CN109685811A - PET/CT hypermetabolism lymph node dividing method based on dual path U-net convolutional neural networks - Google Patents
PET/CT hypermetabolism lymph node dividing method based on dual path U-net convolutional neural networks Download PDFInfo
- Publication number
- CN109685811A CN109685811A CN201811588646.0A CN201811588646A CN109685811A CN 109685811 A CN109685811 A CN 109685811A CN 201811588646 A CN201811588646 A CN 201811588646A CN 109685811 A CN109685811 A CN 109685811A
- Authority
- CN
- China
- Prior art keywords
- pet
- hypermetabolism
- lymph node
- image
- 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
- 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
- G06T2207/10—Image acquisition modality
- G06T2207/10072—Tomographic images
- G06T2207/10104—Positron emission tomography [PET]
-
- 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
- G06T2207/30096—Tumor; Lesion
Abstract
The present invention relates to field of medical image processing, it is desirable to provide the dividing method of hypermetabolism lymph node in a kind of PET/CT image based on dual path U-net convolutional neural networks.The dividing method includes following processes: the acquisition and processing of experimental data；Construct dual path U-net convolutional neural networks；Training dual path U-net convolutional neural networks；Hypermetabolism lymph node in PET/CT image is split using trained dual path U-net convolutional neural networks, obtains segmentation result.The present invention can be very good to split in the hypermetabolism lymph node in PET/CT image by means of dual path U-net convolutional neural networks.
Description
Technical field
The present invention relates to field of medical image processing more particularly to a kind of based on dual path U-net convolutional neural networks
The dividing method of hypermetabolism lymph node in PET/CT image.
Background technique
Lymthoma seriously threatens the health and lives of the people, but some patientss can cure.18F-FDG-PET/CT for
Hodgkin lymphoma (HL) and Diffuse Large B-Cell Lymphoma (DLBCL) are that the diagnosis and treatment performance of the aggressive lymphomas of representative is important
Effect.Lymthoma for whole body infringement based on lymph node involvement, shown as on PET/CT image FDG intake increase, mesh
The preceding hand dipping that image doctor is depended on to the PET/CT image quided of Lymphoma, a large amount of manually with consumption,
Measure the defects of there are subjectivities.Therefore, the place subregion of lymthoma involved lymph nodes, size, extent of metabolism are carried out automatic
Change analysis to be of great significance, but there is no the intelligent evaluation software appearance for lymthoma PET/CT image at present, ours grinds
Study carefully try hard to find a relatively accurate dividing method, make system automatically identified on PET/CT show as FDG intake increase
Involved lymph nodes (i.e. hypermetabolism lymph node) Lymphoma is carried out in favor of its size of later period automatic measurement and SUV value
It automates by stages, to achieve the purpose that exclude the influence of image doctor subjective factor as far as possible, reduce manpower consumption, in favor of clinic
The correct assessment state of an illness, starts Canonical management as early as possible, improves prognosis.
Existing many algorithms are used to divide the hypermetabolism lymph node in PET/CT image at present, mainly include region growing
Method, completion method and convolution method etc..But most lymph node dividing method cannot achieve full automation, it is still necessary to artificial ginseng
With such as region-growing method needs to manually select the seed point of region growing, and completion method needs to manually select initial interested
Region etc..And the mathematic calculations such as convolution are by engineer, the convolution number of plies is limited, for position feature and shape feature without
Method is easily and accurately extracted.2015, Jonathan Long et al. was by full convolutional network (Fully Convolutional
Networks, FCN) it is successfully applied to image segmentation field, and achieve significant effect.The network can be to right in image
Each pixel of vertical position is classified, to realize the task of image segmentation.FCN may be implemented to predict end to end, such as
The gray scale or color image of a 512*512 size are inputted, output is an equal amount of matrix, includes the class to each pixel
Other prediction result.U-net convolutional neural networks were proposed in international bio Medical conference ICMICCAI in 2015, have with
FCN similar network structure becomes one of the major networks for medical image segmentation in recent years.
The structure of U-net as shown in Figure 1, include constricted path and path expander two parts, because network structure shape with
Alphabetical ' U ' is similar, so referred to as U-net.Constricted path includes to shrink module 1~5, wherein shrinking the internal structure of module 1~4
As shown in Fig. 2 (a), each module has an input and two outputs, and input is that initial image to be split or upper layer are shunk
The pondization of module exports, and input obtains convolution output by two layers of convolution operation, and convolution output obtains pond by maximum pond
Output；It include an input and an output shown in the structure such as Fig. 2 (b) for shrinking module 5, input is the pond for shrinking module 4
Change output, operates to obtain deconvolution output by two layers of convolution operation and one layer of deconvolution.Path expander include expansion module 1~
4, wherein each module has 2 inputs shown in internal structure such as Fig. 2 (c) of expansion module 1~3,1 output is inputted to shrink
The deconvolution output of module 5 (or upper layer expansion module) and the corresponding convolution output for shrinking module, two inputs are by fusion, two
Layer convolution operation and one layer of deconvolution operate to obtain deconvolution output；Expand the internal structure and expansion 1~3 base of module of module 4
This is identical, is only free of warp lamination, and input is exported by several layers convolution operation.As shown in Figure 1, expansion module 4
Output is the segmentation result of network output.At present in research field, U-net convolutional neural networks are usually used in medical image point
It cuts, the segmentation including nerve cell, the segmentation of blood vessel.U-net convolutional neural networks achieve not in many segmentation challenge matches
Wrong effect, such as 2012ISBI cell segmentation challenge match.
It is effective for being split using the realization of U-net convolutional neural networks to the hypermetabolism lymph node in PET/CT image
Method, but the feature for needing according to hypermetabolism lymph node is divided in PET/CT image while analyzing two kinds of images of PET and CT, and U-
Only one input channel of net convolutional neural networks cannot achieve while carry out to the hypermetabolism lymph node in PET/CT image
Segmentation.
Summary of the invention
In order to overcome the deficiencies of the prior art, the present invention is directed to propose a kind of improved U-net convolutional neural networks, are realized
The segmentation of hypermetabolism lymph node in PET/CT image is avoided manually dividing existing inefficient and unstable defect, be provided fast
Speed, the method for reliably dividing hypermetabolism lymph node from PET/CT image, to be diagnosis, treatment and the operation of related disease
Guidance provides accurate foundation.For this purpose, the present invention has changed the network structure of U-net, single path input is revised as dual path
Input, specific steps are as follows:
Step 1, the acquisition and processing of experimental data；
Step 2, dual path U-net convolutional neural networks are constructed；
Step 3, training dual path U-net convolutional neural networks；
Step 4, hypermetabolism lymph node in PET/CT image is carried out using trained dual path U-net convolutional neural networks
Segmentation, obtain segmentation result.
The step 1 specifically includes the following steps:
Step 1.1, the cross-section continuous layer images of plain CT and phase of Lymphoma are read simultaneously by Nuclear Medicine Dept doctor
Corresponding PET/CT merges cross-section continuous layer images, and using drawing software, all lymph nodes that FDG intake is higher than background are existed
Profile is sketched the contours of on CT image；
Further, the Nuclear Medicine Dept doctor is intermediate title or more, there is more than 2000 example PET/CT image diagosis experiences
Doctor；
Step 1.2, randomly select parts of images in step 1.1 as training sample, remaining image as test sample,
Data enhancing processing is carried out to training sample；
Further, the method for the data enhancing processing are as follows: using Random-Rotation, scaling, mirror surface treatment, noise is added
The modes such as signal expand training sample, to reduce over-fitting；
Further, each training sample includes CT figure, PET figure and label figure, CT is schemed, PET schemes and the ruler of label figure
It is very little to be consistent；
Further, it is consistent the size of CT figure, PET figure and label figure using bilinear interpolation.
The step 2 specifically includes the following steps:
U-net convolutional neural networks structure is changed, single path input is changed to dual path input；
Specifically: as shown in figure 3, design one and the symmetrical another constricted path of former constricted path, choose wherein one
Input of the paths as CT image, another input as PET image.
More specifically: the overall structure of expanding channel does not change, but the internal structure for expanding module changes, and expands
The input for opening module becomes 3, shown in present input module such as Fig. 2 (d): the shrinking die in the path 1.CT and the path PET by 2
The deconvolution of block 5 (or upper layer expansion module) exports, and the convolution output of module is accordingly shunk in the path 2.CT, and the path 3.PET is corresponding
Shrink the convolution output of module；
Further, structure can also be adjusted according to the feature of image of CT and PET, such as due to PET image constant interval
It is small, therefore BN layers are added in PET constricted path, adjust the quantity etc. of convolutional channel.
The mode of training dual path U-net convolutional neural networks includes training method and loss function in the step 3；
Further, the training method uses Adam optimal way；
Further, the loss function is that weighting intersects entropy function；
The weighting intersects entropy function are as follows:
Wherein, Pi indicates that pixel i belongs to the probability of prospect,The practical generic of sample is represented, N represents sample
Quantity, wclassClass refers to the penalty coefficient of different classes of sample.
The step 4 specifically includes the following steps:
After dual path U-net convolutional neural networks train, any PET/CT image is carried out using trained model
The segmentation of hypermetabolism lymph node, and the test sample in segmentation result and step 1.2 is compared.
Compared with prior art, the beneficial effects of the present invention are: the present invention passes through change U-net convolutional neural networks knot
Single path input is revised as dual path input, proposes a kind of improved U-net convolutional neural networks, realize PET/ by structure
The segmentation of hypermetabolism lymph node in CT image avoids manually dividing existing inefficient and unstable defect, provides one kind
Quickly, reliably divide the method for hypermetabolism lymph node from PET/CT image, to be diagnosis, treatment and the hand of related disease
Art guidance provides accurate foundation.
Detailed description of the invention
Fig. 1 is the schematic network structure of U-net in the present invention.
Fig. 2 is the concrete structure schematic diagram of each module in the network structure of U-net in the present invention.
Fig. 3 is the structural schematic diagram of improved dual path U-net convolutional neural networks in embodiment 2.
Fig. 4 is using trained dual path U-net convolutional neural networks in embodiment 4 to hypermetabolism in PET/CT image
The segmentation result figure that lymph node is split.
Specific embodiment
The present invention is further illustrated in the following with reference to the drawings and specific embodiments.
The acquisition and processing of 1 experimental data of embodiment
(1) more than You Yiming intermediate title, the Nuclear Medicine Dept doctor of more than 2000 example PET/CT image diagosis experiences reads
The cross-section continuous layer images of the plain CT of 10 Lymphoma trunks (from neck root to femur upper segment) and corresponding PET/
CT merges cross-section continuous layer images, and using the drawing software of WIN7 system, all lymph nodes that FDG intake is higher than background are existed
Profile is sketched the contours of on plain CT image.
(2) in step (1) 80% image is randomly selected as training sample, and residual image is as test data.Each
Training sample includes CT figure, PET figure and label figure, and wherein CT figure and label figure be having a size of 512*512, and PET figure having a size of
128*128.Therefore PET is schemed to carry out bilinear interpolation, adjusts size to 512*512, is consistent with CT figure；Mark bitmap-format
For the region corresponding label of involved lymph nodes is 1, other region corresponding labels are 0.Training sample is carried out at data enhancing
Using Random-Rotation, scaling, mirror surface treatment, the modes such as noise signal are added in reason, and 4 times of amplification training sample to original quantity,
To reduce over-fitting.
Embodiment 2 constructs dual path U-net convolutional neural networks
U-net convolutional neural networks structure is changed, single path input is changed to dual path input.
Specifically: as shown in figure 3, design one and the symmetrical another constricted path of former constricted path, choose wherein one
Input of the approach as CT image, another input as PET image.The overall structure of expanding channel does not change,
But the internal structure for expanding module changes, the input for expanding module becomes three by two, expands module in former network
Input such as Fig. 2 (c) shown in: 1. shrink the deconvolution output of module 5 (or upper layer expansion module), the 2. corresponding volumes for shrinking modules
Product output.Shown in present input such as Fig. 2 (d): the contraction module 5 (or upper layer expansion module) in the path 1.CT and the path PET
The convolution output of module is accordingly shunk in deconvolution output, the path 2.CT, and the convolution output of module is accordingly shunk in the path 3.PET.
Embodiment 3 trains dual path U-net convolutional neural networks
Using Adam optimal way as optimization method, while using and intersecting entropy function as loss function in step 1
Neck and clavicle area the PET/CT figure of patient carries out the enhanced picture of data (training sample) and is trained, wherein cross entropy
Function refers to that weighting intersects entropy function；
The weighting intersects entropy function are as follows:
Wherein, PiIndicate that pixel i belongs to the probability of prospect,The practical generic of sample is represented, N represents sample
Quantity, wclassClass refers to the penalty coefficient of different classes of sample.
Embodiment 4 is using trained dual path U-net convolutional neural networks to hypermetabolism lymph node in PET/CT image
It is split
Training is utilized after dual path U-net convolutional neural networks train with the network that test sample test was trained
Model the segmentation of hypermetabolism lymph node, and the segmentation to segmentation result and attending physician are carried out to any one PET/CT image
As a result it is compared.Segmentation result such as Fig. 4 sketches the contours hypermetabolism lymph node with solid white line, and the label in (a) figure is mark
Note, (b) label in figure is the segmentation result of convolutional neural networks, and experimental result is shown: base provided by the present invention
It can be very good to split in hypermetabolism lymph node in PET/CT image in the U-net convolutional neural networks of dual path.
The foregoing is merely illustrative of the preferred embodiments of the present invention, is not intended to limit the invention, all in essence of the invention
Within mind and principle, any modification, equivalent replacement, improvement and so on be should all be included in the protection scope of the present invention.
Claims (10)
1. the dividing method of hypermetabolism lymph node in a kind of PET/CT image based on dual path U-net convolutional neural networks,
It is characterized in that, includes the following steps:
Step 1, the acquisition and processing of experimental data；
Step 2, dual path U-net convolutional neural networks are constructed；
Step 3, training dual path U-net convolutional neural networks；
Step 4, hypermetabolism lymph node in PET/CT image is divided using trained dual path U-net convolutional neural networks
It cuts, obtains segmentation result.
2. the dividing method of hypermetabolism lymph node in PET/CT image according to claim 1, which is characterized in that the step
Rapid 1 detailed process includes:
Step 1.1, at the same read the cross-section continuous layer images of Lymphoma plain CT and corresponding PET/CT fusion it is cross-section
All lymph nodes that FDG intake is higher than background are sketched the contours of profile on CT image using drawing software by continuous layer images；
Step 1.2, the parts of images in step 1.1 is randomly selected as training sample, and remaining image is as test sample, to instruction
Practice sample and carry out data enhancing processing, expands training sample.
3. the dividing method of hypermetabolism lymph node in PET/CT image according to claim 2, which is characterized in that the step
The method that data enhancing processing is carried out to training sample in rapid 1.2 are as follows: using Random-Rotation, scaling, mirror surface treatment, noise is added
Aspect expands training sample, to reduce over-fitting.
4. the dividing method of hypermetabolism lymph node in PET/CT image according to claim 2, which is characterized in that the step
Each training sample in rapid 1.2 includes CT figure, PET figure and label figure, and adjustment CT figure, PET image and the size for marking figure make
Be consistent.
5. the dividing method of hypermetabolism lymph node in PET/CT image according to claim 1, which is characterized in that the step
Rapid 2 detailed process are as follows: single path input is changed to dual path input by the network structure for changing U-net.
6. the dividing method of hypermetabolism lymph node in PET/CT image according to claim 5, which is characterized in that it is described more
Single path input is changed to the operating method of dual path input by the network structure for changing U-net are as follows: design one and former constricted path
Symmetrical another constricted path, chooses wherein input of the paths as CT image, another as the defeated of PET image
Enter.
7. the dividing method of hypermetabolism lymph node in PET/CT image according to claim 1, which is characterized in that the step
The mode of training dual path U-net convolutional neural networks includes training method and loss function in rapid 3.
8. the dividing method of hypermetabolism lymph node in PET/CT image according to claim 7, which is characterized in that the instruction
Practicing method is Adam optimization method, and the loss function is to intersect entropy function.
9. the dividing method of hypermetabolism lymph node in PET/CT image according to claim 8, which is characterized in that the friendship
Pitching entropy function is that weighting intersects entropy function；
The weighting intersects entropy function are as follows:
Wherein PiIndicate that pixel i belongs to the probability of prospect,The practical generic of sample is represented, N represents the quantity of sample,
wclassClass refers to the penalty coefficient of different classes of sample.
10. the dividing method of hypermetabolism lymph node in PET/CT image according to claim 1, which is characterized in that described
The detailed process of step 4 includes: after dual path U-net convolutional neural networks train, using trained model to any
The cross-section image of PET/CT carries out the segmentation of hypermetabolism lymph node.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811588646.0A CN109685811B (en) | 2018-12-24 | 2018-12-24 | PET/CT high-metabolism lymph node segmentation method based on dual-path U-net convolutional neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811588646.0A CN109685811B (en) | 2018-12-24 | 2018-12-24 | PET/CT high-metabolism lymph node segmentation method based on dual-path U-net convolutional neural network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109685811A true CN109685811A (en) | 2019-04-26 |
| CN109685811B CN109685811B (en) | 2019-12-13 |
Family
ID=66189273
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811588646.0A Active CN109685811B (en) | 2018-12-24 | 2018-12-24 | PET/CT high-metabolism lymph node segmentation method based on dual-path U-net convolutional neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109685811B (en) |
Cited By (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110363797A (en) * | 2019-07-15 | 2019-10-22 | 东北大学 | A kind of PET and CT method for registering images inhibited based on excessive deformation |
| CN110555830A (en) * | 2019-08-15 | 2019-12-10 | 浙江工业大学 | Deep neural network skin detection method based on deep Labv3+ |
| CN110619633A (en) * | 2019-09-10 | 2019-12-27 | 武汉科技大学 | Liver image segmentation method based on multi-path filtering strategy |
| CN111062252A (en) * | 2019-11-15 | 2020-04-24 | 浙江大华技术股份有限公司 | Real-time dangerous article semantic segmentation method and device and storage device |
| CN111096767A (en) * | 2020-01-08 | 2020-05-05 | 南京市第一医院 | Deep learning-based mediastinal lymph node ultrasound elastic image segmentation and classification method |
| CN111340825A (en) * | 2020-03-05 | 2020-06-26 | 上海市肺科医院（上海市职业病防治院） | Method and system for generating mediastinal lymph node segmentation model |
| CN111986213A (en) * | 2020-08-21 | 2020-11-24 | 四川大学华西医院 | Processing method, training method and device of slice image and storage medium |
| CN112233058A (en) * | 2019-07-15 | 2021-01-15 | 上海交通大学医学院附属第九人民医院 | Method for detecting lymph nodes in head and neck CT image |
| CN112288708A (en) * | 2020-10-27 | 2021-01-29 | 推想医疗科技股份有限公司 | Method, device, medium, and electronic device for detecting lymph node in CT image |
| CN112349392A (en) * | 2020-11-25 | 2021-02-09 | 北京大学第三医院（北京大学第三临床医学院） | Human cervical vertebra medical image processing system |
| CN112365443A (en) * | 2020-10-16 | 2021-02-12 | 珠海市奥德维科技有限公司 | Hexahedron defect detection method and medium based on deep learning |
| CN112686875A (en) * | 2021-01-04 | 2021-04-20 | 浙江明峰智能医疗科技有限公司 | Tumor prediction method of PET-CT image based on neural network and computer readable storage medium |
| CN113256500A (en) * | 2021-07-02 | 2021-08-13 | 北京大学第三医院（北京大学第三临床医学院） | Deep learning neural network model system for multi-modal image synthesis |
| CN113393443A (en) * | 2021-06-17 | 2021-09-14 | 华南理工大学 | HE pathological image cell nucleus segmentation method and system |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| AU2017101166A4 (en) * | 2017-08-25 | 2017-11-02 | Lai, Haodong MR | A Method For Real-Time Image Style Transfer Based On Conditional Generative Adversarial Networks |
| CN107886081A (en) * | 2017-11-23 | 2018-04-06 | 武汉理工大学 | Two-way U Net deep neural network mine down-holes hazardous act is intelligently classified discrimination method |
-
2018
- 2018-12-24 CN CN201811588646.0A patent/CN109685811B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| AU2017101166A4 (en) * | 2017-08-25 | 2017-11-02 | Lai, Haodong MR | A Method For Real-Time Image Style Transfer Based On Conditional Generative Adversarial Networks |
| CN107886081A (en) * | 2017-11-23 | 2018-04-06 | 武汉理工大学 | Two-way U Net deep neural network mine down-holes hazardous act is intelligently classified discrimination method |
Non-Patent Citations (1)
| Title |
|---|
| 徐峰，等: ""基于U-net的结节分割方法"", 《软件导刊》 * |
Cited By (19)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112233058A (en) * | 2019-07-15 | 2021-01-15 | 上海交通大学医学院附属第九人民医院 | Method for detecting lymph nodes in head and neck CT image |
| CN110363797B (en) * | 2019-07-15 | 2023-02-14 | 东北大学 | PET and CT image registration method based on excessive deformation inhibition |
| CN110363797A (en) * | 2019-07-15 | 2019-10-22 | 东北大学 | A kind of PET and CT method for registering images inhibited based on excessive deformation |
| CN110555830A (en) * | 2019-08-15 | 2019-12-10 | 浙江工业大学 | Deep neural network skin detection method based on deep Labv3+ |
| CN110619633A (en) * | 2019-09-10 | 2019-12-27 | 武汉科技大学 | Liver image segmentation method based on multi-path filtering strategy |
| CN111062252B (en) * | 2019-11-15 | 2023-11-10 | 浙江大华技术股份有限公司 | Real-time dangerous goods semantic segmentation method, device and storage device |
| CN111062252A (en) * | 2019-11-15 | 2020-04-24 | 浙江大华技术股份有限公司 | Real-time dangerous article semantic segmentation method and device and storage device |
| CN111096767A (en) * | 2020-01-08 | 2020-05-05 | 南京市第一医院 | Deep learning-based mediastinal lymph node ultrasound elastic image segmentation and classification method |
| CN111340825A (en) * | 2020-03-05 | 2020-06-26 | 上海市肺科医院（上海市职业病防治院） | Method and system for generating mediastinal lymph node segmentation model |
| CN111340825B (en) * | 2020-03-05 | 2023-05-09 | 上海市肺科医院（上海市职业病防治院） | Method and system for generating mediastinum lymph node segmentation model |
| CN111986213A (en) * | 2020-08-21 | 2020-11-24 | 四川大学华西医院 | Processing method, training method and device of slice image and storage medium |
| CN111986213B (en) * | 2020-08-21 | 2022-12-23 | 四川大学华西医院 | Processing method, training method and device of slice image and storage medium |
| CN112365443A (en) * | 2020-10-16 | 2021-02-12 | 珠海市奥德维科技有限公司 | Hexahedron defect detection method and medium based on deep learning |
| CN112288708A (en) * | 2020-10-27 | 2021-01-29 | 推想医疗科技股份有限公司 | Method, device, medium, and electronic device for detecting lymph node in CT image |
| CN112349392A (en) * | 2020-11-25 | 2021-02-09 | 北京大学第三医院（北京大学第三临床医学院） | Human cervical vertebra medical image processing system |
| CN112686875A (en) * | 2021-01-04 | 2021-04-20 | 浙江明峰智能医疗科技有限公司 | Tumor prediction method of PET-CT image based on neural network and computer readable storage medium |
| CN113393443A (en) * | 2021-06-17 | 2021-09-14 | 华南理工大学 | HE pathological image cell nucleus segmentation method and system |
| CN113393443B (en) * | 2021-06-17 | 2023-02-14 | 华南理工大学 | HE pathological image cell nucleus segmentation method and system |
| CN113256500A (en) * | 2021-07-02 | 2021-08-13 | 北京大学第三医院（北京大学第三临床医学院） | Deep learning neural network model system for multi-modal image synthesis |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109685811B (en) | 2019-12-13 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109685811A (en) | PET/CT hypermetabolism lymph node dividing method based on dual path U-net convolutional neural networks | |
| US10990851B2 (en) | Method and device for performing transformation-based learning on medical image | |
| CN106296699A (en) | Cerebral tumor dividing method based on deep neural network and multi-modal MRI image | |
| CN109146879B (en) | Method and device for detecting bone age | |
| CN107766874B (en) | Measuring method and measuring system for ultrasonic volume biological parameters | |
| CN109214375A (en) | A kind of embryo's pregnancy outcome prediction meanss based on block sampling video features | |
| CN106780434A (en) | Underwater picture visual quality evaluation method | |
| CN110544264A (en) | Temporal bone key anatomical structure small target segmentation method based on 3D deep supervision mechanism | |
| CN109087298B (en) | Alzheimer's disease MRI image classification method | |
| CN109840913A (en) | The method and system of lump segmentation in a kind of mammography X | |
| CN108665454A (en) | A kind of endoscopic image intelligent classification and irregular lesion region detection method | |
| Foth et al. | Macroevolutionary and morphofunctional patterns in theropod skulls: a morphometric approach | |
| JP2022501737A (en) | 3D model generation method, device, device, and storage medium This application is a Chinese patent application filed with the China Patent Office on September 30, 2018, application number 20111160166.4, invention title "Generation of 3D model". Priority is claimed for "methods, devices, devices, and storage media", the entire contents of which are incorporated by reference in this application. | |
| CN110934587A (en) | Alzheimer disease auxiliary diagnosis method based on atlas neural network | |
| CN112232407A (en) | Neural network model training method and device for pathological image sample | |
| CN113724206B (en) | Fundus image blood vessel segmentation method and system based on self-supervision learning | |
| CN110859624A (en) | Brain age deep learning prediction system based on structural magnetic resonance image | |
| CN108629772A (en) | Image processing method and device, computer equipment and computer storage media | |
| CN112613428B (en) | Resnet-3D convolution cattle video target detection method based on balance loss | |
| Spitzer et al. | Parcellation of visual cortex on high-resolution histological brain sections using convolutional neural networks | |
| CN111563903A (en) | MRI whole brain tissue segmentation method based on deep learning | |
| CN112820399A (en) | Method and device for automatically diagnosing benign and malignant thyroid nodules | |
| CN109919216B (en) | Counterlearning method for computer-aided diagnosis of prostate cancer | |
| CN113643297B (en) | Computer-aided age analysis method based on neural network | |
| CN113393456B (en) | Automatic quality control method of early pregnancy fetus standard section based on multiple tasks |
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 |