CN109636808A - A kind of lobe of the lung dividing method based on full convolutional neural networks - Google Patents
A kind of lobe of the lung dividing method based on full convolutional neural networks Download PDFInfo
- Publication number
- CN109636808A CN109636808A CN201811425317.4A CN201811425317A CN109636808A CN 109636808 A CN109636808 A CN 109636808A CN 201811425317 A CN201811425317 A CN 201811425317A CN 109636808 A CN109636808 A CN 109636808A
- Authority
- CN
- China
- Prior art keywords
- lung
- lobe
- data
- neural networks
- convolutional neural
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- 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
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- 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
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30004—Biomedical image processing
- G06T2207/30061—Lung
Abstract
The present invention provides a kind of lobe of the lung dividing methods based on full convolutional neural networks.Its technical solution includes: building lobe of the lung partitioned data set；The 3D for obtaining lung organ surrounds frame；The lung 3D data surrounded in frame are pre-processed；Data block is input in full convolutional neural networks and is trained；Data block is input in trained network and is predicted.Due to using full convolutional neural networks, realizing training end to end and predicting, be not necessarily to manual intervention, predetermined speed is fast；And it is split using being surrounded in frame in lung, eliminates lung 3D and surround the interference that outer frame information divides the lobe of the lung, the integrality and details of lobe of the lung segmentation are substantially better than conventional method；For there is the lung CT data of obvious illness also can preferably realize lobe of the lung region segmentation, so that assessing pulmonary lesion for further quantitative and qualitative provides technical support.Compared to traditional algorithm, this inventive method significantly improves the precision of lobe of the lung segmentation, realizes full-automatic lobe of the lung segmentation.
Description
Technical field
The present invention designs medical image processing field, in particular to a kind of lobe of the lung segmentation side based on full convolutional neural networks
Method.
Background technique
In recent years, due to the fast development of Medical Imaging and computer technology, computer technology is better blended into
Developed in Medical Imaging is already trend of the times.The CT image of fine definition, high contrast, is generally applied to lung
The diagnosis of disease.Lung mechanics are observed by chest CT and functional character is current clinical for the important auxiliary of the various diseases of lung
Assistant's section is conducive to find and treat conditions of patients early, it usually needs to chest to provide physicians with reliable diagnostic data
CT image carries out subsequent processing, extracts i.e. segmentation lung tissue image.
It is cut currently, many dividing methods are applied to lung differentiation, technology has: (1) threshold method is the most common lung segmentation
Method, although simply, quickly, cannot be removed effectively background and tracheae branch, and threshold value is more difficult, often rule of thumb really
It is fixed.(2) region growth method is the method used in most of work, and the omission that this method can effectively make up Edge Following lacks
It falls into, but usually needs to manually select seed point, be a kind of semi-automatic partition method for needing manually to participate in；(3) based on mode point
The method of class.This method can extract the characteristics of image of some data, but need a large amount of training sample, and segmentation result is to sample
Strong with the dependence of feature, the processing time is longer.(4) method based on image registration and shape, the general effect of this method
Preferably, but its by training set data influence to will lead to result variability big, it is more difficult and computationally intensive to establish model, from
And cause speed slow, it is difficult to meet the real-time demand of clinical application.
In conclusion numerous traditional dividing methods by the multi-party factor such as each separating step due to being influenced, it is difficult to point
Ideal effect is cut out, robustness is not strong, needs to be improved, and promotes splitting speed and precision, meets medical diagnosis to lung
The requirement of image.
Summary of the invention
The object of the present invention is to provide a kind of lobe of the lung dividing methods based on full convolutional neural networks, it is intended to improve lung
Leaf divides efficiency, and the accuracy of lobe of the lung segmentation is turned up, and compared to the lobe of the lung dividing method of basic traditional images algorithm,
This method has higher robustness.
To achieve the above object, a kind of technical solution provided by the present invention are as follows: lobe of the lung based on full convolutional neural networks
Dividing method, key step include: (a) building training dataset: acquisition CT images carry out lobe of the lung region different classes of
Mark, and data are pre-processed.(b) it obtains lung and surrounds frame: in the training stage, according to the mark side in step (a)
Method or lung segmentation, the 3D for obtaining lung areas in CT images surround frame；In forecast period, divided using existing lung partitioning algorithm
Lung areas, the 3D for obtaining lung areas in CT images surround frame.(c) it data stripping and slicing: is surrounded in the lung candidate 3D of step (b)
Carry out data stripping and slicing in frame, lung encirclement frame be cut into several data blocks, data block is filled or is sheared or
It does not handle to meet data size requirement.(d) data block in the step (c) training pattern: is supplied to full convolutional Neural
Network is trained, and obtains lobe of the lung parted pattern.(e) lobe of the lung is divided: by data by the image preprocessing in step (a), so
Afterwards by step (b) and step (c) process, multiple data blocks that lung 3D is surrounded in frame are obtained, data block passes through step (d) lung
Leaf parted pattern obtains the lobe of the lung segmentation result of data block, and the lobe of the lung point of entire CT images is finally obtained by data back-filling way
Cut result.
Further, in step (a), doctor carries out the mark in five regions to lung, respectively according to Clinical anatomic structure
Are as follows: upper lobe of left lung, lobe of left lung, superior lobe of right lung, middle lobe of right lung and inferior lobe of right lung.
Further, in step (a), the mode of the data prediction is that data are normalized, mode
To carry out window truncation to Hu value, and be normalized to 0 to 1 codomain range, then zoom between -1 to 1；Data are carried out slotting
Value, 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
It counts, the physical picture element interval value on usual three directions is identical, between 0.5 to 1.4 millimeters.
Further, in step (b), in the training stage by obtaining the lobe of the lung tab area in x-axis (horizontal axis), y-axis
(longitudinal axis), the maximum of z-axis (vertical axis), the range of minimum value surround frame as lung's standard 3D, and pass through random offset
Transformation obtains a candidate 3D and surrounds frame.In forecast period, divided by the existing lung based on the full convolutional neural networks of 2D/3D
Model (can also be obtained) by traditional image segmentation algorithm, be partitioned into lung areas, and the lung obtained according to segmentation
Portion region surrounds frame as the lung candidate 3D in x-axis, y-axis, the maximum of z-axis, the range of minimum value.
Further, in step (c), stripping and slicing is carried out along the sagittal plane or coronal-plane of CT images or horizontal plane direction, is obtained
To 3D data block, still, if the video memory of the equipment of training and prediction is enough, also can choose without stripping and slicing.Training stage,
The processing such as constant value filling or cutting redundant data is carried out to the data block to require (it is tall and big to fix length and width to meet data size
It is small), fixed length, width and height size is surrounded frame size by the lung of the data after analysis interpolation and is obtained.Forecast period, block size
Non- sliding window direction without strict demand, but prediction data block is needed comprising all lung areas, and lung is complete when guaranteeing prediction
Whole property, in step (c), stripping and slicing data size can be significantly greater than training stage size to accelerate predetermined speed, if data block
Size is less than fixed size when training, then is carried out constant value and filled to the fixed size, otherwise do not handled.
Further, in step (d), the used convolutional neural networks are full convolution 3D-U type neural network, are led to
Cross concatenate(stacking) mode merges deep layer meaning of one's words feature and shallow-layer local feature.Network inputs are the 3D in step (c)
Data block exports the lobe of the lung segmentation result with input block with size for neural network forecast.
Compared with the prior art, the invention has the following advantages: (1) lobe of the lung dividing method of the invention has better Shandong
Stick, accuracy, as long as and more training datas are provided are trained, the segmentation result of model can be made to become better and better；
(2) by the present invention in that being not necessarily to manual intervention with the full automatic lobe of the lung segmentation of neural fusion；(3) it surrounds in frame and carries out in lung
Segmentation eliminates the interference that lung surrounds outer frame data, and accuracy rate has been turned up and has reduced false sun；(4) present invention is obvious for having
The lung CT data of illness also can preferably Ground Split lobe of the lung region, thus for further quantitative and qualitative assessment pulmonary lesion base is provided
Plinth support.
Detailed description of the invention
It is a kind of flow chart of lobe of the lung dividing method based on full convolutional neural networks described in Fig. 1.
It is the lobe of the lung segmentation result (two dimension view) that the present invention predicts described in Fig. 2.
It is the lobe of the lung segmentation result (3-D view) that the present invention predicts described in Fig. 3.
It is full convolutional neural networks (U-shaped network) of the present invention for lobe of the lung segmentation described in Fig. 4.
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
Describe to one step.Obviously, described embodiments are only a part of the embodiments of the present invention, instead of all the embodiments.Base
Embodiment in the present invention, it is obtained by those of ordinary skill in the art without making creative efforts it is all its
Its embodiment, shall fall within the protection scope of the present invention.
Fig. 1 is the flow chart of lobe of the lung dividing method provided in an embodiment of the present invention.Its key step includes: the building lobe of the lung point
Cut data set；The 3D for obtaining lung organ surrounds frame；The lung 3D data surrounded in frame are pre-processed；Data block is inputted
It is trained into full convolutional neural networks；Data block is input in trained network and is predicted.It is sent out for convenience of understanding
Every details in bright is described in detail by taking training pattern to the prediction segmentation result lobe of the lung as an example.
(1) lobe of the lung partitioned data set, label for labelling and data prediction including five lobe of the lung regions are constructed.The present invention
It needs to carry out Pixel-level to five lobe of the lung regions (upper lobe of left lung, lobe of left lung, superior lobe of right lung, middle lobe of right lung and inferior lobe of right lung)
Mark, notation methods are as follows: veteran doctor is labeled data by 3D slicer software, and will mark
As a result it gives another doctor to check, after doctor's verification confirmation mark is errorless, adopts data as training data, otherwise
It abandons this labeled data or marks again.220 number of cases evidences, and ill agnogenic diffusivity tuberculosis and the number without agnogenic diffusivity tuberculosis are marked altogether
According to ratio be 1:1.
The mode of data prediction is, by the x of data, y, z(, that is, horizontal axis, the longitudinal axis, vertical axis) physical picture element interval interpolation
To 1.4 millimeters, Hu value interception window ranges are [- 1000, -200] and are normalized to 0 to 1 codomain range, then zoom to -1 to 1
Between.
(2) 3D for obtaining lung organ surrounds frame, lung 3D encirclement frame acquisition and forecast period including the training stage
Lung 3D surrounds frame and obtains.In the training stage, lung can be obtained by lobe of the lung mark and surround frame；In forecast period, by existing
Lung parted pattern of some based on the full convolutional neural networks of 3D, is partitioned into lung areas, finally obtains lung and surrounds frame.
(3) the lung 3D data surrounded in frame are pre-processed, is divided into training stage and test phase.In training rank
Section takes lung 3D to surround the data in frame, discards 3D and surrounds the data of outer frame, it is specified that training data block when neural network big
Small is 48*196*256 (being sequentially x, y, z, unit: pixel).Along x-axis (horizontal axis) direction stripping and slicing when stripping and slicing, stripping and slicing with a thickness of
48 pixels, stripping and slicing step-length are 8 pixels, and the stripping and slicing starting point that retracts if stripping and slicing is more than to surround frame boundary makes it not cut out boundary, are obtained
To the data block with a thickness of 48 pixels；Then the y-axis of data block and z-axis are handled as follows, for data block y-axis, if
Data block y-axis length fills constant 0 to data both sides along y-axis less than 196, right if surrounding frame y-axis length and being greater than 196
Random shearing is carried out to data along y-axis, is not handled if surrounding frame y-axis length and being equal to 196, finally makes the y of data block
Shaft length is 196；For data block z-axis, also similarly the processing mode with data block y-axis, final data make the z-axis of data block
Length is 256, finally obtains the data block of 48*196*256 size.
In forecast period, lung is taken to surround the data in frame, discard the data for surrounding outer frame, due to the full convolution mind of 3 dimensions
Through network, so the data of input network can be any dimension.Along the x-axis direction to data progress stripping and slicing in frame is surrounded, cut
For block with a thickness of 48 pixels, stripping and slicing step-length is 16 pixels, and the starting point that retracts if stripping and slicing is more than to surround frame boundary makes it not cut out side
Boundary obtains data block；Then the data of the y-axis of data block and z-axis direction are handled, for y-axis, if data block y-axis
Length then fills constant 0 to data block both sides along y-axis less than 196 pixels, if surrounding frame y-axis length is more than or equal to 196 pictures
It is plain then without operation, finally obtain data block y-axis length more than or equal to 196 pixels；For data z-axis, also carry out similarly with
The data processing method of data block y-axis finally makes data block z-axis length be more than or equal to 256 pixels.
(4) data block is input in full convolutional neural networks and is trained.The full convolution mind of training lobe of the lung parted pattern
It is the 3 U-shaped neural networks of dimension through network, comprising 3 down-sampling layers (maximum pond layer) and 3 up-sampling layers (warp lamination), and
And it is intermediate by stacking connection (concat), each down-sampling layer and up-sampling 2 convolution blocks of layer heel, each convolution block include
3 dimensions convolution (3DConv), batch normalization (Batch normalization), nonlinear activation (ReLU), full convolutional neural networks
The last layer be Softmax activation primitive, network output port number be 6 channels, respectively represent: background, upper lobe of left lung, a left side
6 regions such as lobi inferior, superior lobe of right lung, middle lobe of right lung and inferior lobe of right lung, the optimizer of training network are Adam, initial learning rate
It is 0.001.The output size of network is equal to input size.
When training will the data after (1) to (3) step process be input to above-mentioned 3 tie up it is trained in U-shaped neural networks, when testing
Deconditioning when loss on card collection no longer declines.
(5) data block is input in trained network and is predicted.When carrying out lobe of the lung segmentation to test data, it will count
According to the processing of pretreatment, step (2) and step (3) in (1) through the above steps, data block is obtained, it is finally that data block is defeated
Enter into the full convolutional neural networks of step (4) training and predicted, obtains the segmentation result of each data block.Because predicting number
According to stripping and slicing have overlapping, so the probability of obtained segmentation result overlapping region by be added merged, last image
The classification that the classification of upper each pixel passes through argmax(i.e. 6 channel maximum probability is the classification of current pixel) it obtains.Most
Lobe of the lung segmentation result is obtained by way of backfill eventually.
The careful section of embodiment described above is one of present invention preferably case, not limits the present invention with this and implements model
Enclose, therefore, those skilled in the art various changes and modifications can be made to the invention without departing from spirit of the invention and
Range, these improvements and modifications also should be regarded as protection scope of the present invention.It includes excellent that the following claims are intended to be interpreted as above
It selects embodiment and falls into all change and modification of the scope of the invention.
Claims (6)
1. a kind of lobe of the lung dividing method based on full convolutional neural networks, step include:
(a) construct training dataset: acquisition CT images carry out different classes of mark to lobe of the lung region, are located in advance to data
Reason；
(b) obtain lung and surround frame: in the training stage, according in the step (a) the mask method or lung divide, obtain
The 3D of lung areas surrounds frame in CT images；In forecast period, divides the lung areas using existing lung partitioning algorithm, obtain
The 3D for obtaining lung areas in CT images surrounds frame；
(c) it data stripping and slicing: is surrounded in the lung candidate 3D of the step (b) and carries out data stripping and slicing in frame, the lung is surrounded
Frame is cut into several data blocks, and the data block is filled or is sheared or is not handled to meet data size requirement；
(d) training pattern: being supplied to full convolutional neural networks for the data block in the step (c) and be trained, and obtains
The lobe of the lung parted pattern；
(e) lobe of the lung is divided: data are pre-processed by the described image in the step (a), then by the step (b) with
Step (c) process, obtains multiple data blocks that lung 3D is surrounded in frame, and data block passes through the step (d) lobe of the lung
Parted pattern obtains the lobe of the lung segmentation result of data block, is divided by the lobe of the lung that data back-filling way finally obtains entire CT images
As a result.
2. a kind of lobe of the lung dividing method based on full convolutional neural networks according to claim 1, it is characterised in that: described
In step (a), doctor carries out the mark in five regions to lung, is respectively as follows: under upper lobe of left lung, left lung according to Clinical anatomic structure
Leaf, superior lobe of right lung, middle lobe of right lung and inferior lobe of right lung.
3. a kind of lobe of the lung dividing method based on full convolutional neural networks according to claim 1, it is characterised in that: described
In step (a), the mode of the data prediction is that data are normalized；Interpolation is carried out to data, so that data
D1, d2, d3, and d1, d2 are divided between the physical picture element on x, tri- directions y, z, d3 is the number greater than 0.
4. a kind of lobe of the lung dividing method based on full convolutional neural networks according to claim 1, it is characterised in that: described
In step (b), in the training stage by obtaining the lobe of the lung tab area in x-axis (horizontal axis), y-axis (longitudinal axis), z-axis (vertical axis)
Maximum, the range of minimum value as lung's standard 3D surround frame, and by random offset transformation acquisition one candidate 3D
Surround frame.The lung is partitioned by the existing lung parted pattern based on the full convolutional neural networks of 2D/3D in forecast period
Portion region, and according to the obtained lung areas of segmentation in x-axis, y-axis, the maximum of z-axis, the range of minimum value as the lung
Portion candidate 3D surrounds frame.
5. a kind of lobe of the lung dividing method based on full convolutional neural networks according to claim 1, it is characterised in that: described
In step (c), stripping and slicing is carried out along the sagittal plane or coronal-plane of CT images or horizontal plane direction, obtains 3D data block.Training rank
Section carries out the processing such as constant value filling or cutting redundant data to the data block and requires (fixed length, width and height to meet data size
Size).Forecast period, block size is without strict demand, the step of method as described in claim 1 in (c), stripping and slicing data ruler
The very little training stage size that can be significantly greater than is to accelerate predetermined speed, if the fixation described in when block size is less than training is big
It is small, then it is filled with to the fixed size, does not otherwise handle.
6. a kind of lobe of the lung dividing method based on full convolutional neural networks according to claim 1, it is characterised in that: described
In step (d), the convolutional neural networks used are melted for full convolution 3D-U type neural network by concatenate mode
Close deep layer meaning of one's words feature and shallow-layer local feature.Network inputs are 3D data block described in step (d) described in claim 1,
Output is for neural network forecast with input block with the lobe of the lung segmentation result of size.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811425317.4A CN109636808B (en) | 2018-11-27 | 2018-11-27 | Lung lobe segmentation method based on full convolution neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811425317.4A CN109636808B (en) | 2018-11-27 | 2018-11-27 | Lung lobe segmentation method based on full convolution neural network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109636808A true CN109636808A (en) | 2019-04-16 |
| CN109636808B CN109636808B (en) | 2022-08-12 |
Family
ID=66069348
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811425317.4A Active CN109636808B (en) | 2018-11-27 | 2018-11-27 | Lung lobe segmentation method based on full convolution neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109636808B (en) |
Cited By (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110060262A (en) * | 2019-04-18 | 2019-07-26 | 北京市商汤科技开发有限公司 | A kind of image partition method and device, electronic equipment and storage medium |
| CN110136119A (en) * | 2019-05-16 | 2019-08-16 | 杭州健培科技有限公司 | A kind of lung based on deep learning splits the method and system of segmentation and integrity assessment |
| CN110390261A (en) * | 2019-06-13 | 2019-10-29 | 北京汽车集团有限公司 | Object detection method, device, computer readable storage medium and electronic equipment |
| CN110942453A (en) * | 2019-11-21 | 2020-03-31 | 山东众阳健康科技集团有限公司 | CT image lung lobe identification method based on neural network |
| CN111028248A (en) * | 2019-12-19 | 2020-04-17 | 杭州健培科技有限公司 | Method and device for separating static and dynamic pulses based on CT (computed tomography) image |
| 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 |
| CN111563902A (en) * | 2020-04-23 | 2020-08-21 | 华南理工大学 | Lung lobe segmentation method and system based on three-dimensional convolutional neural network |
| CN112541892A (en) * | 2020-12-07 | 2021-03-23 | 山东众阳健康科技集团有限公司 | Cerebral aneurysm image segmentation method based on 3D activation function |
| CN112598669A (en) * | 2021-03-04 | 2021-04-02 | 之江实验室 | Lung lobe segmentation method based on digital human technology |
| CN114170493A (en) * | 2021-12-02 | 2022-03-11 | 江苏天汇空间信息研究院有限公司 | Method for improving semantic segmentation precision of remote sensing image |
| CN115147359A (en) * | 2022-06-06 | 2022-10-04 | 北京医准智能科技有限公司 | Lung lobe segmentation network model training method and device, electronic equipment and storage medium |
| CN116797596A (en) * | 2023-08-17 | 2023-09-22 | 杭州健培科技有限公司 | Lung segment recognition model and training method for lung nodule |
| US11948305B2 (en) | 2020-06-11 | 2024-04-02 | GE Precision Healthcare LLC | Method and system for segmenting lung image, and storage medium |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106504232A (en) * | 2016-10-14 | 2017-03-15 | 北京网医智捷科技有限公司 | A kind of pulmonary nodule automatic testing method based on 3D convolutional neural networks |
| WO2017210690A1 (en) * | 2016-06-03 | 2017-12-07 | Lu Le | Spatial aggregation of holistically-nested convolutional neural networks for automated organ localization and segmentation in 3d medical scans |
| US20180240235A1 (en) * | 2017-02-23 | 2018-08-23 | Zebra Medical Vision Ltd. | Convolutional neural network for segmentation of medical anatomical images |
| US20180260957A1 (en) * | 2017-03-08 | 2018-09-13 | Siemens Healthcare Gmbh | Automatic Liver Segmentation Using Adversarial Image-to-Image Network |
-
2018
- 2018-11-27 CN CN201811425317.4A patent/CN109636808B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2017210690A1 (en) * | 2016-06-03 | 2017-12-07 | Lu Le | Spatial aggregation of holistically-nested convolutional neural networks for automated organ localization and segmentation in 3d medical scans |
| CN106504232A (en) * | 2016-10-14 | 2017-03-15 | 北京网医智捷科技有限公司 | A kind of pulmonary nodule automatic testing method based on 3D convolutional neural networks |
| US20180240235A1 (en) * | 2017-02-23 | 2018-08-23 | Zebra Medical Vision Ltd. | Convolutional neural network for segmentation of medical anatomical images |
| US20180260957A1 (en) * | 2017-03-08 | 2018-09-13 | Siemens Healthcare Gmbh | Automatic Liver Segmentation Using Adversarial Image-to-Image Network |
Non-Patent Citations (2)
| Title |
|---|
| GONZALEZ，ET AL: "《3D brain tumor segmentation：Narrow unet cnn》", 《IEEE MIT URTC》 * |
| 李正，等: "《一类用于肺结节检测的深度学习加速方法》", 《高校应用数学学报》 * |
Cited By (16)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110060262A (en) * | 2019-04-18 | 2019-07-26 | 北京市商汤科技开发有限公司 | A kind of image partition method and device, electronic equipment and storage medium |
| CN110136119A (en) * | 2019-05-16 | 2019-08-16 | 杭州健培科技有限公司 | A kind of lung based on deep learning splits the method and system of segmentation and integrity assessment |
| CN110390261A (en) * | 2019-06-13 | 2019-10-29 | 北京汽车集团有限公司 | Object detection method, device, computer readable storage medium and electronic equipment |
| CN110942453A (en) * | 2019-11-21 | 2020-03-31 | 山东众阳健康科技集团有限公司 | CT image lung lobe identification method based on neural network |
| CN111028248A (en) * | 2019-12-19 | 2020-04-17 | 杭州健培科技有限公司 | Method and device for separating static and dynamic pulses based on CT (computed tomography) image |
| 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 |
| CN111563902B (en) * | 2020-04-23 | 2022-05-24 | 华南理工大学 | Lung lobe segmentation method and system based on three-dimensional convolutional neural network |
| CN111563902A (en) * | 2020-04-23 | 2020-08-21 | 华南理工大学 | Lung lobe segmentation method and system based on three-dimensional convolutional neural network |
| US11948305B2 (en) | 2020-06-11 | 2024-04-02 | GE Precision Healthcare LLC | Method and system for segmenting lung image, and storage medium |
| CN112541892A (en) * | 2020-12-07 | 2021-03-23 | 山东众阳健康科技集团有限公司 | Cerebral aneurysm image segmentation method based on 3D activation function |
| CN112598669A (en) * | 2021-03-04 | 2021-04-02 | 之江实验室 | Lung lobe segmentation method based on digital human technology |
| CN112598669B (en) * | 2021-03-04 | 2021-06-01 | 之江实验室 | Lung lobe segmentation method based on digital human technology |
| CN114170493A (en) * | 2021-12-02 | 2022-03-11 | 江苏天汇空间信息研究院有限公司 | Method for improving semantic segmentation precision of remote sensing image |
| CN115147359A (en) * | 2022-06-06 | 2022-10-04 | 北京医准智能科技有限公司 | Lung lobe segmentation network model training method and device, electronic equipment and storage medium |
| CN116797596A (en) * | 2023-08-17 | 2023-09-22 | 杭州健培科技有限公司 | Lung segment recognition model and training method for lung nodule |
| CN116797596B (en) * | 2023-08-17 | 2023-11-28 | 杭州健培科技有限公司 | Lung segment recognition model and training method for lung nodule |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109636808B (en) | 2022-08-12 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109636808A (en) | A kind of lobe of the lung dividing method based on full convolutional neural networks | |
| CN109063710B (en) | 3D CNN nasopharyngeal carcinoma segmentation method based on multi-scale feature pyramid | |
| CN109583440B (en) | Medical image auxiliary diagnosis method and system combining image recognition and report editing | |
| CN111798462B (en) | Automatic delineation method of nasopharyngeal carcinoma radiotherapy target area based on CT image | |
| CN105574859B (en) | A kind of liver neoplasm dividing method and device based on CT images | |
| RU2739713C1 (en) | Training annotation of objects in an image | |
| CN111951221B (en) | Glomerular cell image recognition method based on deep neural network | |
| CN109389584A (en) | Multiple dimensioned rhinopharyngeal neoplasm dividing method based on CNN | |
| CN108447062A (en) | A kind of dividing method of the unconventional cell of pathological section based on multiple dimensioned mixing parted pattern | |
| CN109146899A (en) | CT image jeopardizes organ segmentation method and device | |
| CN109872325B (en) | Full-automatic liver tumor segmentation method based on two-way three-dimensional convolutional neural network | |
| CN110310289A (en) | Lung tissue's image partition method based on deep learning | |
| CN110415230A (en) | A kind of CT sectioning image semantic segmentation system and method based on deep learning | |
| CN110652312B (en) | Blood vessel CTA intelligent analysis system and application | |
| CN111047605B (en) | Construction method and segmentation method of vertebra CT segmentation network model | |
| CN106780522A (en) | A kind of bone marrow fluid cell segmentation method based on deep learning | |
| CN115546605A (en) | Training method and device based on image labeling and segmentation model | |
| CN108830149A (en) | A kind of detection method and terminal device of target bacteria | |
| CN107767362A (en) | A kind of early screening of lung cancer device based on deep learning | |
| CN113436173A (en) | Abdomen multi-organ segmentation modeling and segmentation method and system based on edge perception | |
| CN111275712A (en) | Residual semantic network training method oriented to large-scale image data | |
| CN116452618A (en) | Three-input spine CT image segmentation method | |
| Hu et al. | Breastsam: A study of segment anything model for breast tumor detection in ultrasound images | |
| CN114565601A (en) | Improved liver CT image segmentation algorithm based on DeepLabV3+ | |
| CN112884792A (en) | Lung image segmentation method and device, electronic equipment and storage medium |
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 | ||
| CB02 | Change of applicant information |
Address after: 311200 floor 4-6, East Block E, Hangzhou Bay Information Port, Xiaoshan District, Hangzhou City, Zhejiang Province Applicant after: HANGZHOU JIANPEI TECHNOLOGY Co.,Ltd. Address before: 310018 No. 3, No. 3 Street, Xiasha economic and Technological Development Zone, Hangzhou, Zhejiang Applicant before: HANGZHOU JIANPEI TECHNOLOGY Co.,Ltd. |
|
| CB02 | Change of applicant information | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |