CN110136119A - A kind of lung based on deep learning splits the method and system of segmentation and integrity assessment - Google Patents
A kind of lung based on deep learning splits the method and system of segmentation and integrity assessment Download PDFInfo
- Publication number
- CN110136119A CN110136119A CN201910406447.1A CN201910406447A CN110136119A CN 110136119 A CN110136119 A CN 110136119A CN 201910406447 A CN201910406447 A CN 201910406447A CN 110136119 A CN110136119 A CN 110136119A
- Authority
- CN
- China
- Prior art keywords
- lung
- splits
- segmentation
- split
- data
- 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
- G06T7/10—Segmentation; Edge detection
- G06T7/11—Region-based segmentation
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Medical Informatics (AREA)
- Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
- Radiology & Medical Imaging (AREA)
- Quality & Reliability (AREA)
- Apparatus For Radiation Diagnosis (AREA)
- Image Processing (AREA)
Abstract
The invention discloses the method and systems that a kind of lung based on deep learning splits segmentation and integrity assessment.Compared to conventional method, hence it is evident that improve precision and computational efficiency that lung splits segmentation, realize that full automatic lung splits segmentation and lung splits integrity assessment.Wherein the key step of summary of the invention includes: that building lung splits partitioned data set;Lung of the training based on full convolutional neural networks splits parted pattern;Prediction lung split plot domain simultaneously identifies the left fissura obliqua pulmonis of acquisition, right lung oblique segmentation, horizontal fissure of right lung;Estimate that complete lung is split;Assessment lung splits integrated degree.Due to using full convolutional neural networks, it realizes lung end to end and splits model training and prediction, without manual intervention, predetermined speed is fast, and use segmentation framework from coarse to fine, it solves class label quantity extreme imbalance problem when being split task using full convolutional neural networks, and eliminates the vacation sun that lung splits segmentation generation by introducing lobe of the lung segmentation, so that lung is split integrity assessment more accurate.
Description
Technical field
The present invention relates to medical image processing field, in particular to a kind of lung based on deep learning splits segmentation and integrality
The method and system of assessment.
Background technique
Lung is split to be had very important effect in the detection, classification and assessment of pulmonary disease, therefore lung is accurately positioned and splits
Simultaneously divide lung and split and had a very important significance in terms of Diagnosis of Pulmonary Diseases in region.In CT images, lung is split in two dimensional slice knot
The curve for Curvature varying very little is cashed in structure, then shows as ribbon structure or planar structure in a three-dimensional structure.It is examined in clinic
In disconnected, understanding lung, which splits structure feature, facilitates the positioning of pulmonary lesions and the qualitative assessment of pulmonary disease.And it is wanted in CT images
At automatic lung split segmentation be it is very difficult, lung, which is split often, has imperfect, deformation, fracture and attached phenomena such as splitting.Lung is split at present
Partitioning algorithm is faced with maximum challenge: lung splits segmentation.
A large amount of detection dividing method (such as segmentation of air flue, blood vessel and lung) is related with the CT images of lung and develops into
Ripe, but the detection dividing method that lung is split is still under study for action, lung splits segmentation to be generally concentrated at computational geometry method automatic at present
In the method that detection segmentation lung is split, but this scheme, there are limitation, the method needs can reach under the conditions of " advantageous "
Good effect, i.e. its detection need precondition;The execution efficiency of this scheme is low.It and is based on full volume in the present invention
Product neural network algorithm is analyzed CT images, is handled, and realizes that automatic lung splits segmentation, full convolutional neural networks algorithm has certainly
The characteristics of I learns, constantly improve has lesser limitation, and completes lung using neural network algorithm for doctor and split segmentation
Implementation rate will be improved, error is reduced.
Summary of the invention
This method provides the method and system that a kind of lung based on deep learning splits segmentation and integrity assessment, and feature exists
Split partitioned data set in, comprising the following steps: S1) building lung: acquisition chest CT images carry out classification mark to lung split plot domain,
Line number of going forward side by side Data preprocess;S2) training lung splits parted pattern: the labeled data based on step S1) builds full convolutional neural networks
It is trained, obtains lung and split parted pattern;S3 lung split plot domain) is predicted: the data prediction by data by step S1), then
It is input to step S2) obtained lung splits parted pattern, and it obtains lung and splits segmentation result, and is false using the segmentation removal of the existing lobe of the lung
Sun;S4) estimate that complete lung is split: the one way in which for dividing or being fitted three-dimension curved surface by the lobe of the lung obtains complete lung and splits;
S5) lung splits integrality calculating: calculating lung to different lung split plot domain and splits integrality, calculation formula are as follows: lung splits integrality=pre-
The lung of survey splits the complete lung broken face product of region area/estimation.
Optionally, step S1) in, lung is split according to Clinical anatomic structure and is divided into three regions and is labeled by doctor, respectively
Are as follows: left fissura obliqua pulmonis, right lung oblique segmentation, horizontal fissure of right lung.
Optionally, step S1) in, data prediction further comprises: being normalized to data, carries out to data
Interpolation, so that data are divided into d1, d2, d3, and d1, d2 between the physical picture element on x, tri- directions y, z, d3 is greater than 0
Number;The size requirement that sliding stripping and slicing makes data block meet neural network input is carried out to data;Data augmentation is carried out, including is put
Contracting, rotation, changes axis, gaussian filtering, the mapping modes such as Lightness disposal.
Optionally, step S2) in, the full convolutional neural networks structure for splitting segmentation to training lung further comprises: this is complete
Convolutional neural networks are parted pattern from coarse to fine;The full convolutional neural networks are made of 2 concatenated sub-networks, the subnet
Network is input size full convolutional neural networks identical with Output Size;Also, the input of first sub-network is original image, output
For the ROI region that the lung of prediction is split, the input of second sub-network be original image with first network ROI predict to be multiplied as a result,
Output splits accurate region for the lung of prediction;The output of first sub-network is activated by Softmax function, is exported as 2 classifications,
Respectively background and ROI region;The output of second sub-network is activated by Softmax function, is exported as 4 classifications, difference
For left fissura obliqua pulmonis, right lung oblique segmentation, horizontal fissure of right lung and background.
Optionally, step S3) in, for the data that lung to be predicted is split, handled by the data normalization of step S1) and sliding
Dynamic stripping and slicing, and be input to step S2) the obtained lung of training splits parted pattern and is predicted, and prediction result is backfilled, it obtains
Lung to prediction splits segmentation result, if it is there is overlapping stripping and slicing, is then averaged to probability.
Optionally, step S3) in, the mode for removing false sun is, to the data progress lobe of the lung segmentation that lung to be predicted is split, if
It obtains lung and splits the connected domain of segmentation result only falling in a lobe of the lung, then it is false sun that the lung that this segmentation obtains, which splits connected domain, will
This block connected domain is deleted.
Optionally, step S4) in, split and further comprised using the complete lung of lobe of the lung partitioning estimation: lobe of the lung segmentation obtains
The superior lobe of right lung face adjacent with the middle lobe of right lung smoothly obtains complete horizontal fissure of right lung by curved surface, and area is horizontal fissure of right lung
Complete area;The inferior lobe of right lung that lobe of the lung segmentation the obtains face adjacent with middle lobe of right lung, inferior lobe of right lung is smoothly obtained by curved surface
Complete right lung oblique segmentation, area are the complete area of right lung oblique segmentation;The upper lobe of left lung and lobe of left lung phase that lobe of the lung segmentation obtains
Adjacent face smoothly obtains complete left fissura obliqua pulmonis by curved surface, and area is the complete area of left fissura obliqua pulmonis.
Optionally, step S5) in, it is step S3 that the lung of prediction, which splits region area) predict the surface area that obtained lung is split, it is complete
Whole each lung broken face product is step S4) obtained complete lung broken face product, by the formula evaluate respectively left fissura obliqua pulmonis, right lung oblique segmentation,
The lung of horizontal fissure of right lung splits integrated degree.
Optionally, when pre-processing stripping and slicing, can in the encirclement frame of lung to data along the sagittal plane or hat of CT images
Shape face or horizontal plane direction carry out stripping and slicing, obtain 3D data block, and stripping and slicing size is determined according to network inputs size and video memory size.
Optionally, the sub-network of full convolutional network is the 3 U-shaped full convolutional neural networks of dimension, (maximum comprising 3 down-sampling layers
Pond layer) and 3 up-sampling layers (warp lamination), and it is intermediate by stacking connection (Concatenate), each down-sampling layer
With up-sampling several convolution blocks of layer heel, each convolution block includes 3 dimensions convolution (3DConv), crowd normalization (Batch
Normalization), nonlinear activation (ReLU), the last layer of full convolutional neural networks is Softmax activation primitive, defeated
Port number out is determined according to classification number.
Compared with the prior art, the invention has the following advantages: 1) lung of the invention splits dividing method with better Shandong
Stick, accuracy, and more mark training datas can be increased to keep model more accurate;2) lung of the invention is split completely
Property appraisal procedure be based on full convolutional neural networks, it is achieved that full automatic lung splits integrity assessment, without artificial dry
In advance;3) present invention adds lobe of the lung segmentations to remove false sun, so that lung is split integrity assessment more accurate;4) the present invention provides from
It is thick to arrive thin full convolutional neural networks model, it provides and segmentation is split according to accurate lung.
Detailed description of the invention
It is a kind of structural representation for the method and system that the lung based on deep learning splits segmentation and integrity assessment described in Fig. 1
Figure.
It is that the lung that the present invention predicts splits segmentation result (two dimension view) described in Fig. 2.
Be step S3 of the present invention described in Fig. 3) described in the lung of prediction split segmentation result (3-D view).
Be step S4 of the present invention described in Fig. 4) described in complete lung split (3-D view).
It is the network structure that the present invention splits segmentation for lung described in Fig. 5.
Specific embodiment
To make the objectives, technical solutions, and advantages of the present invention clearer, below in conjunction with attached drawing to the present invention make into
It is described in detail to one step, it is clear that described embodiments are only a part of the embodiments of the present invention, rather than whole implementation
Example.Based on the embodiments of the present invention, obtained by those of ordinary skill in the art without making creative efforts
All other embodiment, shall fall within the protection scope of the present invention.
Fig. 1 is a kind of structural schematic diagram for the method and system that lung based on deep learning splits segmentation and integrity assessment.
Its key step includes: that building lung splits partitioned data set;Lung of the training based on full convolutional neural networks splits parted pattern;Predict lung
Split plot domain simultaneously identifies the left fissura obliqua pulmonis of acquisition, right lung oblique segmentation, horizontal fissure of right lung;Estimate that complete lung is split;Assessment lung splits integrated degree.For
Facilitate the every details understood in invention, with construct lung split partitioned data set to assessment lung split integrality, be described in detail.
(1) building lung splits partitioned data set, including lung splits label and pretreatment.
Lung splits the mode of mark are as follows: veteran doctor to visible left fissura obliqua pulmonis, right lung oblique segmentation, horizontal fissure of right lung this
Three kinds of category regions carry out Pixel-level label, and give the result marked to another doctor and check, Yi Shenghe
After errorless to confirmation mark, data are adopted as training data, this labeled data is otherwise abandoned or marks again.220 are marked altogether
Number of cases evidence, and the ratio of ill agnogenic diffusivity tuberculosis and the data without agnogenic diffusivity tuberculosis is 1:1.
After the completion of mark, by the x of data, y, z(, that is, horizontal axis, the longitudinal axis, vertical axis) physical picture element interval be interpolated into 1.2 milli
Rice, Hu value interception window ranges are [- 1000, -200] and are normalized to 0 to 1 codomain range, then are zoomed between -1 to 1.
And lung is taken to surround the data in frame, and it takes lung 3D to surround the data in frame, discards the data that 3D surrounds outer frame, rule
Surely it is x, y, z, unit: pixel that data block size when trained neural network, which is 48*196*256(sequence).Along x-axis when stripping and slicing
(horizontal axis) direction stripping and slicing, stripping and slicing size is 48*196*256, if block size is unsatisfactory for 48*196*256 size, with central point
Do filling or intercepting process.And data augmentation, including scaling are carried out, it rotates, changes axis, gaussian filtering, the processing such as Lightness disposal.
(2) lung of the training based on full convolutional neural networks splits parted pattern.
The full convolutional neural networks that training lung splits segmentation are that two U-shaped convolutional neural networks are composed in series, and constitute a W type
Network, as shown in figure 5, first sub- role of network is the ROI region predicting lung and splitting, the effect of second sub-network is prediction
Accurate lung split plot domain in ROI region.The input of first sub-network is original image, exports the ROI region split for the lung of prediction, the
The input of two sub-networks is original image multiplied by the output in first network ROI region channel, exports and splits accurate area for the lung of prediction
Domain.The output of first sub-network is activated by Softmax, is exported as 2 classifications, respectively background, ROI region, second
The output of sub-network is activated by Softmax function, is exported as 4 classifications (channel), respectively background, left fissura obliqua pulmonis, right lung
Oblique segmentation, horizontal fissure of right lung.
It is exactly the port number difference exported that the hyper parameter of two sub-networks, which is uniquely distinguished, other hyper parameters are all identical
, comprising: sub-network is the 3 U-shaped neural networks of dimension, includes 3 down-sampling layers (maximum pond layer) and 3 up-sampling layer (warps
Lamination), and it is intermediate by stacking connection (Concatenate), each down-sampling layer and up-sampling 2 convolution blocks of layer heel,
Each convolution block includes 3 dimensions convolution (3DConv), criticizes and normalizes (Batch normalization), nonlinear activation (ReLU),
The last layer of full convolutional neural networks is Softmax activation primitive.The optimizer of training network is Adam, and initial learning rate is
0.001.By training in step (1) treated data are input to above-mentioned 3 dimension W type neural networks when training, when on verifying collection
Loss continuous 3 epoch then deconditioning when no longer declining.
(3) it predicts lung split plot domain and identifies to obtain left fissura obliqua pulmonis, right lung oblique segmentation, horizontal fissure of right lung.
A new CT images data are obtained, data are passed through to the pretreatment of step (1), output data to step (2)
It is predicted in trained model, if having overlapping when prediction data stripping and slicing, obtained segmentation result overlapping region
Probability by be added merged, on last image the classification of each pixel pass through argmax(i.e. 4 channel maximum probability
Classification be current pixel classification) obtain, network divides the image into the region of four classifications, respectively left fissura obliqua pulmonis, the right side
Fissura obliqua pulmonis, horizontal fissure of right lung and background split segmentation result eventually by the lung that the mode of backfill obtains entire lung.
Further, the vacation sun that lung is split is removed by carrying out lobe of the lung segmentation to data, the mode for removing false sun is logarithm
It according to lobe of the lung segmentation is carried out, obtains lung and splits the connected region of segmentation result only falling in a lobe of the lung, then this segmentation obtains lung and splits
Connected region is false sun, this block region is deleted.
(4) estimate that complete lung is split.
Calculating lung and splitting integrality and need to obtain lung to split hypothesis is complete area, i.e., splits lack part by completion lung to estimate
Count whole face product.The complete area split by lobe of the lung partitioning estimation lung, comprising: in the superior lobe of right lung and right lung that lobe of the lung segmentation obtains
The adjacent face of leaf smoothly obtains complete horizontal fissure of right lung by curved surface, and area is the complete area of horizontal fissure of right lung;The lobe of the lung
Divide the inferior lobe of right lung obtained the face adjacent with middle lobe of right lung, inferior lobe of right lung and smoothly obtain complete right lung oblique segmentation by curved surface,
Its area is the complete area of right lung oblique segmentation;The upper lobe of left lung that lobe of the lung segmentation the obtains face adjacent with lobe of left lung is flat by curved surface
Sliding to obtain complete left fissura obliqua pulmonis, area is the complete area of left fissura obliqua pulmonis.
(5) assessment lung splits integrated degree.
The lung split plot field surface product predicted by step (3), the complete lung broken face estimated by step (4)
Product, if it is incomplete (in CT images lung split there are a part of invisible) that lung, which is split, step (3) to be imperfect
Lung split.And what step (4) obtained is that complete lung splits surface area.Lung splits integrality can be by incomplete area ratio
The ratio of complete area obtains, it may be assumed that lung splits integrality=prediction lung and splits the complete lung broken face product of region area/estimation.
Claims (11)
1. the method and system that a kind of lung based on deep learning splits segmentation and integrity assessment, which is characterized in that including following
Step:
S1) building lung splits partitioned data set: acquisition chest CT images carry out classification mark to lung split plot domain, and it is pre- to carry out data
Processing;
S2) training lung splits parted pattern: the labeled data based on step S1) is built full convolutional neural networks and is trained, obtains
Lung splits parted pattern;
S3) predict lung split plot domain: data passed through into step S1) data prediction, be then input to step S2) obtained lung splits
Parted pattern obtains lung and splits segmentation result, and utilizes the false sun of existing lobe of the lung segmentation removal;
S4) estimate that complete lung is split: the one way in which for dividing or being fitted three-dimension curved surface by the lobe of the lung obtains complete lung and splits;
S5) lung splits integrality calculating: calculating lung to different lung split plot domain and splits integrality, calculation formula are as follows: lung splits integrality=pre-
The lung of survey splits the complete lung broken face product of region area/estimation.
2. the method as described in claim 1, which is characterized in that the step S1) in, doctor, will according to Clinical anatomic structure
Lung, which is split, to be divided into three regions and is labeled, and left fissura obliqua pulmonis, right lung oblique segmentation, horizontal fissure of right lung are respectively as follows:.
3. the method as described in claim 1, which is characterized in that the step S1) in, data prediction further comprises: right
Data are normalized, and interpolation are carried out to data, so that data are divided between the physical picture element on x, tri- directions y, z
D1, d2, d3, and d1, d2, d3 are the number greater than 0;Carrying out sliding stripping and slicing to data makes data block meet neural network input
Size requirement;Data augmentation, including scaling are carried out, axis, gaussian filtering, the mapping modes such as Lightness disposal are changed in rotation.
4. the method as described in claim 1, which is characterized in that the step S2) in, the full convolution of segmentation is split to training lung
Neural network structure further comprises: the full convolutional neural networks are parted pattern from coarse to fine;The full convolutional neural networks
It is made of 2 concatenated sub-networks, which is input size full convolutional neural networks identical with Output Size;Also,
The input of first sub-network is original image, exports the ROI region split for the lung of prediction, the input of second sub-network be original image with
What first network ROI prediction was multiplied splits accurate region as a result, exporting for the lung of prediction;The output of first sub-network passes through
The activation of Softmax function, exports as 2 classifications, respectively background and ROI region;The output of second sub-network passes through
The activation of Softmax function, exports as 4 classifications, respectively left fissura obliqua pulmonis, right lung oblique segmentation, horizontal fissure of right lung and background.
5. the method as described in claim 1, which is characterized in that the step S3) in, for the data that lung to be predicted is split, warp
Cross step S1) data normalization processing and sliding stripping and slicing, and be input to step S2) the obtained lung of training splits parted pattern progress
Prediction, and prediction result is backfilled, the lung predicted splits segmentation result.
6. the method as described in claim 1, which is characterized in that the step S3) in, the mode for removing false sun is, to pre-
It surveys the data split of lung and carries out lobe of the lung segmentation, split the connected domain of segmentation result if claim 5 obtains lung and only fall in a lobe of the lung
In, then this segmentation obtains lung to split connected domain being false sun, this block connected domain is deleted.
7. the method as described in claim 1, which is characterized in that the step S4) in, it is split using the complete lung of lobe of the lung partitioning estimation
Further comprise: the superior lobe of right lung that lobe of the lung segmentation the obtains face adjacent with the middle lobe of right lung smoothly obtains the complete right side by curved surface
Edema with the lung involved plane fracture, area are the complete area of horizontal fissure of right lung;Under inferior lobe of right lung and middle lobe of right lung, right lung that lobe of the lung segmentation obtains
The adjacent face of leaf smoothly obtains complete right lung oblique segmentation by curved surface, and area is the complete area of right lung oblique segmentation;Lobe of the lung segmentation
The upper lobe of left lung of the acquisition face adjacent with lobe of left lung smoothly obtains complete left fissura obliqua pulmonis by curved surface, and area is that left lung is oblique
The complete area split.
8. the method as described in claim 1, which is characterized in that the step S5) in, it is step that the lung of prediction, which splits region area,
S3) the surface area that the lung that prediction obtains is split, the complete lung broken face product that complete each lung broken face product obtains for step S4), by the public affairs
Formula evaluate respectively left fissura obliqua pulmonis, right lung oblique segmentation, horizontal fissure of right lung lung split integrated degree.
9. method as claimed in claim 3, which is characterized in that data along the sagittal plane of CT images in the encirclement frame of lung
Or coronal-plane or horizontal plane direction carry out stripping and slicing, obtain 3D data block.
10. method as claimed in claim 4, which is characterized in that the sub-network is the 3 U-shaped full convolutional neural networks of dimension, includes
3 down-sampling layers (maximum pond layer) and 3 up-sampling layers (warp lamination), and it is intermediate by stacking connection
(Concatenate), each down-sampling layer and up-sampling several convolution blocks of layer heel, each convolution block include 3 dimension convolution
(3DConv), batch normalization (Batch normalization), nonlinear activation (ReLU), full convolutional neural networks it is last
One layer is Softmax activation primitive.
11. the method and system that a kind of lung based on deep learning splits segmentation and integrity assessment characterized by comprising adopt
Collect chest CT image data, is pre-processed by data normalization as claimed in claim 3 and interpolation;It surrounds in frame in lung to data
Stripping and slicing is carried out, stripping and slicing is input to step S2) obtained lung splits in parted pattern and predicts, and prediction result is returned
It fills out, the lung predicted splits segmentation result;Mode according to claim 6 removes lung and splits false sun;Mode according to claim 7
Estimate that complete lung is split;According to formula: lung splits integrality=prediction lung and splits the complete lung broken face product of region area/estimation, calculates not
Same lung split plot domain calculates lung and splits integrality.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910406447.1A CN110136119A (en) | 2019-05-16 | 2019-05-16 | A kind of lung based on deep learning splits the method and system of segmentation and integrity assessment |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910406447.1A CN110136119A (en) | 2019-05-16 | 2019-05-16 | A kind of lung based on deep learning splits the method and system of segmentation and integrity assessment |
Publications (1)
Publication Number | Publication Date |
---|---|
CN110136119A true CN110136119A (en) | 2019-08-16 |
Family
ID=67574472
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910406447.1A Pending CN110136119A (en) | 2019-05-16 | 2019-05-16 | A kind of lung based on deep learning splits the method and system of segmentation and integrity assessment |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110136119A (en) |
Cited By (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111260669A (en) * | 2020-02-17 | 2020-06-09 | 北京推想科技有限公司 | Lung lobe segmentation method and device based on CT image |
CN111292339A (en) * | 2020-01-22 | 2020-06-16 | 北京航空航天大学 | Clinical temporal bone CT multi-structure extraction method and device based on W-type network structure |
CN111402231A (en) * | 2020-03-16 | 2020-07-10 | 杭州健培科技有限公司 | Automatic evaluation system and method for lung CT image quality |
CN111415341A (en) * | 2020-03-17 | 2020-07-14 | 北京推想科技有限公司 | Pneumonia stage evaluation method, pneumonia stage evaluation device, pneumonia stage evaluation medium and electronic equipment |
CN112907651A (en) * | 2021-03-29 | 2021-06-04 | 山东捷瑞数字科技股份有限公司 | Measuring method of oyster external form based on semantic segmentation network |
CN113160186A (en) * | 2021-04-27 | 2021-07-23 | 青岛海信医疗设备股份有限公司 | Lung lobe segmentation method and related device |
CN114092470A (en) * | 2021-12-08 | 2022-02-25 | 之江实验室 | Deep learning-based automatic detection method and device for pulmonary fissure |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106875379A (en) * | 2017-01-10 | 2017-06-20 | 陕西渭南神州德信医学成像技术有限公司 | Lung splits integrity degree appraisal procedure, device and system |
CN109584252A (en) * | 2017-11-03 | 2019-04-05 | 杭州依图医疗技术有限公司 | Lobe of the lung section dividing method, the device of CT images based on deep learning |
CN109598727A (en) * | 2018-11-28 | 2019-04-09 | 北京工业大学 | A kind of CT image pulmonary parenchyma three-dimensional semantic segmentation method based on deep neural network |
CN109636808A (en) * | 2018-11-27 | 2019-04-16 | 杭州健培科技有限公司 | A kind of lobe of the lung dividing method based on full convolutional neural networks |
-
2019
- 2019-05-16 CN CN201910406447.1A patent/CN110136119A/en active Pending
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106875379A (en) * | 2017-01-10 | 2017-06-20 | 陕西渭南神州德信医学成像技术有限公司 | Lung splits integrity degree appraisal procedure, device and system |
CN109584252A (en) * | 2017-11-03 | 2019-04-05 | 杭州依图医疗技术有限公司 | Lobe of the lung section dividing method, the device of CT images based on deep learning |
CN109636808A (en) * | 2018-11-27 | 2019-04-16 | 杭州健培科技有限公司 | A kind of lobe of the lung dividing method based on full convolutional neural networks |
CN109598727A (en) * | 2018-11-28 | 2019-04-09 | 北京工业大学 | A kind of CT image pulmonary parenchyma three-dimensional semantic segmentation method based on deep neural network |
Cited By (11)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111292339A (en) * | 2020-01-22 | 2020-06-16 | 北京航空航天大学 | Clinical temporal bone CT multi-structure extraction method and device based on W-type network structure |
CN111292339B (en) * | 2020-01-22 | 2023-01-10 | 北京航空航天大学 | Clinical temporal bone CT multi-structure extraction method and device based on W-type network structure |
CN111260669A (en) * | 2020-02-17 | 2020-06-09 | 北京推想科技有限公司 | Lung lobe segmentation method and device based on CT image |
CN111402231A (en) * | 2020-03-16 | 2020-07-10 | 杭州健培科技有限公司 | Automatic evaluation system and method for lung CT image quality |
CN111402231B (en) * | 2020-03-16 | 2023-05-23 | 杭州健培科技有限公司 | Automatic evaluation system and method for lung CT image quality |
CN111415341A (en) * | 2020-03-17 | 2020-07-14 | 北京推想科技有限公司 | Pneumonia stage evaluation method, pneumonia stage evaluation device, pneumonia stage evaluation medium and electronic equipment |
CN112907651A (en) * | 2021-03-29 | 2021-06-04 | 山东捷瑞数字科技股份有限公司 | Measuring method of oyster external form based on semantic segmentation network |
CN113160186A (en) * | 2021-04-27 | 2021-07-23 | 青岛海信医疗设备股份有限公司 | Lung lobe segmentation method and related device |
CN113160186B (en) * | 2021-04-27 | 2022-10-25 | 青岛海信医疗设备股份有限公司 | Lung lobe segmentation method and related device |
CN114092470A (en) * | 2021-12-08 | 2022-02-25 | 之江实验室 | Deep learning-based automatic detection method and device for pulmonary fissure |
CN114092470B (en) * | 2021-12-08 | 2022-08-09 | 之江实验室 | Deep learning-based automatic detection method and device for pulmonary fissure |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110136119A (en) | A kind of lung based on deep learning splits the method and system of segmentation and integrity assessment | |
JP6993371B2 (en) | Computed tomography lung nodule detection method based on deep learning | |
JP7154322B2 (en) | Medical image processing method and apparatus, electronic equipment and storage medium | |
US9898825B2 (en) | Segmentation of magnetic resonance imaging data | |
De-Maeztu et al. | Linear stereo matching | |
Digne | Similarity based filtering of point clouds | |
CN109767440A (en) | A kind of imaged image data extending method towards deep learning model training and study | |
CN106600571A (en) | Brain tumor automatic segmentation method through fusion of full convolutional neural network and conditional random field | |
CN106355570A (en) | Binocular stereoscopic vision matching method combining depth characteristics | |
CN109727253A (en) | Divide the aided detection method of Lung neoplasm automatically based on depth convolutional neural networks | |
US10878574B2 (en) | 3D quantitative analysis of retinal layers with deep learning | |
Barman et al. | MRI image segmentation using level set method and implement an medical diagnosis system | |
CN107767362A (en) | A kind of early screening of lung cancer device based on deep learning | |
CN111383215A (en) | Focus detection model training method based on generation of confrontation network | |
CN109887008B (en) | Method, device and equipment for parallax stereo matching based on forward and backward smoothing and O (1) complexity | |
CN112734822A (en) | Stereo matching algorithm based on infrared and visible light images | |
CN113658165A (en) | Cup-to-tray ratio determining method, device, equipment and storage medium | |
CN117315210B (en) | Image blurring method based on stereoscopic imaging and related device | |
CN111815610A (en) | Lesion focus detection method and device of lesion image | |
Ramana | Alzheimer disease detection and classification on magnetic resonance imaging (MRI) brain images using improved expectation maximization (IEM) and convolutional neural network (CNN) | |
EP3018626B1 (en) | Apparatus and method for image segmentation | |
CN111967462A (en) | Method and device for acquiring region of interest | |
CN116342444A (en) | Dual-channel multi-mode image fusion method and fusion imaging terminal | |
CN110163847A (en) | Liver neoplasm dividing method and device based on CT/MR image | |
Mohammed et al. | Esophageal Cancer Detection Using Feed-Forward Neural Network |
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 | ||
WD01 | Invention patent application deemed withdrawn after publication | ||
WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20190816 |