CN109829918A - A kind of liver image dividing method based on dense feature pyramid network - Google Patents
A kind of liver image dividing method based on dense feature pyramid network Download PDFInfo
- Publication number
- CN109829918A CN109829918A CN201910001654.9A CN201910001654A CN109829918A CN 109829918 A CN109829918 A CN 109829918A CN 201910001654 A CN201910001654 A CN 201910001654A CN 109829918 A CN109829918 A CN 109829918A
- Authority
- CN
- China
- Prior art keywords
- network
- liver
- feature pyramid
- dividing method
- segmentation
- 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
Landscapes
- Apparatus For Radiation Diagnosis (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of liver image dividing methods based on dense feature pyramid network, it is related to technical field of image processing, this method handles the segmentation problem of more phase CT image for liver using dense feature pyramid network, the network is based on full convolution and divides network, introduced feature pyramid, and intensive link enhancement feature stream is used, the liver segmentation of Pixel-level is realized on more phase CT;This method Dice value in public 3DIRCADb database is 95.0%, improves segmentation performance;And confirmed by common data sets and set of clinical data, the smaller CT image of thickness can be seamlessly generalized to the DPFN of the larger CT image training of thickness.
Description
Technical field
The present invention relates to technical field of image processing, particularly relate to a kind of liver figure based on dense feature pyramid network
As dividing method.
Background technique
It is liver diagnosis and the important step that treatment is planned that liver in more phase CT images is divided automatically.Accurate liver point
It is essential for cutting in many clinical applications, such as the diagnosis of liver diseases, functional test and surgery planning.Artificial segmentation
It is a Xiang Fanchong, error-prone and time-consuming work, especially in a large amount of CT data.Therefore, the automatic segmentation of liver is very
It is necessary.However this is a challenging job because in CT image liver anatomical structure is complicated, obscurity boundary,
Form of diverse.
Several dividing methods based on ct images have been proposed in scholar at present.It can be mainly divided into non-machine learning and machine
The method of device study.The method of non-machine learning often relies on the statistical of Hounsfield unit (HU) value in CT data
Cloth, including being based on atlas, being based on active shape model (ASM), based on level set and the method cut based on figure.Such as Wang etc.
People (Wang et al.A new segmentation framework based on sparse shape composition
In liver surgery planning system.Medical Physics.2013,40 (5): 051913.) by sparse shape
Shape composition model is combined with iteratively faster Level Set Method, realizes the synchronous Accurate Segmentation of liver, vena hepatica and tumour.
AlShaikhli et al. (AlShaikhli, Yang, Rosenhahn.Automatic 3D liver segmentation
using sparse representation of global and local image information via level
Set formulation.Computer Science.2015) it develops and a kind of utilizes the sparse of global and local image information
Indicate to come the Level Set Method for dividing 3D liver automatically.Li et al. people (Li et al.Automatic liver segmentation
based on shape con-straints and deformable graph cut in CT images.IEEE Trans-
Actions on Image Processing.2015,24 (12): it 5315.) proposes and a kind of cuts shape constraining involvement figure
The deformation map of region cost and boundary cost is cut.The method of machine learning trains classifier with reality using the feature of manual designs
Existing good segmentation.
In recent years, deep learning is excellent in the various challenging tasks such as classification, segmentation, detection.It is existing
Scholar proposes several liver automatic division methods based on convolutional neural networks.(the Lu et al.Automatic 3d such as Lu
liver location and segmentation via convolutional neural network and graph
Cut.International Journal of Computer Assisted Radiology Surgery.2017,12 (2):
171.) it proposes a 3D FCN and is post-processed with the method that figure is cut.But inventor is in research about Hepatic CT figure
It is found in cutting procedure, current dividing method still has the segmentation mould that segmentation performance is low, and the CT image of different thickness is trained
The problem of popularization cannot be compatible between type.
Summary of the invention
In view of this, it is an object of the invention to propose a kind of liver image segmentation based on dense feature pyramid network
Method, to overcome all or part of deficiency in the prior art.
Based on a kind of above-mentioned purpose liver image dividing method based on dense feature pyramid network provided by the invention,
The dividing method is that building is intensive to connect composition by dividing network, feature pyramid based on full convolution in image procossing
Dense feature pyramid model network, the liver of Pixel-level is then realized on the more phase CT of liver using the prototype network
The method of segmentation.
In some optional embodiments, described that network is divided for end-to-end full convolution based on full convolution segmentation network.
In some optional embodiments, the end-to-end full convolution segmentation network is by an encoder and a decoder group
At, and on the same scale, the feature of the feature of the encoder and the decoder is connected directly.
In some optional embodiments, the encoder is made of multiple convolution blocks, for extracting semantic feature and compressing
Feature Mapping, the convolution block are made of two concatenated convolutional layers and a maximum pond layer.
In some optional embodiments, the decoder is made of multiple warp blocks, and the warp block is anti-by one
Convolutional layer and two concatenated convolutional layer compositions.
In some optional embodiments, the feature pyramid is input module, realizes the down-sampling of different multiples to obtain
Analysis On Multi-scale Features figure, and after being integrated into each layer of network, the feature of different scale is matched with figure, serial operation general
Analysis On Multi-scale Features snap to carries out convolution together, and meets following relationship:
Ci=Concat (D (I), D (Oi-1))
Wherein, Feature Mapping, O I: are originally inputtedi: the output of each convolution block, Ci: the input of each convolution block, Concat:
That one-dimensional attended operation along channel, D are down-sampling operations.
In some optional embodiments, the intensive connection includes the close connection of different stage layer, and every layer closely connects
The output connect meets following relationship:
xl=Hl([D(x0... D (xl-2), xl-1])
Wherein, xl: layer of output, Hl: it is the combination of the operations such as convolution, pond and activation, D: being a kind of matching preceding layer
The scale of output is to xl-1Down-sampling operation.
In some optional embodiments, the dense feature pyramid model network is realized with Tensorflow 1.4,
Network parameter uses Gaussian Profile random initializtion, and weights intersection entropy loss and meet following relationship:
Wherein,Pixel x belongs to the probability of corresponding class, ω i: weight factor, Ci: class, n: sum of all pixels, N:
The quantity of class.
From the above it can be seen that a kind of liver image based on dense feature pyramid network provided by the invention point
Segmentation method, the segmentation problem of more phase CT image for liver is handled using dense feature pyramid network (DPFN), which is based on
Full convolution segmentation network (FCN), introduced feature pyramid, and intensive link enhancement feature stream is used, picture is realized on more phase CT
The liver segmentation of plain grade.This method Dice value in public 3DIRCADb database is 95.0%, improves segmentation performance.
And confirmed by common data sets and set of clinical data, it can seamlessly be promoted with the DPFN of the biggish CT image training of thickness
To the lesser CT image of thickness.
Detailed description of the invention
Fig. 1 is that structure is complicated, obscurity boundary, form of diverse figure for liver anatomical in CT image in the prior art of the invention;
The cross section a-, b- coronal-plane, c- sagittal plane;
Fig. 2 is DPFN Each part schematic diagram in the embodiment of the present invention;
Box-characteristic pattern, thin arrow-down-sampling operation are used, block arrow-convolution sum maximum pond;
Fig. 3 is the architecture diagram of DPFN in the embodiment of the present invention;
Fig. 4 is that the DPFN of the larger CT image training of layer thicknesses of the embodiment of the present invention is generalized to the segmentation of the smaller CT image of thickness
Result figure.
Specific embodiment
To make the objectives, technical solutions, and advantages of the present invention clearer, below in conjunction with specific embodiment, and reference
Attached drawing, the present invention is described in more detail.
In order to solve the more phase CT pictures of liver in the prior art in image partition method there are segmentation performances low, different thickness
The training of CT image parted pattern between the problem of cannot being compatible with popularization, the embodiment of the invention provides one kind based on intensive special
The liver image dividing method of pyramid network is levied, the dividing method is that building is divided by being based on full convolution in image procossing
Network, feature pyramid are cut, then the dense feature pyramid model network of intensive connection composition uses the prototype network
The method of the liver segmentation of Pixel-level is realized on the more phase CT of liver.
In order to realize a kind of liver image segmentation based on dense feature pyramid network provided in the embodiment of the present invention
Method, the specific steps are as follows: DPFN shown in Fig. 3, it is made of three parts: end-to-end FCN, pyramid feature input module
With intensive connection.
1) end-to-end FCN:FCN solves to ask by the hierarchical structure of learning characteristic automatic from the CT image marked
Topic, Fig. 2 (a) show the hierarchical structure that FCN has learnt feature.Model of the invention used for reference U-Net (Ronneberger,
Fischer, Brox 2015) carry out liver segmentation experience.Model is made of a decoder and an encoder, and same
On one scale.The encoder is made of multiple convolution blocks, for extracting semantic feature and compressive features mapping.Each convolution block
It is made of two concatenated convolutional layers and a maximum pond layer.Due to the presence of maximum pond layer, the output of convolution block has not
Same scale.Convolution block is replaced with warp block by decoder, to improve each intermediate resolution ratio exported.Warp block by
One warp lamination and two concatenated convolutional layer compositions.On the same scale, the feature of encoder and the feature of decoder are straight
Connect it is connected, to synthesize finer segmentation result.Generally speaking, the FCN of a depth segmentation is constructed, it can be with end-to-end
Mode training.In view of the Pixel-level of organ size is uneven, weighting is intersected into entropy loss and is defined as
Wherein,Pixel x belongs to the probability of corresponding class, ω i: weight factor, Ci: class, n: sum of all pixels, N:
The quantity of class.
2) feature pyramid input module
By integrating to Analysis On Multi-scale Features, image pyramid input energy effectively improves segmentation performance.Some model difference
Multi-scale image is applied to multiple-limb network, then synthesizes last characteristic pattern in the last layer, in contrast to this, DPFN makes
Analysis On Multi-scale Features figure is obtained with the down-sampling of different multiples, and after being integrated into each layer of network.Smoothly integrate FCN volume
The multiple dimensioned characteristic of code device calculates the hierarchical structure comprising Analysis On Multi-scale Features mapping with spread step 2.Use maximum pond
Layer realizes down-sampling, and constructs multiple dimensioned input on the encoder, carries out down-sampling to feature as Fig. 2 (b) is shown, constructs more rulers
Degree input.
Specifically, Feature Mapping will be originally inputted and be expressed as I, the output of each convolution block is expressed as Oi.Then each convolution
The input C of blockiIs defined as:
Ci=Concat (D (I), D (Oi-1))
Wherein Concat is the attended operation that is one-dimensional along channel, and D is down-sampling operation (in method in the present embodiment
It is maximum pond).Figure is matched the feature of different scale by down-sampling operation, and serial operation snaps to Analysis On Multi-scale Features together
Carry out convolution.Therefore it may only be necessary to which Analysis On Multi-scale Features can be integrated into network by seldom additional parameter.
3) intensive connection
In more phase CNN, by xlIt is expressed as l layers of output, xlIt can be with is defined as:
xl=Hl(xl-1)
Wherein HlIt is the combination of the operations such as convolution, pond and activation.In order to accelerate to restrain, avoids gradient from disappearing, introduce
Residual error study, the connection is by HlResponse and the identity map of upper one layer of feature integrate, transmitted with enhancement information, i.e., it is residual
Poor block.It may be defined as:
xl=Hl(xl-1)+xl-1
However, the output of Liang Ge branch is directly added, the information flow in network can be made to reduce.In order to further improve net
Information flow in network links together the output of preceding layer with subsequent all layers of output using intensive connection.Intensively
Connection expands to the concept that residual error learns ultimate attainment.Specifically, xlIt can be with is defined as:
xl=Hl([x0, x1..., xl-1])
Wherein [x0, x1..., xl-1] refer to the articulamentum 0 ..., l-1 for generating Feature Mapping.Intensive connectivity there is only
Between continuous convolution layer (referred to as intensive block) under same scale.In order to further improve the information between different scale feature
Stream, the close connection that different stage is constructed in the constricted path of FCN are built on different scale characteristic pattern as Fig. 2 (c) is shown
Found intensive connectivity.So xlIs defined as:
xl=Hl([D(x0... D (xl-2), xl-1])
D is a kind of scale of matching preceding layer output to xl-1Down-sampling operation.
DPFN predicts segmentation result, then connected domain analysis (CCA) is carried out, to reject false positive.It is partitioned into
The largest connected domain come is retained, remaining is rejected.
As shown in figure 3, DPFN includes the composition 1 that is connected with 4 layer coders and decoder-path in the embodiment of the present invention
Basic 5 layers of FCN and 4 residual error study.In encoder path, each layer has 2 convolutional layers, followed by 1 maximum pond
Change layer.In decoder-path, each layer has 2 convolutional layers, followed by 1 warp lamination.Intensive connection features pyramid quilt
It applies on encoder path.The step-length of convolution sum deconvolution is 1, and the step-length in maximum pond is 2.Convolution kernel size is 3, instead
Convolution kernel and maximum Chi Huahe size are 2.In order to utilize front and back inter-frame relation, before inputting network, continuous CT image edge
That one-dimensional stacking of channel.
DPFN is realized with Tensorflow 1.4.Network parameter using Gaussian Profile random initializtion (μ=0, σ=
0.01).The Adam optimizer that initial learning rate is 0.0001 is updated for parameter.The size for considering background and liver, will intersect
The weight of entropy loss is respectively set to 1 and 16.
The test that data set and evaluation index are carried out at above-mentioned DPFN is commented on open library access 3DIRCADb database
The DPFN proposed in the embodiment of the present invention is estimated.It is made of the 3D CT image of 10 women and 10 males, 75% patient suffers from
There is liver neoplasm.In order to explore the generalization ability of dividing method, the data set in MICCAI 2017LiTS challenge match is used.
LiTS data set includes that 131 and 70 venous phases enhance three-dimensional abdominal CT scan.The data set is in the different medical treatment in Europe
The heart, plane intrinsic resolution change very big from 0.55mm to 1.0mm, and thickness is from 0.45mm to 6.0mm.In an experiment, target is
Whether experiment can be generalized to the lesser model of thickness using the model of the biggish CT image training of thickness, pick 51 CT figures
Picture, slice thickness are less than 1mm.
In order to carry out data prediction, according to the suggestion of clinician, the HU value of all CT images is clipped to [- 75,
175], to eliminate extraneous tissue.Segmentation performance is measured by commonly using Dice, and this method calculates mark and model prediction manually
As a result the ratio between intersection point and union.Prospect in manual mark is known as A, prediction prospect is B, then defines Dice similarity and refers to
Number:
Here | | indicate the pixel quantity for belonging to prospect in binary segmentation, | A ∩ B | indicate that A and B belongs to prospect jointly
Pixel quantity.For Comprehensive Evaluation, volume aliasing error (VOE) also is used, relative volume difference (RVD), even symmetrical surface
Distance (ASD), the accuracy of root mean square symmetrical surface distance (RMSD) Lai Hengliang segmentation result.For this four evaluation indexes,
It is worth smaller, segmentation result is better, in order to verify the validity and robustness of the dividing method provided in the embodiment of the present invention,
It is tested on 3DIRCADb data set, the data set is as shown in Table 1 and Table 2.
1 3Dircadb liver segmentation quantitative result of table
Table 1 shows most advanced method (Li et al.2018) on liver segmentation performance and 3DIRCADb data set
The comparison of (Christ et al.2017) (Han 2017) (Chlebus et al.2017).As it can be seen that being provided in inventive embodiments
Dividing method better effect is achieved in terms of liver segmentation.Compared by experiment, demonstrates in inventive embodiments and provide
The superiority that is compared with other methods of dividing method.In work from now on, intensive pyramid feature input module will be applied
In three-dimensional network.
It, can be by liver segmentation from the larger CT image of thickness it has also been found that DPFN has good generalization ability in research process
Expand to the smaller CT image of thickness.In order to prove this point, the training DPFN on 3DIRCADb.It is chosen from LiTS data set thick
Degree is less than volume 51 of 1mm as test set, and the results are shown in Table 2.Although the training set of this experiment is volume 60, venous phase only has 20
Volume, but DPFN still obtains 90.97% better result.Fig. 4 is two samples that slice thickness is respectively 0.8mm and 0.7mm
Segmentation result figure.The first behavior original CT image in figure, the second behavior mark by hand, the knot that third behavior DPFN is generated
Fruit.First three columns are respectively cross section, sagittal plane and the coronal-plane that slice thickness is 0.8mm in figure.Three column are respectively slice thick afterwards
Degree is cross section, sagittal plane and the coronal-plane of 0.7mm.Relative to Fig. 1, the boundary of Fig. 4 is more clear.
The quantitative result of the smaller CT image of thickness of the larger CT image training of thickness of table 2
Method | Dice Similarity Index Per Case (%) |
Baseline | 85.01±14.50 |
DPFN | 90.97±4.43 |
In embodiments of the present invention, a kind of liver image dividing method based on dense feature pyramid network, segmentation side
Method is constructed in the image processor by dividing network, feature pyramid, the dense feature of intensive connection composition based on full convolution
Pyramid model network, that is, DPFN introduces in FCN for carrying out the automatic segmentation of liver in the CT data of enhancing contrast
Pyramid feature and intensive connection.This method achieves good effect in common data sets, has stronger extensive energy
Power.The smaller CT image of thickness can be seamlessly generalized to the DPFN of the larger CT image training of thickness.It can also be easily
3D network is expanded to, and is applied in other medical image segmentations.
It should be understood by those ordinary skilled in the art that: the discussion of any of the above embodiment is exemplary only, not
It is intended to imply that the scope of the present disclosure (including claim) is limited to these examples;Under thinking of the invention, above embodiments
Or can also be combined between the technical characteristic in different embodiments, step can be realized with random order, and be existed such as
Many other variations of the upper different aspect of the invention, for simplicity, they are not provided in details.
The embodiment of the present invention be intended to cover fall into all such replacements within the broad range of appended claims,
Modifications and variations.Therefore, all within the spirits and principles of the present invention, any omission, modification, equivalent replacement, the improvement made
Deng should all be included in the protection scope of the present invention.
Claims (8)
1. a kind of liver image dividing method based on dense feature pyramid network, which is characterized in that the dividing method
It is that building intensively connects the dense feature gold word of composition by dividing network, feature pyramid based on full convolution in image procossing
Then tower prototype network realizes the method for the liver segmentation of Pixel-level using the prototype network on the more phase CT of liver.
2. the liver image dividing method according to claim 1 based on dense feature pyramid network, which is characterized in that
It is described that network is divided for end-to-end full convolution based on full convolution segmentation network.
3. the liver image dividing method according to claim 2 based on dense feature pyramid network, which is characterized in that
The end-to-end full convolution segmentation network is made of an encoder and a decoder, and on the same scale, the volume
The feature of code device and the feature of the decoder are connected directly.
4. the liver image dividing method according to claim 3 based on dense feature pyramid network, which is characterized in that
The encoder is made of multiple convolution blocks, and for extracting semantic feature and compressive features mapping, the convolution block is by two grades
Join convolutional layer and a maximum pond layer composition.
5. the liver image dividing method according to claim 3 based on dense feature pyramid network, which is characterized in that
The decoder is made of multiple warp blocks, and the warp block is made of a warp lamination and two concatenated convolutional layers.
6. the liver image dividing method according to claim 1 based on dense feature pyramid network, which is characterized in that
The feature pyramid is input module, realizes the down-sampling of different multiples to obtain Analysis On Multi-scale Features figure, and after being integrated into
In each layer of network, the feature of different scale is matched with figure, Analysis On Multi-scale Features are snapped to and rolled up together by serial operation
Product, and meet following relationship:
Ci=Concat (D (I), D (Oi-1))
Wherein, Feature Mapping, O I: are originally inputtedi: the output of each convolution block, Ci: the input of each convolution block, Concat: along logical
That one-dimensional attended operation of road, D are down-sampling operations.
7. the liver image dividing method according to claim 1 based on dense feature pyramid network, which is characterized in that
The intensive connection includes the close connection of different stage layer, and every layer of close-connected output meets following relationship:
xl=Hl([D(x0... D (xl-2), xl-1])
Wherein, xl: layer of output, Hl: it is the combination of the operations such as convolution, pond and activation, D: being a kind of matching preceding layer output
Scale to xl-1Down-sampling operation.
8. the liver image dividing method according to claim 1 based on dense feature pyramid network, which is characterized in that
The dense feature pyramid model network realizes that network parameter is initial at random using Gaussian Profile with Tensorflow 1.4
Change, and weight intersection entropy loss and meet following relationship:
Wherein,Pixel x belongs to the probability of corresponding class, ω i: weight factor, Ci: class, n: sum of all pixels, N: class
Quantity.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910001654.9A CN109829918B (en) | 2019-01-02 | 2019-01-02 | Liver image segmentation method based on dense feature pyramid network |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910001654.9A CN109829918B (en) | 2019-01-02 | 2019-01-02 | Liver image segmentation method based on dense feature pyramid network |
Publications (2)
Publication Number | Publication Date |
---|---|
CN109829918A true CN109829918A (en) | 2019-05-31 |
CN109829918B CN109829918B (en) | 2022-10-11 |
Family
ID=66860210
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910001654.9A Active CN109829918B (en) | 2019-01-02 | 2019-01-02 | Liver image segmentation method based on dense feature pyramid network |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109829918B (en) |
Cited By (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110310281A (en) * | 2019-07-10 | 2019-10-08 | 重庆邮电大学 | Lung neoplasm detection and dividing method in a kind of Virtual Medical based on Mask-RCNN deep learning |
CN110728238A (en) * | 2019-10-12 | 2020-01-24 | 安徽工程大学 | Personnel re-detection method of fusion type neural network |
CN110728683A (en) * | 2019-09-29 | 2020-01-24 | 吉林大学 | Image semantic segmentation method based on dense connection |
CN110853046A (en) * | 2019-10-12 | 2020-02-28 | 沈阳航空航天大学 | Pancreatic tissue segmentation method based on deep learning |
CN110910408A (en) * | 2019-11-28 | 2020-03-24 | 慧影医疗科技(北京)有限公司 | Image segmentation method and device, electronic equipment and readable storage medium |
CN111145170A (en) * | 2019-12-31 | 2020-05-12 | 电子科技大学 | Medical image segmentation method based on deep learning |
CN111815628A (en) * | 2020-08-24 | 2020-10-23 | 武汉精测电子集团股份有限公司 | Display panel defect detection method, device, equipment and readable storage medium |
CN112767407A (en) * | 2021-02-02 | 2021-05-07 | 南京信息工程大学 | CT image kidney tumor segmentation method based on cascade gating 3DUnet model |
CN112967294A (en) * | 2021-03-11 | 2021-06-15 | 西安智诊智能科技有限公司 | Liver CT image segmentation method and system |
CN115578404A (en) * | 2022-11-14 | 2023-01-06 | 南昌航空大学 | Liver tumor image enhancement and segmentation method based on deep learning |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2018125580A1 (en) * | 2016-12-30 | 2018-07-05 | Konica Minolta Laboratory U.S.A., Inc. | Gland segmentation with deeply-supervised multi-level deconvolution networks |
US20180260957A1 (en) * | 2017-03-08 | 2018-09-13 | Siemens Healthcare Gmbh | Automatic Liver Segmentation Using Adversarial Image-to-Image Network |
CN109035197A (en) * | 2018-05-31 | 2018-12-18 | 东南大学 | CT contrastographic picture tumor of kidney dividing method and system based on Three dimensional convolution neural network |
CN109063710A (en) * | 2018-08-09 | 2018-12-21 | 成都信息工程大学 | Based on the pyramidal 3D CNN nasopharyngeal carcinoma dividing method of Analysis On Multi-scale Features |
-
2019
- 2019-01-02 CN CN201910001654.9A patent/CN109829918B/en active Active
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2018125580A1 (en) * | 2016-12-30 | 2018-07-05 | Konica Minolta Laboratory U.S.A., Inc. | Gland segmentation with deeply-supervised multi-level deconvolution networks |
US20180260957A1 (en) * | 2017-03-08 | 2018-09-13 | Siemens Healthcare Gmbh | Automatic Liver Segmentation Using Adversarial Image-to-Image Network |
CN109035197A (en) * | 2018-05-31 | 2018-12-18 | 东南大学 | CT contrastographic picture tumor of kidney dividing method and system based on Three dimensional convolution neural network |
CN109063710A (en) * | 2018-08-09 | 2018-12-21 | 成都信息工程大学 | Based on the pyramidal 3D CNN nasopharyngeal carcinoma dividing method of Analysis On Multi-scale Features |
Non-Patent Citations (2)
Title |
---|
刘哲等: "结合改进的U-Net和Morphsnakes的肝脏分割", 《中国图象图形学报》 * |
孙明建等: "基于新型深度全卷积网络的肝脏CT影像三维区域自动分割", 《中国生物医学工程学报》 * |
Cited By (14)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110310281A (en) * | 2019-07-10 | 2019-10-08 | 重庆邮电大学 | Lung neoplasm detection and dividing method in a kind of Virtual Medical based on Mask-RCNN deep learning |
CN110310281B (en) * | 2019-07-10 | 2023-03-03 | 重庆邮电大学 | Mask-RCNN deep learning-based pulmonary nodule detection and segmentation method in virtual medical treatment |
CN110728683A (en) * | 2019-09-29 | 2020-01-24 | 吉林大学 | Image semantic segmentation method based on dense connection |
CN110728238A (en) * | 2019-10-12 | 2020-01-24 | 安徽工程大学 | Personnel re-detection method of fusion type neural network |
CN110853046A (en) * | 2019-10-12 | 2020-02-28 | 沈阳航空航天大学 | Pancreatic tissue segmentation method based on deep learning |
CN110910408A (en) * | 2019-11-28 | 2020-03-24 | 慧影医疗科技(北京)有限公司 | Image segmentation method and device, electronic equipment and readable storage medium |
CN111145170B (en) * | 2019-12-31 | 2022-04-22 | 电子科技大学 | Medical image segmentation method based on deep learning |
CN111145170A (en) * | 2019-12-31 | 2020-05-12 | 电子科技大学 | Medical image segmentation method based on deep learning |
CN111815628A (en) * | 2020-08-24 | 2020-10-23 | 武汉精测电子集团股份有限公司 | Display panel defect detection method, device, equipment and readable storage medium |
CN112767407A (en) * | 2021-02-02 | 2021-05-07 | 南京信息工程大学 | CT image kidney tumor segmentation method based on cascade gating 3DUnet model |
CN112767407B (en) * | 2021-02-02 | 2023-07-07 | 南京信息工程大学 | CT image kidney tumor segmentation method based on cascade gating 3DUnet model |
CN112967294A (en) * | 2021-03-11 | 2021-06-15 | 西安智诊智能科技有限公司 | Liver CT image segmentation method and system |
CN115578404A (en) * | 2022-11-14 | 2023-01-06 | 南昌航空大学 | Liver tumor image enhancement and segmentation method based on deep learning |
CN115578404B (en) * | 2022-11-14 | 2023-03-31 | 南昌航空大学 | Liver tumor image enhancement and segmentation method based on deep learning |
Also Published As
Publication number | Publication date |
---|---|
CN109829918B (en) | 2022-10-11 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109829918A (en) | A kind of liver image dividing method based on dense feature pyramid network | |
CN110111313B (en) | Medical image detection method based on deep learning and related equipment | |
CN109410185B (en) | A kind of image partition method, device and storage medium | |
Liu et al. | CVM-Cervix: A hybrid cervical Pap-smear image classification framework using CNN, visual transformer and multilayer perceptron | |
Wang et al. | Automatic brain tumor segmentation based on cascaded convolutional neural networks with uncertainty estimation | |
CN112465827B (en) | Contour perception multi-organ segmentation network construction method based on class-by-class convolution operation | |
CN111091573B (en) | CT image pulmonary vessel segmentation method and system based on deep learning | |
CN109523521A (en) | Lung neoplasm classification and lesion localization method and system based on more slice CT images | |
CN107977952A (en) | Medical image cutting method and device | |
CN110060235A (en) | A kind of thyroid nodule ultrasonic image division method based on deep learning | |
CN110084318A (en) | A kind of image-recognizing method of combination convolutional neural networks and gradient boosted tree | |
CN109635812B (en) | The example dividing method and device of image | |
CN110047080A (en) | A method of the multi-modal brain tumor image fine segmentation based on V-Net | |
CN111179237A (en) | Image segmentation method and device for liver and liver tumor | |
CN110288611A (en) | Coronary vessel segmentation method based on attention mechanism and full convolutional neural networks | |
CN115578404A (en) | Liver tumor image enhancement and segmentation method based on deep learning | |
Yang et al. | Semantic segmentation in architectural floor plans for detecting walls and doors | |
Huang et al. | Automatic Retinal Vessel Segmentation Based on an Improved U‐Net Approach | |
Abbasi-Sureshjani et al. | Boosted exudate segmentation in retinal images using residual nets | |
Vang et al. | SynergyNet: a fusion framework for multiple sclerosis brain MRI segmentation with local refinement | |
Lu et al. | Regional perception and multi-scale feature fusion network for cardiac segmentation | |
Wang et al. | Multi-scale hierarchical transformer structure for 3d medical image segmentation | |
Jia et al. | Amo-net: abdominal multi-organ segmentation in mri with a extend unet | |
Liu et al. | Right Ventricle Segmentation of Cine MRI Using Residual U-net Convolutinal Networks | |
Xu et al. | Multi-scale context-aware networks for quantitative assessment of colorectal liver metastases |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |