CN110211139A - Automatic segmentation Radiotherapy of Esophageal Cancer target area and the method and system for jeopardizing organ - Google Patents
Automatic segmentation Radiotherapy of Esophageal Cancer target area and the method and system for jeopardizing organ Download PDFInfo
- Publication number
- CN110211139A CN110211139A CN201910504373.5A CN201910504373A CN110211139A CN 110211139 A CN110211139 A CN 110211139A CN 201910504373 A CN201910504373 A CN 201910504373A CN 110211139 A CN110211139 A CN 110211139A
- Authority
- CN
- China
- Prior art keywords
- network
- organ
- area
- interest
- region
- 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.)
- Withdrawn
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
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/11—Region-based segmentation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/25—Determination of region of interest [ROI] or a volume of interest [VOI]
-
- 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
- G06T2207/20016—Hierarchical, coarse-to-fine, multiscale or multiresolution image processing; Pyramid transform
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V2201/00—Indexing scheme relating to image or video recognition or understanding
- G06V2201/03—Recognition of patterns in medical or anatomical images
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Data Mining & Analysis (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Multimedia (AREA)
- Apparatus For Radiation Diagnosis (AREA)
- Image Processing (AREA)
Abstract
The invention discloses a kind of automatic method and system divided Radiotherapy of Esophageal Cancer target area and jeopardize organ, it is related to medical image segmentation field, method includes the following steps: extracting the feature of the CT image of input by residual error network, pass through feature pyramid network integration Analysis On Multi-scale Features figure；Suggest that network screens the area-of-interest of point each in characteristic pattern by region；In conjunction with area-of-interest aligned layer by the region area-of-interest pond that filters out of suggestion network a to fixed dimension；Pond to fixed-size area-of-interest is input to full articulamentum and carries out organ classes, and carries out organ site frame recurrence；Pond is input to organ segmentation's network to fixed-size area-of-interest simultaneously.The present invention has the advantages that improving automatic segmentation Radiotherapy of Esophageal Cancer target area and a variety of accuracy for jeopardizing organ.
Description
Technical field
The present invention relates to medical image segmentation fields, more particularly to one kind to divide Radiotherapy of Esophageal Cancer target area automatically and jeopardize device
The method and system of official.
Background technique
The cancer of the esophagus is the primary malignant tumor of oesophagus, in China, has up to 200,000 people to be affected every year.Radiotherapy
It is one of the primary treatments of the cancer of the esophagus, treatment plan is highly dependent on to planning target volume PTV and multiple jeopardizes organ
The accurate segmentation of OAR, the accuracy of segmentation determine the quality of radiotherapy treatment planning injectivity optimizing, to directly affect radiotherapy
The incidence of success or failure or complication.Since cancer of the esophagus target region shape is changeable, had differences between human body, between target area and organ
Boundary is smudgy, in clinical practice, is generally delineated manually by doctor.However, the accuracy manually delineated is highly dependent on
The clinical experience of radiologist, and inefficiency.Therefore, multiple organ divides the research side of always many scholars automatically
To a variety of different Automatic medical image segmentation algorithms also come into being.
The prior art one uses the Region Segmentation Algorithm based on provincial characteristics, and the basic thought of the algorithm is from one group of growth
Point starts, and growing point can be single pixel, or some zonule, it will adjacent pixel similar with the growing point property
Or region merges with growing point, forms new growing point, repeats this process until cannot grow；The algorithm is for gray scale
The little region segmentation of distributional difference is ineffective.
The prior art two uses the Medical image segmentation algorithm based on deformation model, which provides one on the image first
The closed curve of corresponding objects boundary Position Approximate, i.e. initial model；Then curve is in image information and curve self-information
Guide lower movement, the i.e. deformation process of model；Final curves move at correct object bounds, stop motion, i.e., model is received
It holds back；The algorithm is sensitive to model quality, and segmentation effect is difficult to meet clinical requirement.
Summary of the invention
Technical problem to be solved by the invention is to provide a kind of automatic segmentations of accuracy that can be improved and divide automatically
Radiotherapy of Esophageal Cancer target area and the method and system for jeopardizing organ.
The present invention is to solve above-mentioned technical problem by the following technical programs: automatic segmentation Radiotherapy of Esophageal Cancer target area and danger
And the method for organ, comprising the following steps:
Step A extracts the feature of the CT image of input by residual error network, multiple dimensioned by the feature pyramid network integration
Characteristic pattern suggests that network screens the area-of-interest of point each in characteristic pattern by region；
Step B, in conjunction with area-of-interest aligned layer by the region area-of-interest pond that filters out of suggestion network to one
Fixed dimension；
Pond to fixed-size area-of-interest is input to full articulamentum and carries out organ classes, and carries out device by step C
Official position frame returns；Pond is input to organ segmentation's network to fixed-size area-of-interest simultaneously.
The method that the present invention divides Radiotherapy of Esophageal Cancer target area automatically and jeopardizes organ more succinctly has compared to the prior art
Effect directly predicts input picture in pixel level, avoids the process of range searching, region merging technique；Before not needing
The range searching comprising target image of phase does not need the operation such as to merge in the later period yet；Convolution sum pondization extracts feature, contracting
Small Feature Mapping carries out contrary operation by warp lamination, feature is remapped on original image；It solves existing
The technology region segmentation little for intensity distribution difference is ineffective, and to model quality sensitive issue.
As the technical solution of optimization, specific step is as follows by step A:
Step a1, input one include multiple organ medicine CT image, by residual error network carry out depth convolution, Res2,
Res3, Res4, Res5 are ResNet layers of output respectively；
Step a2, using pyramid network obtain a semantic information, specific practice be to the feature above Res4 into
The operation of row dimensionality reduction, that is, add the convolutional layer of one layer of 1*1, operation is up-sampled to the feature above P5, so that they are having the same
Size；
Step a3, to treated P5 and treated that Res4 executes add operation, and corresponding element is added, by the knot of acquisition
Fruit is output to P4；To treated P4 and treated that Res3 executes add operation, and corresponding element is added, and the result of acquisition is defeated
P3 is arrived out；To treated P3 and treated that Res2 executes add operation, and corresponding element is added, the result of acquisition is output to
P2；
Step a4, using area suggest that network is predicted in P2, P3, P4, P5 difference output layer, obtain region of interest
Domain.
As the technical solution of optimization, residual error network+feature pyramid network is a preparatory trained convolution+pond
Neural network.
As the technical solution of optimization, region suggests that network is the neural network of a lightweight, it uses sliding window
Scan image finds region existing for organ, while obtaining each point on characteristic pattern and generating multiple candidate area-of-interests.
As the technical solution of optimization, in step B, area-of-interest aligned layer traverses each area-of-interest first, protects
It holds floating number boundary and does not do and quantify；Then area-of-interest is divided into k × k cell, the boundary of each cell also not into
Row quantization；Then 4 fixed coordinate positions are calculated in each cell, and the value of this 4 positions is calculated by bilinear interpolation；
Maximum pond operation is finally carried out, obtains fixed-size characteristic pattern on the basis of quantization error is not present.
As the technical solution of optimization, in step C, organ segmentation's network is one and organ classes and organ site frame
The parallel branch of recurrence.
As the technical solution of optimization, in step C, organ segmentation's network includes 4 continuous convolutional layers and 1 deconvolution
Layer, the core size of each convolutional layer are 3 × 3, and deconvolution is 2 times of up-samplings.
As the technical solution of optimization, in step C, it is classified as the heart, right lung, left lung, planning target volume (PTV), clinical target
The region volume (CTV), 6 class of background.
Automatic segmentation Radiotherapy of Esophageal Cancer target area and the system for jeopardizing organ, including area-of-interest screening module, Chi Huamo
Block, organ classes' module；
Area-of-interest screening module extracts the feature of the CT image of input by residual error network, passes through feature pyramid network
Network merges Analysis On Multi-scale Features figure, suggests that network screens the area-of-interest of point each in characteristic pattern by region；
Pond module combination area-of-interest aligned layer is by the region area-of-interest pond that filters out of suggestion network to one
A fixed dimension；
Pond to fixed-size area-of-interest is input to full articulamentum and carries out organ classes by organ classes' module, and
Carry out the recurrence of organ site frame；Pond is input to organ segmentation's network to fixed-size area-of-interest simultaneously.
As the technical solution of optimization, area-of-interest screening module include residual error network module, pyramid network module,
Network module is suggested in add operation module, region；
Residual error network module inputs the medicine CT image comprising multiple organ, carries out depth convolution by residual error network,
Res2, Res3, Res4, Res5 are ResNet layers of output respectively；
Pyramid network module obtains a semantic information using pyramid network, and specific practice is to above Res4
Feature carries out dimensionality reduction operation, that is, adds the convolutional layer of one layer of 1*1, operation is up-sampled to the feature above P5, so that they have
Identical size；
Add operation module is to treated P5 and treated that Res4 executes add operation, and corresponding element is added, will obtain
The result obtained is output to P4, P2, P3 result and so on；
Region suggests that network module using area suggests that network is predicted in P2, P3, P4, P5 difference output layer, obtains
Obtain area-of-interest.
The present invention has the advantages that the present invention utilizes Mask R-cnn algorithm, automatic segmentation Radiotherapy of Esophageal Cancer target is improved
Area and a variety of accuracy for jeopardizing organ；The algorithm can accurately and effectively on a medicine CT image simultaneously be partitioned into the heart,
Right lung, left lung, planning region target volume (PTV) and clinical target volume (CTV)；Missing inspection or false detection rate are low, have very big clinic
Application potential.
Detailed description of the invention
Fig. 1 is that the embodiment of the present invention divides Radiotherapy of Esophageal Cancer target area automatically and jeopardizes the flow chart of the method for organ.
Fig. 2 residual error of embodiment of the present invention network, feature pyramid network, region suggest network flow chart.
Fig. 3 is the flow chart of area-of-interest alignment operation of the embodiment of the present invention.
The structure chart of Fig. 4 organ segmentation's network of the embodiment of the present invention.
The test segmentation result figure of Fig. 5 embodiment of the present invention.
Specific embodiment
The present invention divides Radiotherapy of Esophageal Cancer target area automatically and jeopardizes the method for organ using Mask R-CNN algorithm, Mask
R-CNN algorithm is an example partitioning algorithm, can be used to do target detection, object instance segmentation, the detection of target critical point.
As shown in Figure 1, the present invention divides Radiotherapy of Esophageal Cancer target area automatically and jeopardizes the method for organ, comprising the following steps:
Step A extracts the features such as texture, the color of CT image of input by residual error network (ResNet), passes through feature
Pyramid network (FPN) merges Analysis On Multi-scale Features figure, suggests network (RPN) to the interested of point each in characteristic pattern by region
(ROI) is screened in region.
As shown in Fig. 2, specific step is as follows by step A:
Step a1, one medicine CT image comprising multiple organ of input, carries out depth volume by residual error network (ResNet)
Product, Res2, Res3, Res4, Res5 are ResNet layers of output respectively.
Step a2, using pyramid network (FPN), the thought of FPN is fusion multilayer feature, obtains a strong semantic letter
Breath improves detection accuracy.Its specific practice is that dimensionality reduction operation is carried out to the feature above Res4, that is, adds the convolution of one layer of 1*1
Layer up-samples operation to the feature above P5, so that they are of the same size.
Step a3, to treated P5 and treated that Res4 executes add operation, and corresponding element is added, by the knot of acquisition
Fruit is output to P4；To treated P4 and treated that Res3 executes add operation, and corresponding element is added, and the result of acquisition is defeated
P3 is arrived out；To treated P3 and treated that Res2 executes add operation, and corresponding element is added, the result of acquisition is output to
P2。
Step a4, using area suggest that network (RPN) is predicted in P2, P3, P4, P5 difference output layer, the sense of access
Interest region (ROI), R2, R3, R4, R5, that is, area-of-interest (ROI) in Fig. 2.
Residual error network+feature pyramid network is a preparatory trained convolution+pond neural network, defeated for extracting
The feature of the CT image entered generates corresponding characteristic pattern.Accuracy and speed can be greatly improved using residual error network.Low-level features
Low-level features are extracted in mapping, and advanced features are extracted in advanced features mapping.Feature pyramid network can merge Analysis On Multi-scale Features, make
Our result can divide multiple organic regions simultaneously.
Region suggests that network is the neural network of a lightweight, it uses sliding window scan image, finds organ and deposits
Region, while obtaining on characteristic pattern each point and generating multiple candidate area-of-interests, and to these candidate region of interest
Domain carries out classification and frame returns.
As shown in figure 3, step B, filters out region suggestion network in conjunction with area-of-interest aligned layer (ROI Align)
Area-of-interest pond is to a fixed dimension (7 × 7).
Since area-of-interest is usually to pass through what model returned, so it is usually a floating number, and input
Characteristic pattern to full articulamentum needs a fixed size, so needing area-of-interest pond a to fixed dimension.
In order to avoid quantization error, area-of-interest aligned layer traverses each area-of-interest first, keeps floating number side
Boundary, which is not done, to be quantified；Then area-of-interest is divided into k × k cell, the boundary of each cell is also without quantization；Then
4 fixed coordinate positions are calculated in each cell, and the value of this 4 positions is calculated by bilinear interpolation；It finally carries out most
Great Chiization operation obtains fixed-size characteristic pattern on the basis of quantization error is not present.
Pond to fixed-size area-of-interest is input to full articulamentum and carries out organ classes, is classified as by step C
The heart, right lung, left lung, planning target volume (PTV), the region clinical target volume (CTV), 6 class of background, and carry out organ site frame and return
Return；Pond is input to organ segmentation's network to fixed-size area-of-interest simultaneously, i.e., full convolutional neural networks are each
Operation under area-of-interest.
Organ segmentation's network is the parallel branch returned with organ classes and organ site frame.Organ classes, device
Official position frame returns, organ segmentation's network these three output branchs be it is parallel, it is not only simple but also efficiently.
As shown in figure 4, organ segmentation's network includes 4 continuous convolutional layers and 1 warp lamination, the core of each convolutional layer
Size is 3 × 3, and deconvolution is 2 times of up-samplings.Organ segmentation's network is used to predict the binary pixel values of each category mask.
One mask exports an area-of-interest.
As shown in figure 5, being tested with the model trained to test set, test segmentation result is as schemed.
The present invention divides Radiotherapy of Esophageal Cancer target area and the method for jeopardizing organ using Mask R-cnn algorithm automatically, improves
Automatic segmentation Radiotherapy of Esophageal Cancer target area and a variety of accuracy for jeopardizing organ；The algorithm can be accurately and effectively in a medicine CT
The heart, right lung, left lung, planning region target volume (PTV) and clinical target volume (CTV) are partitioned on image simultaneously；We are with 30
As training set, 10 collect the CT image of patient as verifying, and 4 are used to test, the results show that the case of missing inspection or erroneous detection is not
To 5%, there are very big clinical application potentiality.
Automatic segmentation Radiotherapy of Esophageal Cancer target area and the system for jeopardizing organ, including area-of-interest screening module, Chi Huamo
Block, organ classes' module.
Area-of-interest screening module extracts the feature of the CT image of input by residual error network, passes through feature pyramid network
Network merges Analysis On Multi-scale Features figure, suggests that network screens the area-of-interest of point each in characteristic pattern by region.
Area-of-interest screening module includes that residual error network module, pyramid network module, add operation module, region are built
Discuss network module.
Residual error network module inputs the medicine CT image comprising multiple organ, carries out depth convolution by residual error network,
Res2, Res3, Res4, Res5 are ResNet layers of output respectively.
Pyramid network module obtains a semantic information using pyramid network, and specific practice is to above Res4
Feature carries out dimensionality reduction operation, that is, adds the convolutional layer of one layer of 1*1, operation is up-sampled to the feature above P5, so that they have
Identical size.
Add operation module is to treated P5 and treated that Res4 executes add operation, and corresponding element is added, will obtain
The result obtained is output to P4, P2, P3 result and so on.
Region suggests that network module using area suggests that network is predicted in P2, P3, P4, P5 difference output layer, obtains
Obtain area-of-interest.
Pond module combination area-of-interest aligned layer is by the region area-of-interest pond that filters out of suggestion network to one
A fixed dimension.
Pond to fixed-size area-of-interest is input to full articulamentum and carries out organ classes by organ classes' module, and
Carry out the recurrence of organ site frame；Pond is input to organ segmentation's network to fixed-size area-of-interest simultaneously.
The foregoing is merely presently preferred embodiments of the present invention, is not intended to limit the invention, it is all in spirit of the invention and
Made any modifications, equivalent replacements, and improvements etc., should all be included in the protection scope of the present invention within principle.
Claims (10)
1. a kind of automatic segmentation Radiotherapy of Esophageal Cancer target area and the method for jeopardizing organ, which comprises the following steps:
Step A is extracted the feature of the CT image of input by residual error network, passes through feature pyramid network integration Analysis On Multi-scale Features
Figure suggests that network screens the area-of-interest of point each in characteristic pattern by region；
Step B, in conjunction with area-of-interest aligned layer by the region area-of-interest pond that filters out of suggestion network a to fixation
Size；
Pond to fixed-size area-of-interest is input to full articulamentum and carries out organ classes, and carries out organ position by step C
Set frame recurrence；Pond is input to organ segmentation's network to fixed-size area-of-interest simultaneously.
2. automatic segmentation Radiotherapy of Esophageal Cancer target area as described in claim 1 and the method for jeopardizing organ, it is characterised in that: step
Specific step is as follows by A:
Step a1, input one include multiple organ medicine CT image, by residual error network carry out depth convolution, Res2, Res3,
Res4, Res5 are ResNet layers of output respectively；
Step a2 obtains a semantic information using pyramid network, and specific practice is dropped to the feature above Res4
Dimension operation, that is, add the convolutional layer of one layer of 1*1, operation is up-sampled to the feature above P5, so that they are of the same size；
Step a3, to treated P5 and treated that Res4 executes add operation, and corresponding element is added, and the result of acquisition is defeated
P4 is arrived out；To treated P4 and treated that Res3 executes add operation, and corresponding element is added, the result of acquisition is output to
P3；To treated P3 and treated that Res2 executes add operation, and corresponding element is added, and the result of acquisition is output to P2；
Step a4, using area suggest that network is predicted in P2, P3, P4, P5 difference output layer, obtain area-of-interest.
3. automatic segmentation Radiotherapy of Esophageal Cancer target area as described in claim 1 and the method for jeopardizing organ, it is characterised in that: residual error
Network+feature pyramid network is a preparatory trained convolution+pond neural network.
4. automatic segmentation Radiotherapy of Esophageal Cancer target area as described in claim 1 and the method for jeopardizing organ, it is characterised in that: region
It is recommended that network is the neural network of a lightweight, it uses sliding window scan image, finds region existing for organ, simultaneously
Each point is obtained on characteristic pattern generates multiple candidate area-of-interests.
5. automatic segmentation Radiotherapy of Esophageal Cancer target area as described in claim 1 and the method for jeopardizing organ, it is characterised in that: step
In B, area-of-interest aligned layer traverses each area-of-interest first, keeps floating number boundary not do and quantifies；Then will feel emerging
Interesting region is divided into k × k cell, and the boundary of each cell is also without quantization；Then it calculates and fixes in each cell
4 coordinate positions, the value of this 4 positions is calculated by bilinear interpolation；Maximum pond operation is finally carried out, amount is being not present
Fixed-size characteristic pattern is obtained on the basis of change error.
6. automatic segmentation Radiotherapy of Esophageal Cancer target area as described in claim 1 and the method for jeopardizing organ, it is characterised in that: step
In C, organ segmentation's network is the parallel branch returned with organ classes and organ site frame.
7. automatic segmentation Radiotherapy of Esophageal Cancer target area as described in claim 1 and the method for jeopardizing organ, it is characterised in that: step
In C, organ segmentation's network includes 4 continuous convolutional layers and 1 warp lamination, and the core size of each convolutional layer is 3 × 3, instead
Convolution is 2 times of up-samplings.
8. automatic segmentation Radiotherapy of Esophageal Cancer target area as described in claim 1 and the method for jeopardizing organ, it is characterised in that: step
In C, it is classified as the heart, right lung, left lung, planning target volume (PTV), the region clinical target volume (CTV), 6 class of background.
9. a kind of automatic segmentation Radiotherapy of Esophageal Cancer target area and the system for jeopardizing organ, it is characterised in that: sieved including area-of-interest
Modeling block, pond module, organ classes' module；
Area-of-interest screening module extracts the feature of the CT image of input by residual error network, is melted by feature pyramid network
Analysis On Multi-scale Features figure is closed, suggests that network screens the area-of-interest of point each in characteristic pattern by region；
The area-of-interest pond that pond module combination area-of-interest aligned layer filters out region suggestion network is solid to one
Scale cun；
Pond to fixed-size area-of-interest is input to full articulamentum and carries out organ classes by organ classes' module, and is carried out
Organ site frame returns；Pond is input to organ segmentation's network to fixed-size area-of-interest simultaneously.
10. automatic segmentation Radiotherapy of Esophageal Cancer target area as claimed in claim 9 and the system for jeopardizing organ, it is characterised in that: sense
Interest region screening module includes residual error network module, pyramid network module, add operation module, region suggestion network mould
Block；
Residual error network module input one include multiple organ medicine CT image, by residual error network carry out depth convolution, Res2,
Res3, Res4, Res5 are ResNet layers of output respectively；
Pyramid network module obtains a semantic information using pyramid network, and specific practice is to the feature above Res4
Dimensionality reduction operation is carried out, that is, adds the convolutional layer of one layer of 1*1, operation is up-sampled to the feature above P5, so that they are with identical
Size；
Add operation module is to treated P5 and treated that Res4 executes add operation, and corresponding element is added, by acquisition
As a result P4, P2, P3 result and so on are output to；
Region suggests that network module using area suggests that network is predicted in P2, P3, P4, P5 difference output layer, the sense of access
Interest region.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910504373.5A CN110211139A (en) | 2019-06-12 | 2019-06-12 | Automatic segmentation Radiotherapy of Esophageal Cancer target area and the method and system for jeopardizing organ |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910504373.5A CN110211139A (en) | 2019-06-12 | 2019-06-12 | Automatic segmentation Radiotherapy of Esophageal Cancer target area and the method and system for jeopardizing organ |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110211139A true CN110211139A (en) | 2019-09-06 |
Family
ID=67792171
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910504373.5A Withdrawn CN110211139A (en) | 2019-06-12 | 2019-06-12 | Automatic segmentation Radiotherapy of Esophageal Cancer target area and the method and system for jeopardizing organ |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110211139A (en) |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110688978A (en) * | 2019-10-10 | 2020-01-14 | 广东工业大学 | Pedestrian detection method, device, system and equipment |
| CN110739049A (en) * | 2019-10-10 | 2020-01-31 | 上海联影智能医疗科技有限公司 | Image sketching method and device, storage medium and computer equipment |
| CN111127444A (en) * | 2019-12-26 | 2020-05-08 | 广州柏视医疗科技有限公司 | Method for automatically identifying radiotherapy organs at risk in CT image based on depth semantic network |
| CN111476775A (en) * | 2020-04-07 | 2020-07-31 | 广州柏视医疗科技有限公司 | DR symptom identification device and method |
| CN112233128A (en) * | 2020-10-15 | 2021-01-15 | 推想医疗科技股份有限公司 | Image segmentation method, model training method, device, medium, and electronic device |
| CN113516659A (en) * | 2021-09-15 | 2021-10-19 | 浙江大学 | Medical image automatic segmentation method based on deep learning |
| CN113793304A (en) * | 2021-08-23 | 2021-12-14 | 天津大学 | Intelligent segmentation method for lung cancer target area and organs at risk |
| CN114299080A (en) * | 2021-12-16 | 2022-04-08 | 西安邮电大学 | Throat organ segmentation method based on cavity residual error characteristic pyramid |
| CN116824499A (en) * | 2023-06-28 | 2023-09-29 | 北京建筑大学 | Insect pest detection method, system, equipment and storage medium based on SWT model |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107358600A (en) * | 2017-06-14 | 2017-11-17 | 北京全域医疗技术有限公司 | Automatic hook Target process, device and electronic equipment in radiotherapy planning |
| CN107403201A (en) * | 2017-08-11 | 2017-11-28 | 强深智能医疗科技（昆山）有限公司 | Tumour radiotherapy target area and jeopardize that organ is intelligent, automation delineation method |
| CN108415058A (en) * | 2018-01-23 | 2018-08-17 | 深圳市旭东数字医学影像技术有限公司 | The dose calculation methodology and system of radioactive ray |
| CN109146899A (en) * | 2018-08-28 | 2019-01-04 | 众安信息技术服务有限公司 | CT image jeopardizes organ segmentation method and device |
| CN109300136A (en) * | 2018-08-28 | 2019-02-01 | 众安信息技术服务有限公司 | It is a kind of to jeopardize organs automatic segmentation method based on convolutional neural networks |
-
2019
- 2019-06-12 CN CN201910504373.5A patent/CN110211139A/en not_active Withdrawn
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107358600A (en) * | 2017-06-14 | 2017-11-17 | 北京全域医疗技术有限公司 | Automatic hook Target process, device and electronic equipment in radiotherapy planning |
| CN107403201A (en) * | 2017-08-11 | 2017-11-28 | 强深智能医疗科技（昆山）有限公司 | Tumour radiotherapy target area and jeopardize that organ is intelligent, automation delineation method |
| CN108415058A (en) * | 2018-01-23 | 2018-08-17 | 深圳市旭东数字医学影像技术有限公司 | The dose calculation methodology and system of radioactive ray |
| CN109146899A (en) * | 2018-08-28 | 2019-01-04 | 众安信息技术服务有限公司 | CT image jeopardizes organ segmentation method and device |
| CN109300136A (en) * | 2018-08-28 | 2019-02-01 | 众安信息技术服务有限公司 | It is a kind of to jeopardize organs automatic segmentation method based on convolutional neural networks |
Non-Patent Citations (3)
| Title |
|---|
| KAIMING HE ET AL.: "Mask R-CNN", 《ARXIV:1703.06870V3》 * |
| SHUJUN LIANG ET AL.: "Deep-learning-based detection and segmentation of organ at risk in nasopharyngeal carcinoma computed tomographic images for radiotherapy planning", 《EUROPEAN RADIOLOGY》 * |
| TSUNG-YI LIN ET AL.: "Feature Pyramid Networks for Object Detection", 《ARXIV:1612.03144V2》 * |
Cited By (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110688978A (en) * | 2019-10-10 | 2020-01-14 | 广东工业大学 | Pedestrian detection method, device, system and equipment |
| CN110739049A (en) * | 2019-10-10 | 2020-01-31 | 上海联影智能医疗科技有限公司 | Image sketching method and device, storage medium and computer equipment |
| CN111127444A (en) * | 2019-12-26 | 2020-05-08 | 广州柏视医疗科技有限公司 | Method for automatically identifying radiotherapy organs at risk in CT image based on depth semantic network |
| CN111476775A (en) * | 2020-04-07 | 2020-07-31 | 广州柏视医疗科技有限公司 | DR symptom identification device and method |
| CN112233128A (en) * | 2020-10-15 | 2021-01-15 | 推想医疗科技股份有限公司 | Image segmentation method, model training method, device, medium, and electronic device |
| CN112233128B (en) * | 2020-10-15 | 2021-11-02 | 推想医疗科技股份有限公司 | Image segmentation method, model training method, device, medium, and electronic device |
| CN113793304A (en) * | 2021-08-23 | 2021-12-14 | 天津大学 | Intelligent segmentation method for lung cancer target area and organs at risk |
| CN113516659A (en) * | 2021-09-15 | 2021-10-19 | 浙江大学 | Medical image automatic segmentation method based on deep learning |
| CN114299080A (en) * | 2021-12-16 | 2022-04-08 | 西安邮电大学 | Throat organ segmentation method based on cavity residual error characteristic pyramid |
| CN116824499A (en) * | 2023-06-28 | 2023-09-29 | 北京建筑大学 | Insect pest detection method, system, equipment and storage medium based on SWT model |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110211139A (en) | Automatic segmentation Radiotherapy of Esophageal Cancer target area and the method and system for jeopardizing organ | |
| Yun et al. | Improvement of fully automated airway segmentation on volumetric computed tomographic images using a 2.5 dimensional convolutional neural net | |
| Zhou et al. | Deep learning of the sectional appearances of 3D CT images for anatomical structure segmentation based on an FCN voting method | |
| Cusumano et al. | Artificial Intelligence in magnetic Resonance guided Radiotherapy: Medical and physical considerations on state of art and future perspectives | |
| CN108257135A (en) | The assistant diagnosis system of medical image features is understood based on deep learning method | |
| Lei et al. | 4D-CT deformable image registration using an unsupervised deep convolutional neural network | |
| CN107563383A (en) | A kind of medical image auxiliary diagnosis and semi-supervised sample generation system | |
| CN110544264A (en) | Temporal bone key anatomical structure small target segmentation method based on 3D deep supervision mechanism | |
| CN107203989A (en) | End-to-end chest CT image dividing method based on full convolutional neural networks | |
| Chen et al. | A lung dense deep convolution neural network for robust lung parenchyma segmentation | |
| CN110110723A (en) | A kind of method and device that objective area in image automatically extracts | |
| CN106846330A (en) | Human liver's feature modeling and vascular pattern space normalizing method | |
| Wang et al. | VMAT dose prediction in radiotherapy by using progressive refinement UNet | |
| CN110047075A (en) | A kind of CT image partition method based on confrontation network | |
| CN106127783A (en) | A kind of medical imaging identification system based on degree of depth study | |
| La Rosa | A deep learning approach to bone segmentation in CT scans | |
| Banerjee et al. | A CADe system for gliomas in brain MRI using convolutional neural networks | |
| Peng et al. | Pulmonary lobe segmentation in CT images based on lung anatomy knowledge | |
| Gupta et al. | Implementation of classification system for brain cancer using backpropagation network and MRI | |
| Singh et al. | Semantic segmentation of bone structures in chest X-rays including unhealthy radiographs: A robust and accurate approach | |
| Saha et al. | A survey on artificial intelligence in pulmonary imaging | |
| Lee et al. | Generative adversarial network with radiomic feature reproducibility analysis for computed tomography denoising | |
| Iyer et al. | Deep learning-based auto-segmentation of swallowing and chewing structures | |
| Xu et al. | Improved cascade R-CNN for medical images of pulmonary nodules detection combining dilated HRNet | |
| Zhao et al. | Automatic segmentation method for bone and blood vessel in murine hindlimb |
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 | ||
| WW01 | Invention patent application withdrawn after publication | ||
| WW01 | Invention patent application withdrawn after publication |
Application publication date: 20190906 |