CN110288671A - The low dosage CBCT image rebuilding method of network is generated based on three-dimensional antagonism - Google Patents
The low dosage CBCT image rebuilding method of network is generated based on three-dimensional antagonism Download PDFInfo
- Publication number
- CN110288671A CN110288671A CN201910559108.7A CN201910559108A CN110288671A CN 110288671 A CN110288671 A CN 110288671A CN 201910559108 A CN201910559108 A CN 201910559108A CN 110288671 A CN110288671 A CN 110288671A
- Authority
- CN
- China
- Prior art keywords
- image
- antagonism
- dimensional
- sinusoidal
- projection
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Withdrawn
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T11/00—2D [Two Dimensional] image generation
- G06T11/003—Reconstruction from projections, e.g. tomography
- G06T11/006—Inverse problem, transformation from projection-space into object-space, e.g. transform methods, back-projection, algebraic methods
Landscapes
- Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Algebra (AREA)
- Mathematical Analysis (AREA)
- Mathematical Optimization (AREA)
- Mathematical Physics (AREA)
- Pure & Applied Mathematics (AREA)
- Image Processing (AREA)
- Apparatus For Radiation Diagnosis (AREA)
Abstract
The invention discloses a kind of low dosage CBCT image rebuilding methods that network is generated based on three-dimensional antagonism, belong to technical field of medical image processing.Described method includes following steps: constructing three-dimensional antagonism and generates network model；Network model is generated by sinusoidal image and its three-dimensional antagonism of corresponding data for projection training；Test image is input to trained antagonism and generates network model, the sinusoidal image lack part of incomplete projections is predicted, the sinusoidal image of whole projection is obtained；According to the sinusoidal image CT image reconstruction of the complete projection data.The present invention effectively shortens the acquisition time of cone beam projection data, improves clinical diagnosis efficiency.
Description
Technical field
The present invention relates to technical field of medical image processing, and in particular to generates network based on three-dimensional antagonism to a kind of
Low dosage CBCT image rebuilding method.
Background technique
Pencil-beam computed tomography (CBCT) is a kind of medical imaging that can quickly directly obtain three-dimensional CT image
Technology.But shown by correlative study: primary complete CBCT scanning is usually along with the ionising radiation of higher degree, and high agent
Amount ionising radiation can induce the diseases such as human metabolism's exception or even cancer, leukaemia.Therefore, how to make in reduction X-ray
While with dosage, guarantee the emphasis that reconstructed image quality meets clinical diagnosis requirement as field of medical image processing research.
Clinically reducing one of the important method of sufferer amount of radiation is exactly to reduce CBCT scanning range, i.e., by the rotation of detector
Gyration scope limitation is in the section that some is less than standard, so that X-ray radiation suffered by patient be greatly reduced on the whole
Amount.Although the dose of radiation that patient is subject to is greatly reduced, there is a large amount of star strip artifacts and noise, serious shadow in reconstruction image
The resolution to characteristic point is rung, so that it cannot meet the needs of clinical diagnosis.
To improve the CT image rebuild, method commonly used in the prior art is divided into two major classes: method and base based on projection
In the method for image.Method based on projection is that the projection of missing is estimated before CT image reconstruction, predicts the projection of missing most
Direct method is directional algorihtm, and another method based on projection is image Moment Methods, establishes image moment with it and projects square
Between relationship to estimate the unknown projection from known projection.Disadvantage is that: tripleplane's data can not be handled, it is real
It is general to test effect.
Summary of the invention
In view of the deficiencies of the prior art, the low of network is generated based on three-dimensional antagonism the purpose of the present invention is to provide a kind of
Dosage CBCT image rebuilding method, solve CT image reconstruction existing in the prior art is of poor quality, be unable to satisfy clinical diagnosis need
The technical issues of wanting.
In order to solve the above technical problems, the technical scheme adopted by the invention is that:
A kind of low dosage CBCT image rebuilding method generating network based on three-dimensional antagonism, the method includes walking as follows
It is rapid:
It constructs three-dimensional antagonism and generates network model；
Network model is generated by sinusoidal image and its three-dimensional antagonism of corresponding data for projection training；
Test image is input to trained antagonism and generates network model, to the sinogram of incomplete projections
As lack part is predicted, the sinusoidal image of whole projection is obtained；
According to the sinusoidal image CT image reconstruction of the complete projection data.
Further, it includes: encoder, generator and arbiter that the antagonism, which generates network model,；The encoder
For encoding high-order feature；The generator is used to decode high-order feature and obtains the sinusoidal image of generation；
The arbiter is used to identify the similitude of the sinusoidal image of the sinusoidal image and complete data for projection that generate.
Further, the encoder uses the context of the Three dimensional convolution network derived from from AlexNet framework to compile
Code device.
Further, the encoder, generator and arbiter include Three dimensional convolution layer, the linear elementary layer of amendment, Chi Hua
Layer, batch normalization layer and full articulamentum.
Further, the training method of the antagonism generation network model includes:
Sinusoidal image and its corresponding complete data for projection and incomplete projections are obtained, by incomplete projections
Sinusoidal image is divided into training image and test image；
The high-order feature of the training image is extracted by encoder；
By generator by high-order Feature Mapping be generate sinusoidal image
Measure the similitude of the sinusoidal image of the generation and the sinogram of complete data for projection by arbiter, and according to
Similitude advanced optimizes encoder, generator and arbiter；
It repeats the above steps, until being more than preset model frequency of training.
Further, the similitude is measured by associated losses function；The associated losses function includes rebuilding damage
Antagonism of becoming estranged loss.
Further, the reconstruction loss are as follows:
The antagonism loses A (I) is defined as:
The associated losses function J (I) are as follows:
J (I)=λrR(I)+λaA (I) (4),
Wherein, M is the binary mask of the absent region of I, and G (I) is the output of the generator of input data I, and D is antagonism
Discrimination model, E1For desired value, λaFor antagonism loss, λrTo rebuild loss.
Further, loss is lost and fought come Optimized Coding Based device, generator, arbiter using comprising rebuilding.
Compared with prior art, the beneficial effects obtained by the present invention are as follows being:
The present invention reconstructs the CT image for meeting clinical diagnosis requirement, high quality under the conditions of incomplete projections, i.e.,
Under the premise of guaranteeing reconstructed image quality, x-ray radiation suffered by patient is efficiently reduced, effectively shortens cone-beam projections number
According to acquisition time, improve clinical diagnosis efficiency.
Detailed description of the invention
Fig. 1 is the flow chart of the method for the present invention；
Fig. 2 is the structural schematic diagram that three-dimensional antagonism generates network model in the present invention.
Specific embodiment
The method of the present invention is divided into training stage and test phase, step include: firstly, using complete data for projection sine
Image and the sinusoidal image of incomplete projections train antagonism to generate network model, and high quality sine can be generated in acquisition
The three-dimensional antagonism of image generates network model model；Secondly, using the model trained to incomplete projections sinogram
Lack part is predicted, the sinusoidal image of the whole projection of generation is obtained；Finally, using FDK method from the complete throwing of generation
The sinusoidal image of shadow data reconstructs CT image.The present invention can predict missing projection data and further reconstruct to meet clinic
It diagnoses, the CT image of high quality.
The invention will be further described below in conjunction with the accompanying drawings.Following embodiment is only used for clearly illustrating the present invention
Technical solution, and not intended to limit the protection scope of the present invention.
As shown in Figure 1, being the flow chart of the method for the present invention, include the following steps:
Step 1: it constructs three-dimensional antagonism and generates network model；
As shown in Fig. 2, being the structural schematic diagram that three-dimensional antagonism generates network model in the present invention, three-dimensional antagonism is generated
Network model model includes: encoder, generator and arbiter, and encoder, generator and arbiter mainly include: Three dimensional convolution
Layer, ReLU layers (correcting linear elementary layer), pond layer, Normalization layers of Batch (batch normalization layer) and Full
Connect layers (full articulamentum), convolutional layer step-length is preferred: 3*3*3, and pond layer is using maximum pond.
Step 101: encoding high-order feature using encoder；
Our encoder derives from AlexNet framework.For input picture, we are using preceding 5 convolution figure layers under
The pond figure layer in face indicates to calculate abstract high-dimensional feature.Compared with AlexNet, our model does not pass through
The training of ImageNet classification；On the contrary, network is trained for the context-prediction of random initializtion weight from the beginning.
Step 102: the high-order feature that code coder obtains is solved using generator.
" encoder feature " is fully connected layer using channel and is connected to " decoder characteristic ".Channel is after being fully connected layer
A series of upper convolutional layer of five with learning filters, each activates function with rectification linear unit (ReLU).Upwards
Convolution is the convolution of the image of a generation higher resolution.It is construed as the sampling after convolution or Fractional-step
The convolution of width, until obtaining the sinusoidal image generated.
Step 103: identifying the similitude of the sinusoidal image of the sinusoidal image and complete data for projection that generate using arbiter；
Using the sinusoidal image of the sinusoidal image of the generation of generator and complete data for projection as the input of arbiter；Differentiate
Device, so that the loss function of the sinusoidal image and the sinusoidal image of complete data for projection generated becomes larger, can to the greatest extent may be used by training
The sinusoidal image of sinusoidal image and complete data for projection that the identification of energy generates.
Step 104: using the loss function that arbiter obtains come Reverse optimization generator；
Generator constantly reduces the damage for generating image and true picture by training according to the loss function of arbiter
Function is lost, so that the sinusoidal image generated is similar to the sinusoidal image of complete data for projection as far as possible.
Step 2: it is fought using the sinusoidal image of complete data for projection and the sinusoidal image of incomplete projections to train
Property generate network model；
Training antagonism generates network model, and the specific method is as follows:
Step 201: obtaining the complete data for projection and incomplete projections of three-dimensional CBCT sine image, not by a part
For the sinusoidal image of complete data for projection as training image, the sinusoidal image of complete data for projection, will be another as label image
The sinusoidal image of part incomplete projections is as test image.
Step 202: the high-order feature of training image is extracted using the context coding device based on convolutional neural networks；
In convolution or upper convolutional layer the size and number of filter depend on inputting sinusoidal image size and they
Lack part.For example, if the size of the sinusoidal image of tripleplane and its lack part is respectively 4*360*256 and 4*90*
256, then filter size is 4*4*4,4*4*4,4*4*4,4*4*4,4*4*4, and the quantity of filter is 64,64,128,256,
512 are used for five convolutional layers.Finally obtain 4*4*4*512 high-order feature, the input as generator.
Step 203: encoder being extracted into high-order feature as the input data of generator, high-order character is mapped as
The sinusoidal image of generation；
The target of GAN (generating confrontation network) is to search out one to distinguish really and generate data with maximum possible
Arbiter, while the generation image of generator network being promoted to be more nearly true picture.The target of GAN is maximization or minimum
Change binary system cross entropy:
LGAN=log (D (x))+log (1-D (G (z))) (1),
In formula: D (x) indicates differentiation of the arbiter network to training input picture；Z indicates multidimensional hidden variable feature, G (z)
Indicate that generator network generates image when input is multidimensional hidden variable feature, x indicates input picture.
Step 204: using the sinusoidal image of complete data for projection and the sinusoidal image of generation as the input parameter of arbiter,
The similitude of above-mentioned two image is measured using associated losses function and advanced optimizes encoder, generator, arbiter；
In training, we use associated losses function, and associated losses function includes rebuilding loss and antagonism loss.Make
With the L2 distance of label, we can lose R (I) for rebuilding is defined as:
Wherein M is the binary mask of the absent region of I；G (I) is the output of the generator of input data I.In formula (1)
Reconstruction loss the advantages of be by minimize mean pixel deflection error come approximate target profile well.Nevertheless,
Loss existing defects in terms of capturing high frequency detail are rebuild, this can make prediction result seem fuzzy.Therefore it needs that confrontation is added
Property loss to capture high frequency detail so that image seems to be more clear.
Antagonism loses A (I) is defined as:
D is antagonism discrimination model.In order to optimize generation and discrimination model, training process uses Adam algorithm, formula (3)
In antagonism loss occur when can reduce in training only using the reconstruction loss in formula (2) it is fuzzy.
Then associated losses function J (I) is defined as:
J (I)=λrR(I)+λaA(I) (4)
Wherein, M is the binary mask of the absent region of I, and G (I) is the output of the generator of input data I, and D is antagonism
Discrimination model, E1For desired value, λaFor antagonism loss, λrTo rebuild loss.
It is the respective weights for rebuilding loss and antagonism loss that it, which is neutralized,.Due to the associated losses function knot in formula (4)
The advantages of having closed reconstruction loss and antagonism loss, therefore can further improve prediction
Step 3: being input to trained antagonism for the sinusoidal image of incomplete projections and generate network model,
The sinusoidal image lack part of incomplete projections is predicted, the sinusoidal image of the whole projection of generation is obtained；
These experiments are carried out on the data set being made of 400 CT images, are obtained by filter projection complete
The size of three-dimensional CBCT data for projection is 360*256*256, it means that planar detector is the size of 256*256 and projection is
It is sampled from 360 scanning views of 360 degree of arcs.The data that we input are Incomplete projection three-dimensional datas, lack 0 herein
The three-dimensional CBCT data for projection of 89 scanning angle, therefore what three-dimensional CBCT data for projection was exported having a size of 270*256*4, decoder
It is the prediction to missing projection data, the data for projection predicted is having a size of 90*256*4.
Step 4: the sinusoidal image of the complete projection data of above-mentioned generation is further obtained, is calculated by FDK three-dimensional reconstruction
Method, according to the sinusoidal image CT image reconstruction of the complete projection data of generation.
The above is only a preferred embodiment of the present invention, it is noted that for the ordinary skill people of the art
For member, without departing from the technical principles of the invention, several improvement and deformations can also be made, these improvement and deformations
Also it should be regarded as protection scope of the present invention.
Claims (7)
1. a kind of low dosage CBCT image rebuilding method for generating network based on three-dimensional antagonism, which is characterized in that the method
Include the following steps:
It constructs three-dimensional antagonism and generates network model；
Network model is generated by sinusoidal image and its three-dimensional antagonism of corresponding data for projection training；
Test image is input to trained antagonism and generates network model, the sinusoidal image of incomplete projections is lacked
It loses part to be predicted, obtains the sinusoidal image of whole projection；
According to the sinusoidal image CT image reconstruction of the whole projection.
2. a kind of low dosage CBCT image rebuilding method that network is generated based on three-dimensional antagonism according to claim 1,
It is characterized in that, it includes: encoder, generator and arbiter that the antagonism, which generates network model,；The encoder is for compiling
Code high-order feature；The generator is used to decode high-order feature and obtains the sinusoidal image of generation；The arbiter is for identifying
The similitude of the sinusoidal image of the sinusoidal image and complete data for projection of generation.
3. a kind of low dosage CBCT image rebuilding method that network is generated based on three-dimensional antagonism according to claim 2,
It is characterized in that, the encoder uses the context coding device of the Three dimensional convolution network derived from from AlexNet framework.
4. a kind of low dosage CBCT image rebuilding method that network is generated based on three-dimensional antagonism according to claim 2,
It is characterized in that, the encoder, generator and arbiter include Three dimensional convolution layer, the linear elementary layer of amendment, pond layer, criticize and return
One changes layer and full articulamentum.
5. a kind of low dosage CBCT image rebuilding method that network is generated based on three-dimensional antagonism according to claim 1,
It is characterized in that, the training method that the antagonism generates network model includes:
Sinusoidal image and its corresponding complete data for projection and incomplete projections are obtained, by the sine of incomplete projections
Image is divided into training image and test image；
The high-order feature of the training image is extracted by encoder；
By generator by high-order Feature Mapping be generate sinusoidal image
The similitude of the sinusoidal image of the generation and the sinogram of complete data for projection is measured by arbiter, and according to similar
Property advanced optimizes encoder, generator and arbiter；
It repeats the above steps, until being more than preset model frequency of training.
6. a kind of low dosage CBCT image rebuilding method that network is generated based on three-dimensional antagonism according to claim 5,
It is characterized in that, the similitude is measured by associated losses function；The associated losses function include rebuild loss and it is right
Resistance loss.
7. a kind of low dosage CBCT image rebuilding method that network is generated based on three-dimensional antagonism according to claim 6,
It is characterized in that, the reconstruction loss are as follows:
The antagonism loses A (I) are as follows:
The associated losses function J (I) are as follows:
J (I)=λrR(I)+λaA (I) (4),
Wherein, M is the binary mask of the absent region of I, and G (I) is the output of the generator of input data I, and D is that antagonism differentiates
Model, EIFor desired value, λaWeight, λ are lost for antagonismrTo rebuild loss.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910559108.7A CN110288671A (en) | 2019-06-25 | 2019-06-25 | The low dosage CBCT image rebuilding method of network is generated based on three-dimensional antagonism |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910559108.7A CN110288671A (en) | 2019-06-25 | 2019-06-25 | The low dosage CBCT image rebuilding method of network is generated based on three-dimensional antagonism |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110288671A true CN110288671A (en) | 2019-09-27 |
Family
ID=68005894
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910559108.7A Withdrawn CN110288671A (en) | 2019-06-25 | 2019-06-25 | The low dosage CBCT image rebuilding method of network is generated based on three-dimensional antagonism |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110288671A (en) |
Cited By (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110728729A (en) * | 2019-09-29 | 2020-01-24 | 天津大学 | Unsupervised CT projection domain data recovery method based on attention mechanism |
| CN111260583A (en) * | 2020-01-17 | 2020-06-09 | 南京邮电大学 | Multi-discriminant-based multi-analysis network missing CT projection data estimation method |
| CN111429379A (en) * | 2020-04-01 | 2020-07-17 | 山东大学 | Low-dose CT image denoising method and system based on self-supervision learning |
| CN111445447A (en) * | 2020-03-16 | 2020-07-24 | 东软医疗系统股份有限公司 | CT image anomaly detection method and device |
| CN111489404A (en) * | 2020-03-20 | 2020-08-04 | 深圳先进技术研究院 | Image reconstruction method, image processing device and device with storage function |
| CN111968110A (en) * | 2020-09-02 | 2020-11-20 | 广州海兆印丰信息科技有限公司 | CT imaging method, device, storage medium and computer equipment |
| CN112489156A (en) * | 2020-12-17 | 2021-03-12 | 电子科技大学 | Low-dose CT reconstruction method based on interpolation convolutional neural network |
| CN112508808A (en) * | 2020-11-26 | 2021-03-16 | 中国人民解放军战略支援部队信息工程大学 | CT (computed tomography) dual-domain joint metal artifact correction method based on generation countermeasure network |
| WO2021120069A1 (en) * | 2019-12-18 | 2021-06-24 | 深圳先进技术研究院 | Low-dose image reconstruction method and system on basis of a priori differences between anatomical structures |
| WO2021184389A1 (en) * | 2020-03-20 | 2021-09-23 | 深圳先进技术研究院 | Image reconstruction method, image processing device, and device with storage function |
| CN113610711A (en) * | 2021-08-02 | 2021-11-05 | 南京信息工程大学 | Single-image-guided three-dimensional surface reconstruction method and device |
| CN114241074A (en) * | 2021-12-20 | 2022-03-25 | 四川大学 | CBCT image reconstruction method for deep learning and electronic noise simulation |
| CN116977473A (en) * | 2023-09-21 | 2023-10-31 | 北京理工大学 | Sparse angle CT reconstruction method and device based on projection domain and image domain |
| CN117152365A (en) * | 2023-10-31 | 2023-12-01 | 中日友好医院(中日友好临床医学研究所) | Method, system and device for oral cavity CBCT ultra-low dose imaging |
-
2019
- 2019-06-25 CN CN201910559108.7A patent/CN110288671A/en not_active Withdrawn
Cited By (22)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110728729A (en) * | 2019-09-29 | 2020-01-24 | 天津大学 | Unsupervised CT projection domain data recovery method based on attention mechanism |
| CN110728729B (en) * | 2019-09-29 | 2023-05-26 | 天津大学 | Attention mechanism-based unsupervised CT projection domain data recovery method |
| WO2021120069A1 (en) * | 2019-12-18 | 2021-06-24 | 深圳先进技术研究院 | Low-dose image reconstruction method and system on basis of a priori differences between anatomical structures |
| CN111260583A (en) * | 2020-01-17 | 2020-06-09 | 南京邮电大学 | Multi-discriminant-based multi-analysis network missing CT projection data estimation method |
| CN111445447A (en) * | 2020-03-16 | 2020-07-24 | 东软医疗系统股份有限公司 | CT image anomaly detection method and device |
| CN111445447B (en) * | 2020-03-16 | 2024-03-01 | 东软医疗系统股份有限公司 | CT image anomaly detection method and device |
| CN111489404A (en) * | 2020-03-20 | 2020-08-04 | 深圳先进技术研究院 | Image reconstruction method, image processing device and device with storage function |
| CN111489404B (en) * | 2020-03-20 | 2023-09-05 | 深圳先进技术研究院 | Image reconstruction method, image processing device and device with storage function |
| WO2021184389A1 (en) * | 2020-03-20 | 2021-09-23 | 深圳先进技术研究院 | Image reconstruction method, image processing device, and device with storage function |
| CN111429379B (en) * | 2020-04-01 | 2022-06-10 | 山东大学 | Low-dose CT image denoising method and system based on self-supervision learning |
| CN111429379A (en) * | 2020-04-01 | 2020-07-17 | 山东大学 | Low-dose CT image denoising method and system based on self-supervision learning |
| CN111968110A (en) * | 2020-09-02 | 2020-11-20 | 广州海兆印丰信息科技有限公司 | CT imaging method, device, storage medium and computer equipment |
| CN111968110B (en) * | 2020-09-02 | 2023-12-22 | 广州海兆印丰信息科技有限公司 | CT imaging method, device, storage medium and computer equipment |
| CN112508808A (en) * | 2020-11-26 | 2021-03-16 | 中国人民解放军战略支援部队信息工程大学 | CT (computed tomography) dual-domain joint metal artifact correction method based on generation countermeasure network |
| CN112489156B (en) * | 2020-12-17 | 2023-04-18 | 电子科技大学 | Low-dose CT reconstruction method based on interpolation convolutional neural network |
| CN112489156A (en) * | 2020-12-17 | 2021-03-12 | 电子科技大学 | Low-dose CT reconstruction method based on interpolation convolutional neural network |
| CN113610711A (en) * | 2021-08-02 | 2021-11-05 | 南京信息工程大学 | Single-image-guided three-dimensional surface reconstruction method and device |
| CN114241074A (en) * | 2021-12-20 | 2022-03-25 | 四川大学 | CBCT image reconstruction method for deep learning and electronic noise simulation |
| CN116977473A (en) * | 2023-09-21 | 2023-10-31 | 北京理工大学 | Sparse angle CT reconstruction method and device based on projection domain and image domain |
| CN116977473B (en) * | 2023-09-21 | 2024-01-26 | 北京理工大学 | Sparse angle CT reconstruction method and device based on projection domain and image domain |
| CN117152365A (en) * | 2023-10-31 | 2023-12-01 | 中日友好医院(中日友好临床医学研究所) | Method, system and device for oral cavity CBCT ultra-low dose imaging |
| CN117152365B (en) * | 2023-10-31 | 2024-02-02 | 中日友好医院(中日友好临床医学研究所) | Method, system and device for oral cavity CBCT ultra-low dose imaging |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110288671A (en) | The low dosage CBCT image rebuilding method of network is generated based on three-dimensional antagonism | |
| CN109146988A (en) | Non-fully projection CT image rebuilding method based on VAEGAN | |
| WO2020156195A1 (en) | Ct image generation method and apparatus, computer device and computer-readable storage medium | |
| CN102715906B (en) | Method and system for 3D cardiac motion estimation from single scan of c-arm angiography | |
| CN107203989A (en) | End-to-end chest CT image dividing method based on full convolutional neural networks | |
| Lee et al. | CT image conversion among different reconstruction kernels without a sinogram by using a convolutional neural network | |
| CN112348936B (en) | Low-dose cone-beam CT image reconstruction method based on deep learning | |
| US20130195341A1 (en) | Method for sorting ct image slices and method for constructing 3d ct image | |
| CN112262440A (en) | Method and system for judging cancer treatment response through image omics characteristics | |
| CN109300136A (en) | It is a kind of to jeopardize organs automatic segmentation method based on convolutional neural networks | |
| CN110458859A (en) | A kind of segmenting system of the myelomatosis multiplex stove based on multisequencing MRI | |
| Dai et al. | Limited-view cone-beam CT reconstruction based on an adversarial autoencoder network with joint loss | |
| CN109903356A (en) | Missing CT data for projection estimation method based on the multiple parsing network of depth | |
| TW202141517A (en) | Method for establishing three-dimensional medical image in which a convolutional neural network is used to generate multiple intermediate images between two adjacent ones of X-ray images of a to-be-inspected subject in order to establish a three-dimensional medical image of the subject | |
| Dong et al. | A deep unsupervised learning framework for the 4D CBCT artifact correction | |
| Yang et al. | ImplantFormer: vision transformer-based implant position regression using dental CBCT data | |
| Cong et al. | Monochromatic image reconstruction via machine learning | |
| Liang et al. | Shading correction for volumetric CT using deep convolutional neural network and adaptive filter | |
| CN111179373B (en) | Medical image bone removal model construction method and bone information removal method | |
| Liu et al. | Low-dose CBCT reconstruction via 3D dictionary learning | |
| CN116630738A (en) | Energy spectrum CT imaging method based on depth convolution sparse representation reconstruction network | |
| CN105976412A (en) | Offline-dictionary-sparse-regularization-based CT image reconstruction method in state of low tube current intensity scanning | |
| KR20200021156A (en) | High Quality Four Dimensional Cone Beam Computerized Tomography System Using Prior Image | |
| CN116075273A (en) | Multiple substance decomposition for spectral computed tomography | |
| Deep Learning Based CT Image Reconstruction |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| WW01 | Invention patent application withdrawn after publication | ||
| WW01 | Invention patent application withdrawn after publication |
Application publication date: 20190927 |