5 millimeters and slice thickness is removed in LIDC-IDRI
Part, is divided into 10 sub-folders by 888 in total.In LUNA data set, it is only labelled with the node for being directly greater than 3mm, therefore,
The CT figure for taking out 888 from LIDC-IDRI using the data name in LUNA, is divided into 10 parts, every part of CT number is equal.For this
Data set does 5 folding cross validations.Take a copy of it as test set, remaining nine parts are used as training set.
Initial data is loaded in network, resampling is carried out to it, resolution ratio is unified.Data are trimmed to-
Between 1200~600, value greater than 600 is 600, and the value less than -1200 is -1200, then by data normalization to 0~
255.The lung's mask provided by LUNA16 takes out left lung and right lung and cuts out data by rectangle according to lung edge, removes
Most of data unrelated with lung, and node coordinate is recalculated according to new data.By every CT figure all in accordance with slice side
To building up 3 dimension datas.Using node as center coordinate, it is cut into the three-dimensional data of 32 × 32 × 32 sizes.
Data augmentation: when data are loaded into network, size is added in data and is 4 padding, then is cut at random
Picture by the overturning of data Random Level, flip vertical, mirror image switch, and is taken size by the matrix of 32 × 32 × 32 sizes at random
For 4 matrix zero setting, increases data volume, improve the generalization ability and robustness of model.
Lung tumors node-classification network structure design described in step (2) based on attention mechanism, specific as follows:
Network is divided into two parts, and a part is the 3D sorter network on basis, and a part is the 3D that attention mechanism is added
Sorter network.
As shown in table 1, basic 3D sorter network is by 3D two-way network module (Dual Path Block, DPB) and common
3D CNN composition 3D two-way network (Dual PathNetwork, DPN).Comprising several as shown in Figure 2 residual in 3D DPB
Difference network connection is added, and third and the 4th DPB are respectively labeled as feat8 and feat4.3D based on attention mechanism
Sorter network is which are added attention networks.Addition is based on context attention network (Context after feat8
AttentionNetwork, CoAN) it is denoted as CoA8, wherein CAN such as Fig. 3 is indicated.
CoAN calculation is as follows:
zi=γ oi+xiFormula (1)
Wherein, oiIt is the degree of correlation for the position the i and entire x for inputting x, wherein γ is weight parameter, initial value 0, and
It automatically updates in the training process.
The network (Spatial Attention Network, SpAN) based on spatial attention is added after feat4 to be denoted as
SpA4.The calculation formula of SAN is as follows
Wherein, ai,j,kIt is learnt spatial attention (Spatial Attention, SpA), Xi,j,kIt is upper one layer of biography
The feature entered, i therein, j, k indicate the feature on each channel in the position of three-dimensional.
Node described in step (3), specific as follows:
It is to be denoted as M by the output result of 3D DPN1, the output result of 3D DPN+CoA8 is denoted as M2, by 3D DPN+SpA4
Output result be denoted as M3, the output result of 3D DPN+CoN8+SpA4 is denoted as M4, four output as a result, do weighted sum,
Obtain it is final as a result,
Formula is as follows:
Wherein, αiIt is the weight assigned according to the accuracy of model training, weight is all set as 1 here.
The present invention has the beneficial effect that:
The present invention schemes the CT of low dosage, carries out Lung neoplasm by merging attention mechanism using deep learning network
The classification of benign malicious nodes.For computer-aided diagnosis (CAD) system, which is embedded into CAD, can play auxiliary
And enhance the effect of the work of radiologist.
Detailed description of the invention
Fig. 1 is the Lung neoplasm sorter network block schematic illustration based on attention mechanism;
Fig. 2 is 3D two-way network structure (Dual Path Block, DPB) block schematic illustration
Fig. 3 is based on context attention (ContextAttention, CoA) block schematic illustration
Fig. 4 is based on spatial attention (Spatial Attention, SpA) block schematic illustration
Table 1 is the Lung neoplasm sorter network frame details based on attention mechanism
Table 2 is the final classification results of each model and evaluation criterion
Specific implementation details
The following further describes the present invention with reference to the drawings.
As shown in Figure 1, the frame that the Lung neoplasm based on attention mechanism is classified,
3D DPB as shown in Figure 2 is the main comprising modules of DPN, and basic module formula can be expressed as follows:
Y=ReLU ([x [: d];F(x);F(x)[:d]+x[d:]]);Formula (4)
Wherein x expression input vector, F expression convolution operation, d expression depth, i.e. F (x) [: d] it is attended operation, F (x)
[d :] does phase add operation.
As shown in 3D DPN table in table 1, c indicates that port number, d indicate the depth in DPB, and n indicates to contain in this DPB
N basic module.So basic network, that is, 3D DPN (M1) such as Contextual attentionmechanism in table 1
Shown, third DPB in 3D DPN network is feat8 by we, and the 4th DPB is denoted as feat4.
As shown in Fig. 3 the network of context attention network (Contextual AttentionNetwork, CoAN)
Block schematic illustration, the output after feat8 is input in CoAN by we, is passed through three 1 × 1 × 1 convolution respectively, is denoted as Wf,
Wg, Wh, the operation by the convolution is denoted as f (x)=Wf, g (x)=Wg, h (x)=Wh, and pass through following operation as shown in Figure 2:
It is input in CoAN,
Attention block (Attention Cube) is arrived by what is once operated first
Wherein βjtiIt indicates on i point to be the attention block in Fig. 3 to the degree of correlation of j.
The output of attention block and Attention can be calculated by the following formula and obtained:
The output of Attention is added with member input, obtains the output result of final CAN are as follows:
zi=γ oi+xiFormula (7)
It is so that we obtain the 3D DPN network structure (CA8 (M based on context attention mechanism2))。
It is the network frame of spatial attention network (Spatial AttentionNetwork, SpAN) as shown in Figure 4
Frame schematic diagram, the output after feat4 is input in SpAN by we, and input is denoted as Xi,j,k, it is entered into the calculating of SAN
Formula is as follows
Wherein a is learnt Attention Cube, and so we obtain the 3D based on spatial attention mechanism
DPN network mechanism (SA4 (M3))
As shown in table 2, we will calculate last as a result, the result of each model is first carried out two points by us, less than 0.5
It is benign node, is malicious nodes greater than 0.5.Then the final result of each model is subjected to ballot fusion
(Ensemble) as a result, here using majority voting method, i.e., the weight of each model is 1.
Table 1
Table 2.
Claims (6)
1. a kind of Lung neoplasm classification method based on collaboration attention mechanism, it is characterised in that include the following steps:
Step (1), data processing
Data processing is divided into two parts:
First part pre-processes raw data set;
Second part carries out data augmentation to the data handled well;
Data prediction: being mapped to LIDC-IDRI for the data in LUNA16, and data set is pre-processed, that is, is removed improper
CT figure is simultaneously normalized to rgb space by data, is then stacked into 3D rendering;
Data augmentation: map function is carried out to the data handled well, data is expanded;
Step (2), Lung neoplasm sorter network structure of the design based on attention mechanism;
The data for marking and having handled well with tumor region are input in network, one two points of knot is finally obtained
Fruit, the i.e. tumour are benign or pernicious;
Step (3), node-classification
As a result, be weighted summation, final Fusion Model result will be obtained by different attention mechanism;
Data processing described in step (1), specific as follows:
LUNA16 is that the slice thickness CT scan part inconsistent less than 2.5 millimeters and slice thickness is removed in LIDC-IDRI, always
Totally 888, it is divided into 10 sub-folders;In LUNA data set, therefore node of the diameter dimension greater than 3mm utilizes LUNA
In data name taken out from LIDC-IDRI 888 CT figure, be divided into 10 parts, every part of CT number is equal;For the data set,
Do 5 folding cross validations;Take a copy of it as test set, remaining nine parts are used as training set;
Initial data is loaded into network, resampling is carried out to it, resolution ratio is unified;Data are trimmed to -1200~
Between 600, the value greater than 600 is 600, and the value less than -1200 is -1200, then by data normalization to 0~255;Pass through
Lung's mask that LUNA16 is provided takes out left lung and right lung, according to lung edge, cuts out data by rectangle, remove it is most of with
The unrelated data of lung, and node coordinate is recalculated according to new data;By every CT figure all in accordance with slice direction, 3 are built up
Dimension data;Using node as center coordinate, it is cut into the three-dimensional data of 32 × 32 × 32 sizes;
Data augmentation: when data are loaded into network, being added size for data and be 4 padding, then it is cut into 32 at random ×
The matrix of 32 × 32 sizes takes size for 4 at random by the overturning of data Random Level, flip vertical, mirror image switch, and by picture
Matrix zero setting increases data volume.
2. a kind of Lung neoplasm classification method based on collaboration attention mechanism according to claim 1, it is characterised in that step
Suddenly the lung tumors node-classification network structure design described in (2) based on attention mechanism, specific as follows:
Network is divided into two parts, and a part is the 3D sorter network on basis, and a part is that the 3D classification of attention mechanism is added
Network;
The 3D sorter network on basis forms 3D two-way network DPN by 3D two-way network module DPB and common 3D CNN;3D
It is connected to the network or is added comprising several residual errors in DPB, third and the 4th DPB are respectively labeled as feat8 and feat4;Base
In attention mechanism 3D sorter network which are added attention networks;It adds after feat8 and is paid attention to based on context
Power network C oAN, is denoted as CoA8;
CoAN calculation is as follows:
zi=γ oi+xiFormula (1)
Wherein, oiIt is the degree of correlation for the position the i and entire x for inputting x, wherein γ is weight parameter, initial value 0, and in training
It automatically updates in the process;
The network SpAN based on spatial attention is added after feat4, is denoted as SpA4;The calculation formula of SAN is as follows
Wherein, ai,j,kIt is learnt spatial attention SpA, Xi,j,kIt is upper one layer of incoming feature, i therein, j, k expression is often
Feature on a channel is in the position of three-dimensional.
3. a kind of Lung neoplasm classification method based on collaboration attention mechanism according to claim 2, it is characterised in that step
Suddenly node described in (3), specific as follows:
It is to be denoted as M by the output result of 3D DPN1, the output result of 3D DPN+CoA8 is denoted as M2, by the defeated of 3D DPN+SpA4
Result is denoted as M out3, the output result of 3D DPN+CoN8+SpA4 is denoted as M4, four output as a result, do weighted sum, obtain
It is final as a result,
Formula is as follows:
Wherein αiIt is the weight assigned according to the accuracy of model training, weight is all set as 1 here.
4. a kind of Lung neoplasm classification method based on collaboration attention mechanism according to claim 3, it is characterised in that institute
The 3D DPB stated is the main comprising modules of DPN, and basic module formula is expressed as follows:
Y=ReLU ([x [: d];F(x);F(x)[:d]+x[d:]]);Formula (4)
Wherein, x expression input vector, F expression convolution operation, d expression depth, i.e. F (x) [: d] it is attended operation, F (x) [d :]
Do phase add operation.
5. a kind of Lung neoplasm classification method based on collaboration attention mechanism according to claim 4, it is characterised in that will
Output after feat8, is input in CoAN, passes through three 1 × 1 × 1 convolution respectively, is denoted as Wf, Wg, Wh, by the convolution
Operation is denoted as f (x)=Wf, g (x)=Wg, h (x)=Wh, and perform the following operation: it is input in CoAN,
Attention block is arrived by what is once operated first
Wherein, βJ, iIndicate degree of correlation, that is, attention block on i point to j;
The output of attention block and Attention are calculated by the following formula and obtained:
The output of Attention is added with member input, obtains the output result of final CAN are as follows:
zi=γ oi+xiFormula (7)
To obtain the 3D DPN network structure based on context attention mechanism.
6. a kind of Lung neoplasm classification method based on collaboration attention mechanism according to claim 5, it is characterised in that will
The result of each model carries out two points, is benign node less than 0.5, is malicious nodes greater than 0.5;Then by each model
Final result carry out ballot fusion results, using majority voting method, i.e., the weight of each model is 1.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910276013.4A CN110175979B (en) | 2019-04-08 | 2019-04-08 | Lung nodule classification method based on cooperative attention mechanism |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910276013.4A CN110175979B (en) | 2019-04-08 | 2019-04-08 | Lung nodule classification method based on cooperative attention mechanism |
Publications (2)
Publication Number | Publication Date |
---|---|
CN110175979A true CN110175979A (en) | 2019-08-27 |
CN110175979B CN110175979B (en) | 2021-07-27 |
Family
ID=67689437
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910276013.4A Active CN110175979B (en) | 2019-04-08 | 2019-04-08 | Lung nodule classification method based on cooperative attention mechanism |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110175979B (en) |
Cited By (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110570417A (en) * | 2019-09-12 | 2019-12-13 | 慧影医疗科技(北京)有限公司 | Pulmonary nodule classification method and device and image processing equipment |
CN110766051A (en) * | 2019-09-20 | 2020-02-07 | 四川大学华西医院 | Lung nodule morphological classification method based on neural network |
CN110807764A (en) * | 2019-09-20 | 2020-02-18 | 成都智能迭迦科技合伙企业(有限合伙) | Lung cancer screening method based on neural network |
CN111626969A (en) * | 2020-05-22 | 2020-09-04 | 东北易华录信息技术有限公司 | Attention mechanism-based corn disease image processing method |
CN112419307A (en) * | 2020-12-11 | 2021-02-26 | 长春工业大学 | Lung nodule benign and malignant identification method based on attention mechanism |
CN113782181A (en) * | 2021-07-26 | 2021-12-10 | 杭州深睿博联科技有限公司 | CT image-based lung nodule benign and malignant diagnosis method and device |
Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106650830A (en) * | 2017-01-06 | 2017-05-10 | 西北工业大学 | Deep model and shallow model decision fusion-based pulmonary nodule CT image automatic classification method |
WO2018094294A1 (en) * | 2016-11-18 | 2018-05-24 | Salesforce.Com, Inc. | Spatial attention model for image captioning |
CN108388900A (en) * | 2018-02-05 | 2018-08-10 | 华南理工大学 | The video presentation method being combined based on multiple features fusion and space-time attention mechanism |
CN108537793A (en) * | 2018-04-17 | 2018-09-14 | 电子科技大学 | A kind of pulmonary nodule detection method based on improved u-net networks |
CN108629764A (en) * | 2018-04-17 | 2018-10-09 | 杭州依图医疗技术有限公司 | A kind of good pernicious method and device of determining Lung neoplasm |
CN109102502A (en) * | 2018-08-03 | 2018-12-28 | 西北工业大学 | Pulmonary nodule detection method based on Three dimensional convolution neural network |
CN109543667A (en) * | 2018-11-14 | 2019-03-29 | 北京工业大学 | A kind of text recognition method based on attention mechanism |
-
2019
- 2019-04-08 CN CN201910276013.4A patent/CN110175979B/en active Active
Patent Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2018094294A1 (en) * | 2016-11-18 | 2018-05-24 | Salesforce.Com, Inc. | Spatial attention model for image captioning |
CN106650830A (en) * | 2017-01-06 | 2017-05-10 | 西北工业大学 | Deep model and shallow model decision fusion-based pulmonary nodule CT image automatic classification method |
CN108388900A (en) * | 2018-02-05 | 2018-08-10 | 华南理工大学 | The video presentation method being combined based on multiple features fusion and space-time attention mechanism |
CN108537793A (en) * | 2018-04-17 | 2018-09-14 | 电子科技大学 | A kind of pulmonary nodule detection method based on improved u-net networks |
CN108629764A (en) * | 2018-04-17 | 2018-10-09 | 杭州依图医疗技术有限公司 | A kind of good pernicious method and device of determining Lung neoplasm |
CN109102502A (en) * | 2018-08-03 | 2018-12-28 | 西北工业大学 | Pulmonary nodule detection method based on Three dimensional convolution neural network |
CN109543667A (en) * | 2018-11-14 | 2019-03-29 | 北京工业大学 | A kind of text recognition method based on attention mechanism |
Non-Patent Citations (4)
Title |
---|
KUN ZHANG ET AL.: "Context-Aware Dual-Attention Network for Natural Language Inference", 《PAKDD 2019:ADVANCES IN KNOWLEDGE DISCOVERY AND DATA MINING》 * |
WENTAO ZHU ET AL.: "DeepLung: 3D Deep Convolutional Nets for Automated Pulmonary Nodule Detection and Classification", 《ARXIV》 * |
冯雨 等: "基于三维卷积神经网络的肺结节识别研究", 《HTTP://KNS.CNKI.NET/KCMS/DETAIL/31.1252.O4.20190319.0909.026.HTML》 * |
邓忠豪 等: "基于深度卷积神经网络的肺结节检测算法", 《计算机应用》 * |
Cited By (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110570417A (en) * | 2019-09-12 | 2019-12-13 | 慧影医疗科技(北京)有限公司 | Pulmonary nodule classification method and device and image processing equipment |
CN110766051A (en) * | 2019-09-20 | 2020-02-07 | 四川大学华西医院 | Lung nodule morphological classification method based on neural network |
CN110807764A (en) * | 2019-09-20 | 2020-02-18 | 成都智能迭迦科技合伙企业(有限合伙) | Lung cancer screening method based on neural network |
CN111626969A (en) * | 2020-05-22 | 2020-09-04 | 东北易华录信息技术有限公司 | Attention mechanism-based corn disease image processing method |
CN111626969B (en) * | 2020-05-22 | 2023-05-30 | 张卫东 | Corn disease image processing method based on attention mechanism |
CN112419307A (en) * | 2020-12-11 | 2021-02-26 | 长春工业大学 | Lung nodule benign and malignant identification method based on attention mechanism |
CN113782181A (en) * | 2021-07-26 | 2021-12-10 | 杭州深睿博联科技有限公司 | CT image-based lung nodule benign and malignant diagnosis method and device |
Also Published As
Publication number | Publication date |
---|---|
CN110175979B (en) | 2021-07-27 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110175979A (en) | A kind of Lung neoplasm classification method based on collaboration attention mechanism | |
CN108615237A (en) | A kind of method for processing lung images and image processing equipment | |
CN108765387A (en) | Based on Faster RCNN mammary gland DBT image lump automatic testing methods | |
CN109685776A (en) | A kind of pulmonary nodule detection method based on ct images and system | |
CN108428229A (en) | It is a kind of that apparent and geometric properties lung's Texture Recognitions are extracted based on deep neural network | |
CN108830826A (en) | A kind of system and method detecting Lung neoplasm | |
CN109034221A (en) | A kind of processing method and its device of cervical cytology characteristics of image | |
CN106485695A (en) | Medical image Graph Cut dividing method based on statistical shape model | |
CN110503626A (en) | Based on space-semantic significance constraint CT image modalities alignment schemes | |
Cheng et al. | Modified U-Net block network for lung nodule detection | |
CN114494199A (en) | Liver CT tumor segmentation and classification method based on deep learning | |
Yang et al. | A multi-stage progressive learning strategy for COVID-19 diagnosis using chest computed tomography with imbalanced data | |
CN109671055A (en) | Pulmonary nodule detection method and device | |
CN115409815A (en) | Pulmonary nodule detection method based on three-dimensional deformable transformer | |
CN111784638A (en) | Pulmonary nodule false positive screening method and system based on convolutional neural network | |
Min et al. | Computer-aided detection of pulmonary nodules based on convolutional neural networks: a review | |
Amin et al. | Optimizing convolutional neural networks with transfer learning for making classification report in covid-19 chest x-rays scans | |
Bu et al. | Lung nodule detection based on YOLOv3 deep learning with limited datasets | |
Cao et al. | MSM-ViT: A multi-scale MobileViT for pulmonary nodule classification using CT images | |
CN111724345A (en) | Pneumonia picture verification device and method capable of adaptively adjusting size of receptive field | |
Ma et al. | Automatic pulmonary ground‐glass opacity nodules detection and classification based on 3D neural network | |
Podder et al. | Rethinking the transfer learning architecture for respiratory diseases and COVID-19 diagnosis | |
Zhang et al. | Topological structure and global features enhanced graph reasoning model for non-small cell lung cancer segmentation from CT | |
Yang et al. | 3D multi‐view squeeze‐and‐excitation convolutional neural network for lung nodule classification | |
TWI733627B (en) | Lung nodule detection method on low-dose chest computer tomography images using deep learning and its computer program product |
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 |