CN108876779A - Lung cancer method for early prediction, electronic equipment based on deep learning - Google Patents
Lung cancer method for early prediction, electronic equipment based on deep learning Download PDFInfo
- Publication number
- CN108876779A CN108876779A CN201810648062.1A CN201810648062A CN108876779A CN 108876779 A CN108876779 A CN 108876779A CN 201810648062 A CN201810648062 A CN 201810648062A CN 108876779 A CN108876779 A CN 108876779A
- Authority
- CN
- China
- Prior art keywords
- lung
- image
- design
- data
- shielding film
- 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
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
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/60—Analysis of geometric attributes
- G06T7/62—Analysis of geometric attributes of area, perimeter, diameter or volume
-
- 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 the lung cancer method for early prediction based on deep learning, include the following steps：Lung CT image data is obtained, the lung CT image data is made of several Lung sections；Pretreatment operation is carried out to the lung CT image data；Design Lung neoplasm detection model；Design Lung neoplasm disaggregated model；According to the sample number of every batch of training Lung neoplasm detection model.The present invention is converted into model by preprocessor etc. to original CT data can training data, and the model after training is deployed in server, patient lungs' tubercle after uploading CT data is detected using the model of the server, export nodule position and size, it predicts the good pernicious of tubercle, and the nodularity checked next time is predicted.
Description
Technical field
The present invention relates to CT images technologies, more particularly to the lung cancer method for early prediction based on deep learning and electronics are set
It is standby.
Background technique
Lung cancer is one of highest malignant tumour of disease incidence inside malignant tumour, and five year survival rate is only 15% left side
It is right.Therefore it finds and treats as soon as possible, cure rate could be improved.Lung cancer initial symptoms are very unobvious, it is easy to it is ignored,
And reach an advanced stage, cancer metastasis can occur, cause to treat extremely difficult.Tumour association of the U.S. is a series of studies have shown that inspection
Survey the very effective means that pulmonary nodule is early detection lung cancer.It is small-sized due to pulmonary nodule tumour, pass through
Low-dose CT carries out early screening, can greatly improve the diagnosis of the early stage of lung cancer.
CT tomographic imaging is the very high three-dimensional imaging of resolution ratio, and data volume is huge.Each patient has several hundred tension faults
Image.In lung cancer early stage, since tubercle very little, doctor are possible to generate solution because experience is insufficient or tired and misread at this time
Accidentally.
Someone just put forward to use computer-aided diagnosis in the 1960s, was helped using the macrooperation amount of computer
Doctor carry out diagnosis in traditional cad technique, mainly by have the people of many years of experience for different type disease detection come
Design some characteristic values, such as using texture analysis, the various characteristic functions of edge detection and object detection, such as SIFT or
HoG etc..The extraction of these features is realized by manually completely.This way objectively requires to read a large amount of cases, Cai Nengcong
In sum up experience to instruct feature extraction.On the one hand, characteristic mass and limited amount directly result in the diagnosis of disease
It is low；On the other hand, entirely different feature is designed for various disease, more increases degree of difficulty, this is also traditional CAD
The reason of technology can not be soon applied in the different field of medicine.
Summary of the invention
For overcome the deficiencies in the prior art, one of the objects of the present invention is to provide the lung cancer early stages based on deep learning
Prediction technique, the problem of can solve pulmonary cancer diagnosis low efficiency in the prior art.
The second object of the present invention is to provide a kind of electronic equipment, can solve pulmonary cancer diagnosis low efficiency in the prior art
The problem of.
An object of the present invention is implemented with the following technical solutions：
Lung cancer method for early prediction based on deep learning, includes the following steps：
Obtaining step：Lung CT image data is obtained, the lung CT image data is made of several Lung sections；
Processing step：Pretreatment operation is carried out to the lung CT image data；
Detection design step：Design Lung neoplasm detection model；
Classification design step：Design Lung neoplasm disaggregated model；
Model training step：According to the sample number of every batch of training Lung neoplasm detection model.
Preferably, the pretreatment operation specifically includes following sub-step：
Read lung CT image data；
Calculate every slice in arbitrary point arrive central point geometric distance, and remove geometric distance be greater than image_
Geometric distance of the point of size/2 to central point；Side is carried out using every sectioning image of the Gaussian filter to CT images data
Edge sharpens；
Using HU=600 as threshold value by every sectioning image binaryzation of CT images data；
To every slice by connection domain analysis, Retention area is greater than somin^2, and connection of the eccentricity less than 0.99
Domain；3D image analysing computer is done to CT images data, to remove impurity and background, and fills up duck eye in 3D image；
It repeats the above steps to generate maximum two UNICOM domains, which is left lung and right lung, and generates a left side
The shielding film of lung and right lung；
The area for calculating the convex closure on shielding film judges the area of convex closure half whether bigger than the area of shielding film,
If so, continuing to use shielding film, otherwise, convex closure is substituted into the shielding film；
Expansion process is done to shielding film, by linear transformation, the HU value [- 1200,600] that every is sliced is converted into [0,
255] gray value, and be multiplied the image value that every is sliced to obtain the segmentation data value of lung with gray value respectively；
CT images data are subjected to resampling, make the image as unit pixel representation space size of the corresponding slice of different patients
Unanimously.
Preferably, detection design step specifically includes following sub-step：
The 3D convolutional neural networks based on U-NET and RESNET are designed, nodule detection is done；
Allowable loss function；
Design SGD optimizer.
Classification design step specifically includes following sub-step：
Three layers of 3D convolutional neural networks are designed to do the good pernicious prediction of tubercle；
Allowable loss function；
Design SGD optimizer.
It preferably, further include following steps：
Deploying step：Lung neoplasm detection model after training is deployed to server end, and constructs REST API server.
The second object of the present invention is implemented with the following technical solutions：
A kind of electronic equipment, including memory and processor are stored with the meter that can be executed by processor on the memory
Calculation machine program, the computer program realize following steps when being executed by processor：
Obtaining step：Lung CT image data is obtained, the lung CT image data is made of several Lung sections；
Processing step：Pretreatment operation is carried out to the lung CT image data；
Detection design step：Design Lung neoplasm detection model；
Classification design step：Design Lung neoplasm disaggregated model；
Model training step：According to the sample number of every batch of training Lung neoplasm detection model.
Preferably, the pretreatment operation specifically includes following sub-step：
Read lung CT image data；
Calculate every slice in arbitrary point arrive central point geometric distance, and remove geometric distance be greater than image_
Geometric distance of the point of size/2 to central point；Side is carried out using every sectioning image of the Gaussian filter to CT images data
Edge sharpens；
Using HU=600 as threshold value by every sectioning image binaryzation of CT images data；
To every slice by connection domain analysis, Retention area is greater than somin^2, and connection of the eccentricity less than 0.99
Domain；3D image analysing computer is done to CT images data, to remove impurity and background, and fills up duck eye in 3D image；
It repeats the above steps to generate maximum two UNICOM domains, which is left lung and right lung, and generates a left side
The shielding film of lung and right lung；
The area for calculating the convex closure on shielding film judges the area of convex closure half whether bigger than the area of shielding film,
If so, continuing to use shielding film, otherwise, convex closure is substituted into the shielding film；
Expansion process is done to shielding film, by linear transformation, the HU value [- 1200,600] that every is sliced is converted into [0,
255] gray value, and be multiplied the image value that every is sliced to obtain the segmentation data value of lung with gray value respectively；
CT images data are subjected to resampling, make the image as unit pixel representation space size of the corresponding slice of different patients
Unanimously.
Preferably, detection design step specifically includes following sub-step：
The 3D convolutional neural networks based on U-NET and RESNET are designed, nodule detection is done；
Allowable loss function；
Design SGD optimizer.
Preferably, classification design step specifically includes following sub-step：
Three layers of 3D convolutional neural networks are designed to do the good pernicious prediction of tubercle；
Allowable loss function；
Design SGD optimizer.
Compared with prior art, the beneficial effects of the present invention are：
The present invention to original CT data by preprocessor etc. be converted into model can training data, and by the model after training
It is deployed in server, patient lungs' tubercle after uploading CT data is detected using the model of the server, exports tubercle
Position and size are predicted the good pernicious of tubercle, and are predicted the nodularity checked next time.
Detailed description of the invention
Fig. 1 is the flow chart of the lung cancer method for early prediction of the invention based on deep learning.
Specific embodiment
In the following, being described further in conjunction with attached drawing and specific embodiment to the present invention：
As shown in Figure 1, the present invention provides a kind of lung cancer method for early prediction based on deep learning, following step is specifically included
Suddenly：
S1：Lung CT image data is obtained, the lung CT image data is made of several Lung sections；
The lung CT image data that the present invention uses has LUNAdata, Data Science Bowl2017stage1data,
LUNA data are MHD format, read data using SimpleITK tool, 2017 data of Data Science Bowl are DICOM
Format reads data using dicom tool.
S2：Pretreatment operation is carried out to the lung CT image data；
In this step, lung CT image data is read；
Calculate every slice in arbitrary point arrive central point geometric distance, and remove geometric distance be greater than image_
Geometric distance of the point of size/2 to central point；Side is carried out using every sectioning image of the Gaussian filter to CT images data
Edge sharpens；
Using HU=600 as threshold value by every sectioning image binaryzation of CT images data；It is mono- that CT value unit is also known as Heng Shi
Position, is named by the name of its inventor SirGreoffreyHounsfie1d, abbreviation Hu, for indicating group on CT image
Knit the relative density of structure, that is, body some part for X-ray opaqueness.
To every slice by connection domain analysis, Retention area is greater than somin^2, and connection of the eccentricity less than 0.99
Domain；3D image analysing computer is done to CT images data, to remove impurity and background, and fills up duck eye in 3D image；UNICOM domain is general
Refer in image with same pixel value and position it is adjacent foreground pixel point composition image-region (Region, Blob).Connection
Logical regional analysis (Connected Component Analysis, Connected Component Labeling) refers to and will scheme
Each connected region as in is found out and is marked.Connection regional analysis is a kind of in numerous application fields of image analysis processing
More common and basic method.
It repeats the above steps to generate maximum two UNICOM domains, which is left lung and right lung, and generates a left side
The shielding film of lung and right lung；
The area for calculating the convex closure on shielding film judges the area of convex closure half whether bigger than the area of shielding film,
If so, continuing to use shielding film, otherwise, convex closure is substituted into the shielding film；For two-dimensional image, convex closure is exactly will be outermost
The point of layer connects composition convex polygon, it can concentrate all points comprising point.
Expansion process is done to shielding film, by linear transformation, the HU value [- 1200,600] that every is sliced is converted into [0,
255] gray value, and be multiplied the image value that every is sliced to obtain the segmentation data value of lung with gray value respectively；Expansion is
Most basic operation in morphology goes scanning original image with a core, replaces anchor point position with the max pixel value of core institute overlay area
The pixel set.
CT images data are subjected to resampling, make the image as unit pixel representation space size of the corresponding slice of different patients
Unanimously.
When being the data of DICOM format, treated image data is saved, reads annatations data, it will
Annatations resampling saves treated label data, at that time when the data of MHD format, reads annatations number
The spatial information in annatations is mapped on the voxel coordinate system of image by world coordinates after, then will
Annatations resampling simultaneously saves treated label data.
S3：Design Lung neoplasm detection model；
Specially：The 3D convolutional neural networks based on U-NET and RESNET are designed, nodule detection is done；Lung neoplasm detects net
Network integrally uses 3DFasterRCNN because task only there are two classify (tubercle and background), prediction target directly as
Testing result, without carrying out the classification of second stage.Basic network part uses Unet+Resnet18 framework, and the inside is each
Resnet Block is made of multiple convolutional layers and bn layers and relu layers.Since negative sample is far more than positive sample, distribution is not
It is similar to tubercle to balance some negative samples of, it is easy be accidentally divided into tubercle .hard negative mining and be used to handle this ask
Topic.Originally it was negative example, positive example is but divided into very big probability, this part increases weight when calculating loss.Joint intersection
(IoU) for determining that the sample of output is positive sample or negative sample.The IoU of target nodule is greater than 0.5 and pair less than 0.02
As by respectively as positive sample and negative sample.
Allowable loss function, comprising Bingary Cross Entropy (binary intersect entropy loss, for classifying) and
SmothL1Loss (IoU of target nodule is greater than 0.5 and the object less than 0.02 is by respectively as positive sample and negative sample)；
SGD optimizer is designed, parameter momentum=0.9, weight_decay=1e-4, wherein momentum are set
For momentum, weight_decay is weight.
S4：Design Lung neoplasm disaggregated model；
Three layers of 3D convolutional neural networks are designed to do the good pernicious prediction of tubercle；Lung neoplasm sorter network uses 3DCNN network,
Classification output Malignant Nodules probability.Network is made of two layers of full linking layer.Input data be upper note nodule detection network last
A convolutional layer exports feature, this is the feature of 32 × 32 × 32 cube × 128, therefrom chooses confidence level highest 5
Tubercle feature.
Allowable loss function；Loss function by the imaginary tubercle of the confidence level of this five tubercles and one confidence level joint
Probability is constituted, P=1- (1-Pd)∏i(1-Pi), wherein PdIt is the lung cancer probability for representing imaginary tubercle, PiRepresent 1--5 tubercle
Lung cancer probability.
SGD optimizer is designed, parameter momentum=0.9, weight_decay=1e-4 are set.
S5：According to the sample number of every batch of training Lung neoplasm detection model, including learning rate, number of iterations, while training lung
Nodule detection model and lung cancer prediction model.
Since training sample is 3D data, memory and video memory are expended, so batch quantity cannot be arranged excessive, first trains tubercle
Network is detected, after so that network is exported effective tubercle, then joint training nodule detection and sorter network.
S6：Lung neoplasm detection model after training is deployed to server end, and constructs REST API server.
REST API server environment, including Python, PyTorch, Redis, Flask, modification Apache service are installed
Device configuration file, creates web server code file, and client code file creates model server code file, creation
WSGI configuration file establishes the soft link of project, copies trained model to server, starts Apache Web server,
Start Redis database, Boot Model service.Server carries model, after client accesses server, using on server
Model the CT data of patient are predicted.
Because the CT data of patient belong to sensitive information, model server is deployed in hospital internal network, and outer
It needs to install Nvidia GPU on net isolation server hardware, needs that CUDA, CUDNN are installed on software.Server is installed to need
Software package be：Python,PyTorch,Redis,Flask,Apache,WSGI,virtualenv.Edit the configuration of Apache
File, additional project information, to be directed toward Flask application program.The code file for writing Web server receives client and is transmitted through
The CT data come, and are pushed into Redis queue, continuous cycle request, until obtaining prediction data from model server, and by data
Send back client.HTML the and JS file that client is shown is write, is used to and client carries out data interaction, receive CT number
According to display prediction result.Model server code file is write, stress model is completed, the image in Redis queue is carried out pre-
It surveys, is as a result also pushed into Redis queue.WSGI file is write, server directory is added to system path, and import web and answer
With.The soft default directory for being linked to Apache of creation project.Because using GPU and CUDA, thus the library CUDA be flexible coupling to
Under system directory/usr/lib, so that Apache can find CUDA library file.Model is copied to server, so as to model service
Device load.Start Apache Web server, Redis database, model server with order.
The present invention also provides a kind of electronic equipment, including memory and processor, being stored on the memory can be located
The computer program that device executes is managed, the computer program realizes following steps when being executed by processor：
Obtaining step：Lung CT image data is obtained, the lung CT image data is made of several Lung sections；
Processing step：Pretreatment operation is carried out to the lung CT image data；
Detection design step：Design Lung neoplasm detection model；
Classification design step：Design Lung neoplasm disaggregated model；
Model training step：According to the sample number of every batch of training Lung neoplasm detection model.
The pretreatment operation specifically includes following sub-step：
Read lung CT image data；
Calculate every slice in arbitrary point arrive central point geometric distance, and remove geometric distance be greater than image_
Geometric distance of the point of size/2 to central point；Side is carried out using every sectioning image of the Gaussian filter to CT images data
Edge sharpens；
Using HU=600 as threshold value by every sectioning image binaryzation of CT images data；
To every slice by connection domain analysis, Retention area is greater than somin^2, and connection of the eccentricity less than 0.99
Domain；3D image analysing computer is done to CT images data, to remove impurity and background, and fills up duck eye in 3D image；
It repeats the above steps to generate maximum two UNICOM domains, which is left lung and right lung, and generates a left side
The shielding film of lung and right lung；
The area for calculating the convex closure on shielding film judges the area of convex closure half whether bigger than the area of shielding film,
If so, continuing to use shielding film, otherwise, convex closure is substituted into the shielding film；
Expansion process is done to shielding film, by linear transformation, the HU value [- 1200,600] that every is sliced is converted into [0,
255] gray value, and be multiplied the image value that every is sliced to obtain the segmentation data value of lung with gray value respectively；
CT images data are subjected to resampling, make the image as unit pixel representation space size of the corresponding slice of different patients
Unanimously.
Detection design step specifically includes following sub-step：
The 3D convolutional neural networks based on U-NET and RESNET are designed, nodule detection is done；
Allowable loss function；
Design SGD optimizer.
Classification design step specifically includes following sub-step：
Three layers of 3D convolutional neural networks are designed to do the good pernicious prediction of tubercle；
Allowable loss function；
Design SGD optimizer.
The present invention is mainly pre-processed by CT images, model training, and it is logical that model disposes three part composition original CT data collection
Cross pretreatment, be converted into model can training data, be supplied to model be trained training after the completion of, model is saved and is deployed in
Server end, using the model of server end, examines patient lungs' tubercle after having passed patient CT data on the client
It surveys, exports nodule position and size, predict the good pernicious of tubercle, and predict the nodularity checked next time.
It will be apparent to those skilled in the art that can make various other according to the above description of the technical scheme and ideas
Corresponding change and deformation, and all these changes and deformation all should belong to the protection scope of the claims in the present invention
Within.
Claims (9)
1. the lung cancer method for early prediction based on deep learning, which is characterized in that include the following steps：
Obtaining step：Lung CT image data is obtained, the lung CT image data is made of several Lung sections；
Processing step：Pretreatment operation is carried out to the lung CT image data；
Detection design step：Design Lung neoplasm detection model；
Classification design step：Design Lung neoplasm disaggregated model；
Model training step：According to the sample number of every batch of training Lung neoplasm detection model.
2. lung cancer method for early prediction as described in claim 1, which is characterized in that the pretreatment operation specifically includes as follows
Sub-step：
Read lung CT image data；
It calculates arbitrary point in every slice and arrives the geometric distance of central point, and remove geometric distance and be greater than image_size/2
Geometric distance of the point to central point；Edge sharpening is carried out using every sectioning image of the Gaussian filter to CT images data；
Using HU=600 as threshold value by every sectioning image binaryzation of CT images data；
To every slice by connection domain analysis, Retention area is greater than somin^2, and UNICOM domain of the eccentricity less than 0.99；It is right
CT images data do 3D image analysing computer, to remove impurity and background, and fill up duck eye in 3D image；
Repeat the above steps to generate maximum two UNICOM domains, two UNICOM domains be left lung and right lung, and generate left lung and
The shielding film of right lung；
The area for calculating the convex closure on shielding film judges the area of convex closure half whether bigger than the area of shielding film, if so,
Shielding film is then continued to use, otherwise, convex closure is substituted into the shielding film；
Expansion process is done to shielding film, by linear transformation, the HU value [- 1200,600] that every is sliced is converted into [0,255]
Gray value, and respectively by every be sliced image value be multiplied to obtain the segmentation data value of lung with gray value；
CT images data are subjected to resampling, make the image as unit pixel representation space size one of the corresponding slice of different patients
It causes.
3. lung cancer method for early prediction as claimed in claim 2, which is characterized in that detection design step specifically includes following son
Step：
The 3D convolutional neural networks based on U-NET and RESNET are designed, nodule detection is done；
Allowable loss function；
Design SGD optimizer.
4. lung cancer method for early prediction as described in claim 1, which is characterized in that classification design step specifically includes following son
Step：
Three layers of 3D convolutional neural networks are designed to do the good pernicious prediction of tubercle；
Allowable loss function；
Design SGD optimizer.
5. lung cancer method for early prediction as described in claim 1, which is characterized in that further include following steps：
Deploying step：Lung neoplasm detection model after training is deployed to server end, and constructs REST API server.
6. a kind of electronic equipment, including memory and processor, the calculating that can be executed by processor is stored on the memory
Machine program, which is characterized in that the computer program realizes following steps when being executed by processor：
Obtaining step：Lung CT image data is obtained, the lung CT image data is made of several Lung sections；
Processing step：Pretreatment operation is carried out to the lung CT image data；
Detection design step：Design Lung neoplasm detection model；
Classification design step：Design Lung neoplasm disaggregated model；
Model training step：According to the sample number of every batch of training Lung neoplasm detection model.
7. electronic equipment as claimed in claim 6, which is characterized in that the pretreatment operation specifically includes following sub-step：
Read lung CT image data；
It calculates arbitrary point in every slice and arrives the geometric distance of central point, and remove geometric distance and be greater than image_size/2
Geometric distance of the point to central point；Edge sharpening is carried out using every sectioning image of the Gaussian filter to CT images data；
Using HU=600 as threshold value by every sectioning image binaryzation of CT images data；
To every slice by connection domain analysis, Retention area is greater than somin^2, and UNICOM domain of the eccentricity less than 0.99；It is right
CT images data do 3D image analysing computer, to remove impurity and background, and fill up duck eye in 3D image；
Repeat the above steps to generate maximum two UNICOM domains, two UNICOM domains be left lung and right lung, and generate left lung and
The shielding film of right lung；
The area for calculating the convex closure on shielding film judges the area of convex closure half whether bigger than the area of shielding film, if so,
Shielding film is then continued to use, otherwise, convex closure is substituted into the shielding film；
Expansion process is done to shielding film, by linear transformation, the HU value [- 1200,600] that every is sliced is converted into [0,255]
Gray value, and respectively by every be sliced image value be multiplied to obtain the segmentation data value of lung with gray value；
CT images data are subjected to resampling, make the image as unit pixel representation space size one of the corresponding slice of different patients
It causes.
8. electronic equipment as claimed in claim 6, which is characterized in that detection design step specifically includes following sub-step：
The 3D convolutional neural networks based on U-NET and RESNET are designed, nodule detection is done；
Allowable loss function；
Design SGD optimizer.
9. electronic equipment as claimed in claim 6, which is characterized in that classification design step specifically includes following sub-step：
Three layers of 3D convolutional neural networks are designed to do the good pernicious prediction of tubercle；
Allowable loss function；
Design SGD optimizer.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810648062.1A CN108876779A (en) | 2018-06-22 | 2018-06-22 | Lung cancer method for early prediction, electronic equipment based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810648062.1A CN108876779A (en) | 2018-06-22 | 2018-06-22 | Lung cancer method for early prediction, electronic equipment based on deep learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN108876779A true CN108876779A (en) | 2018-11-23 |
Family
ID=64340415
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810648062.1A Pending CN108876779A (en) | 2018-06-22 | 2018-06-22 | Lung cancer method for early prediction, electronic equipment based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108876779A (en) |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109685776A (en) * | 2018-12-12 | 2019-04-26 | 华中科技大学 | A kind of pulmonary nodule detection method based on ct images and system |
| CN110322435A (en) * | 2019-01-20 | 2019-10-11 | 北京工业大学 | A kind of gastric cancer pathological image cancerous region dividing method based on deep learning |
| CN110415791A (en) * | 2019-01-29 | 2019-11-05 | 四川大学华西医院 | System and method is established in a kind of disease library |
| CN110443268A (en) * | 2019-05-30 | 2019-11-12 | 杭州电子科技大学 | A kind of benign pernicious classification method of liver cancer CT image based on deep learning |
| CN111369524A (en) * | 2020-03-02 | 2020-07-03 | 北京推想科技有限公司 | Nodule recognition model training method, nodule recognition method and device |
| CN112259199A (en) * | 2020-10-29 | 2021-01-22 | 西交利物浦大学 | Medical image classification model training method, system, storage medium and medical image processing device |
| WO2021015913A1 (en) * | 2019-07-22 | 2021-01-28 | The Johns Hopkins University | Lung cancer prediction |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106934228A (en) * | 2017-03-06 | 2017-07-07 | 杭州健培科技有限公司 | Lung's pneumothorax CT image classification diagnostic methods based on machine learning |
| CN107016665A (en) * | 2017-02-16 | 2017-08-04 | 浙江大学 | A kind of CT pulmonary nodule detection methods based on depth convolutional neural networks |
| CN107563383A (en) * | 2017-08-24 | 2018-01-09 | 杭州健培科技有限公司 | A kind of medical image auxiliary diagnosis and semi-supervised sample generation system |
| CN107767362A (en) * | 2017-09-01 | 2018-03-06 | 苏州侠洛信息科技有限公司 | A kind of early screening of lung cancer device based on deep learning |
-
2018
- 2018-06-22 CN CN201810648062.1A patent/CN108876779A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107016665A (en) * | 2017-02-16 | 2017-08-04 | 浙江大学 | A kind of CT pulmonary nodule detection methods based on depth convolutional neural networks |
| CN106934228A (en) * | 2017-03-06 | 2017-07-07 | 杭州健培科技有限公司 | Lung's pneumothorax CT image classification diagnostic methods based on machine learning |
| CN107563383A (en) * | 2017-08-24 | 2018-01-09 | 杭州健培科技有限公司 | A kind of medical image auxiliary diagnosis and semi-supervised sample generation system |
| CN107767362A (en) * | 2017-09-01 | 2018-03-06 | 苏州侠洛信息科技有限公司 | A kind of early screening of lung cancer device based on deep learning |
Non-Patent Citations (1)
| Title |
|---|
| YULEI QIN等: "SIMULTANEOUS ACCURATE DETECTION OF PULMONARY NODULES AND FALSE POSITIVE REDUCTION USING 3D CNNS", 《2018 IEEE INTERNATIONAL CONFERENCE ON ACOUSTICS，SPEECH AND SIGNAL PROCESSING（ICASSP）》 * |
Cited By (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109685776A (en) * | 2018-12-12 | 2019-04-26 | 华中科技大学 | A kind of pulmonary nodule detection method based on ct images and system |
| CN109685776B (en) * | 2018-12-12 | 2021-01-19 | 华中科技大学 | Pulmonary nodule detection method and system based on CT image |
| CN110322435A (en) * | 2019-01-20 | 2019-10-11 | 北京工业大学 | A kind of gastric cancer pathological image cancerous region dividing method based on deep learning |
| CN110415791A (en) * | 2019-01-29 | 2019-11-05 | 四川大学华西医院 | System and method is established in a kind of disease library |
| CN110443268A (en) * | 2019-05-30 | 2019-11-12 | 杭州电子科技大学 | A kind of benign pernicious classification method of liver cancer CT image based on deep learning |
| CN110443268B (en) * | 2019-05-30 | 2022-02-08 | 杭州电子科技大学 | Liver CT image benign and malignant classification method based on deep learning |
| WO2021015913A1 (en) * | 2019-07-22 | 2021-01-28 | The Johns Hopkins University | Lung cancer prediction |
| CN111369524A (en) * | 2020-03-02 | 2020-07-03 | 北京推想科技有限公司 | Nodule recognition model training method, nodule recognition method and device |
| CN111369524B (en) * | 2020-03-02 | 2021-07-27 | 推想医疗科技股份有限公司 | Nodule recognition model training method, nodule recognition method and device |
| CN112259199A (en) * | 2020-10-29 | 2021-01-22 | 西交利物浦大学 | Medical image classification model training method, system, storage medium and medical image processing device |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108876779A (en) | Lung cancer method for early prediction, electronic equipment based on deep learning | |
| CN108898588A (en) | Therapeutic effect appraisal procedure based on time series, electronic equipment | |
| US10347010B2 (en) | Anomaly detection in volumetric images using sequential convolutional and recurrent neural networks | |
| Yan et al. | 3D context enhanced region-based convolutional neural network for end-to-end lesion detection | |
| Zhao et al. | Automatic detection of small lung nodules on CT utilizing a local density maximum algorithm | |
| CN108615237A (en) | A kind of method for processing lung images and image processing equipment | |
| Golosio et al. | A novel multithreshold method for nodule detection in lung CT | |
| CN108986067A (en) | Pulmonary nodule detection method based on cross-module state | |
| Karthik et al. | Contour-enhanced attention CNN for CT-based COVID-19 segmentation | |
| Pezeshk et al. | Seamless lesion insertion for data augmentation in CAD training | |
| CN108549912A (en) | A kind of medical image pulmonary nodule detection method based on machine learning | |
| US10706534B2 (en) | Method and apparatus for classifying a data point in imaging data | |
| McEvoy et al. | Using machine learning to classify image features from canine pelvic radiographs: evaluation of partial least squares discriminant analysis and artificial neural network models | |
| Li et al. | Explainable multi-instance and multi-task learning for COVID-19 diagnosis and lesion segmentation in CT images | |
| Gao et al. | Joint disc and cup segmentation based on recurrent fully convolutional network | |
| Guo et al. | High performance lung nodule detection schemes in CT using local and global information | |
| CN111724344A (en) | Method for generating medical ultrasonic image data based on countermeasure network | |
| US20220285011A1 (en) | Document creation support apparatus, document creation support method, and program | |
| Goel et al. | Computecovid19+: accelerating Covid-19 diagnosis and monitoring via high-performance deep learning on CT images | |
| Huynh et al. | Region-of-interest optimization for deep-learning-based breast cancer detection in mammograms | |
| Li et al. | Scaling the inference of digital pathology deep learning models using cpu-based high-performance computing | |
| Yang et al. | 3D multi‐view squeeze‐and‐excitation convolutional neural network for lung nodule classification | |
| Yang et al. | Dual-path network for liver and tumor segmentation in CT images using Swin Transformer encoding approach | |
| CN115564756A (en) | Medical image focus positioning display method and system | |
| Gupta et al. | Deep learning assisted COVID-19 detection using full CT-scans |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20181123 |