CN109584324B - Positron Emission Tomography (PET) reconstruction method based on automatic encoder network - Google Patents

Positron Emission Tomography (PET) reconstruction method based on automatic encoder network Download PDF

Info

Publication number
CN109584324B
CN109584324B CN201811247856.3A CN201811247856A CN109584324B CN 109584324 B CN109584324 B CN 109584324B CN 201811247856 A CN201811247856 A CN 201811247856A CN 109584324 B CN109584324 B CN 109584324B
Authority
CN
China
Prior art keywords
image
dae
network
pet
automatic encoder
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.)
Active
Application number
CN201811247856.3A
Other languages
Chinese (zh)
Other versions
CN109584324A (en
Inventor
刘且根
周瑾洁
王宗祥
张明辉
王玉皞
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Nanchang University
Original Assignee
Nanchang University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Nanchang University filed Critical Nanchang University
Priority to CN201811247856.3A priority Critical patent/CN109584324B/en
Publication of CN109584324A publication Critical patent/CN109584324A/en
Application granted granted Critical
Publication of CN109584324B publication Critical patent/CN109584324B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T11/002D [Two Dimensional] image generation
    • G06T11/003Reconstruction from projections, e.g. tomography
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2211/00Image generation
    • G06T2211/40Computed tomography
    • G06T2211/424Iterative

Landscapes

  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Nuclear Medicine (AREA)

Abstract

The invention provides a Positron Emission Tomography (PET) reconstruction method based on an automatic encoder network, which comprises the following steps: step A: establishing a Denoising Automatic Encoder (DAE) network model on the basis of a positron emission computed tomography (PET) image, and acquiring prior information of the image by using a trained Denoising Automatic Encoder (DAE); and B: the invention integrates a Denoising Automatic Encoder (DAE) network based on a Positron Emission Tomography (PET) reconstruction method, adopts the Denoising Automatic Encoder (DAE) network to learn the prior information of the PET image without supervision, then combines the DAE network with the image prior information with the traditional PET iterative reconstruction method, and alternately iterates the two to obtain the reconstructed image, thereby achieving good reconstruction effect.

Description

Positron Emission Tomography (PET) reconstruction method based on automatic encoder network
Technical Field
The invention relates to the technical field of medical image processing, which is mainly used in the fields of denoising, recovering, reconstructing and the like of electronic Computed Tomography (CT), positron emission computed tomography (PET), medical noisy images, and the like, in particular to a positron emission computed tomography (PET) reconstruction method based on an automatic encoder network.
Background
Positron Emission Tomography (PET) is currently the only new imaging technique that can display biomolecular metabolism, receptor and neuromediator activity in vivo, and is an important tool for tumor research and clinical diagnosis and treatment. PET reconstruction is a method of reconstructing a functional image acceptable to a clinician from low-tech-rate and noise-affected projection sinusoidal data. Due to the detector fuzzy effect, positron range and photon nonlinearity, the existing PET reconstructed image has the problems of poor resolution, unclear anatomical structure, inaccurate anatomical positioning, large Poisson noise and the like. At the same time, the limited number of photons in the PET data requires spatial smoothing to reduce noise.
In PET reconstruction, the measurement data y ∈ R M×1 Can be modeled as an independent Poisson randomSet of machine variables by affine transformation
Figure BDA0001839972120000011
And unknown image x ∈ R N×1 The relationship of (a) to (b) is as follows:
Figure BDA0001839972120000012
P∈R M×N is a detection probability matrix, P ij Representing the probability of detecting a photon originating from voxel j by detector i. s is formed by R M×1 And R ∈ R M×1 Respectively, representing scatter coincidence and random coincidence data. M is the number of lines of response (LOR) and N is the number of pixels in image space.
PET reconstruction methods include analytical methods and iterative methods. The analytic methods may be classified into a filtered back projection method (FBP), a back projection filtering method (BFP), a ρ -filter, a convolution back projection method, and the like, according to a specific calculation process. Most representative is a filtered back projection method, an algorithm of which is based on Radon (Radon) transform and fourier slice theorem, although an analysis method of which is fast, the noise immunity is poor, noise in original data is large, data is relatively undersampled, and a satisfactory reconstructed image is difficult to obtain.
Iterative reconstruction algorithms include Maximum Likelihood (MLEM), ordered subset maximum likelihood (OSEM), algebraic Reconstruction (ART), simultaneous Iterative Reconstruction (SIRT), conjugate Gradient (CGM), weighted Least Squares (WLS), maximum A Posteriori (MAP), etc. MLEM algorithms since the inverse problem of solving the emission distribution from the measurement data is an ill-defined problem, an iterative expectation-maximization algorithm that seeks a maximum likelihood solution can result in increased noise once the iteration reaches a certain point. The MAP algorithm, while eliminating divergence at higher iterations, the traditional smoothing priors or total variation previously resulted in excessive smoothing or artifacts in the reconstructed image.
The existing Positron Emission Tomography (PET) reconstruction method based on the automatic encoder network has the problems of high noise, excessive smoothness and artifacts.
How to design a Positron Emission Tomography (PET) reconstruction method based on an automatic encoder network to solve the problems proposed in the background art.
Disclosure of Invention
In view of the above-mentioned problems, the present invention aims to provide a Positron Emission Tomography (PET) reconstruction method based on an automatic encoder network, which is practical, excellent in performance and strong in environmental adaptability.
In order to achieve the purpose, the invention provides the following technical scheme: a Positron Emission Tomography (PET) reconstruction method based on an automatic encoder network comprises the following steps:
step A: establishing a Denoising Automatic Encoder (DAE) network model on the basis of a positron emission computed tomography (PET) image, and acquiring prior information of the image by using a trained Denoising Automatic Encoder (DAE);
PET reconstruction image mathematical model: measurement data y ∈ R M×1 Can be modeled as a set of independent Poisson random variables, transformed by affine
Figure BDA0001839972120000021
And unknown image x ∈ R N×1 The relationship of (a) to (b) is as follows:
Figure BDA0001839972120000022
wherein P ∈ R M×N Is a detection probability matrix, P ij Representing the probability of a photon originating from voxel j being detected by detector i, s ∈ R M×1 And R ∈ R M×1 Respectively representing scatter and random coincidence data, M is the number of lines of response (LOR), N is the number of pixels in image space, and the log-likelihood function is:
Figure BDA0001839972120000023
the maximum likelihood estimate of the unknown image x is:
Figure BDA0001839972120000024
and B, step B: combining a Denoising Automatic Encoder (DAE) network with image prior information with a traditional PET iterative reconstruction method, and alternately iterating the Denoising Automatic Encoder (DAE) network and the traditional PET iterative reconstruction method to obtain a reconstructed image.
Further, the step a specifically includes:
(1) The method comprises the following specific steps of establishing a DAE network model:
the DAE network consists of 20 convolution layers, wherein batch normalization is carried out between the layers except the 1 st layer and the 20 th layer, except the 20 th layer, the activation is carried out by using a linear rectification function (ReLU), the size of a convolution kernel is 3 multiplied by 3, the number of input and output channels is 3 (RGB), and the number of the other layers is 64;
the output of the optimal DAE network is the local mean of the true data density, and the denoised auto-encoder (DAE) error (the difference between the output and the input of the trained auto-encoder) is the average shift vector, the size of this average shift vector is used as the negative log-likelihood of the image prior, for image reconstruction, the probability of using the gradient descent is maximized by back-propagating the auto-encoder error, a ση (I) Representing the DAE network, the input image is I, the output of the DAE network is A ση (I),
Figure BDA0001839972120000031
The DAE is trained by minimizing a cost function:
Figure BDA0001839972120000032
wherein the Gaussian noise η has a variance
Figure BDA0001839972120000033
A ση Indicating DAE as noise variance
Figure BDA0001839972120000034
Training, noise variance
Figure BDA0001839972120000035
And the degraded noise and its variance
Figure BDA0001839972120000036
Irrelevant, it is a specified parameter;
(2) Obtaining prior information of an image by using a DAE network, wherein L (I) = | | | A ση (I)-I|| 2 Measuring image I and local average value A thereof in true data density ση (I) The proximity of (a).
Further, the step B specifically includes:
(1) In the step B, the conventional iterative reconstruction method takes a maximum likelihood method (MLEM) as an example, and the corresponding steps are as follows:
and E: get the desired X by ML, construct alternate iterations:
Figure BDA0001839972120000037
and step M: will find Z ij Carry-in, let the partial derivative equal 0:
Figure BDA0001839972120000038
(2) Combining a Denoising Automatic Encoder (DAE) network with image prior information and an MLEM reconstruction method, and alternately iterating the two methods to obtain a reconstructed image;
Figure BDA0001839972120000041
wherein P ∈ R M×N Is a detection probability matrix, x is an unknown image, and gamma is a relative influence used for weighting the data item and the prior information; the DAE network iteration process is as follows: the first step is as follows: computing data items relative to an image
Figure BDA0001839972120000042
A gradient of (a); the second step is that: calculating the gradient of prior information, and averaging the moving vector | | | A ση (I)-I|| 2 The gradient of (A) is required fromDynamic encoder A ση (I) The gradient of (c), back-propagating the computation through the network; the final step updates the image I using a weighted sum of the two gradient terms.
Compared with the prior art, the invention has the beneficial effects that:
the invention integrates a de-noising automatic encoder (DAE) network on the basis of a positron emission computed tomography (PET) reconstruction method, unsupervised learning of prior information of a PET image is carried out by adopting the de-noising automatic encoder (DAE) network, then the DAE network with the image prior information is combined with a traditional PET iterative reconstruction method, and the two are alternately iterated to obtain a reconstructed image, thereby achieving good reconstruction effect.
Drawings
FIG. 1 is a flow chart of the present invention;
FIG. 2 is a diagram of a neural network framework constructed in accordance with the present invention;
FIG. 3 is a flowchart of a loop through the algorithmic process of the present invention;
FIG. 4 is a diagram of a test apparatus according to the present invention; (a) is a true emission image; (b) is an attenuation map; (c) is the sum of the mask and the true emission image; (d) a noise-free projection data map; (e) a noisy projection data map; (f) reconstructing an image for the FBP;
FIG. 5 is a graph of results of a conventional iterative algorithm; (a) is a result graph of a traditional MLEM algorithm; (b) is a result graph of a traditional OSEM algorithm; (c) is a result graph of the E-ML-INC-EM1 algorithm; (d) is a result graph of the E-ML-INC-EM3 algorithm; (e) is a PL-OS-EMDP algorithm result graph; (f) is a PL with point characteristics algorithm result graph;
FIG. 6 is a comparison of the results of a conventional iterative algorithm and the reconstruction of the present invention; (a) is an original image; (b) reconstructing a result graph for the traditional OSEM algorithm; (c) reconstructing a result graph by using a traditional MLEM algorithm; (d) is a reconstruction result graph of the E-ML-INC-EM-3 algorithm; (E) reconstructing a result graph by using an E-PL-OS-EMDP algorithm; (f) reconstructing a result graph according to the invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention will be described in further detail below with reference to the accompanying drawings and embodiments. The embodiments described herein are only for explaining the technical solution of the present invention and are not limited to the present invention.
The invention provides a technical scheme that: a Positron Emission Tomography (PET) reconstruction method based on an automatic encoder network comprises the following steps:
step A: establishing a Denoising Automatic Encoder (DAE) network model on the basis of a positron emission computed tomography (PET) image, and acquiring prior information of the image by using a trained Denoising Automatic Encoder (DAE);
PET reconstructed image mathematical model: measurement data y ∈ R M×1 Can be modeled as a set of independent Poisson random variables, transformed by affine
Figure BDA0001839972120000051
Unknown image x ∈ R N×1 The relationship of (a) to (b) is as follows:
Figure BDA0001839972120000052
wherein P ∈ R M×N Is a detection probability matrix, P ij Representing the probability of a photon originating from voxel j being detected by detector i, s ∈ R M×1 And R ∈ R M×1 Respectively representing scatter and random coincidence data, M is the number of lines of response (LOR), N is the number of pixels in image space, and the log-likelihood function is:
Figure BDA0001839972120000053
the maximum likelihood estimate of the unknown image x is:
Figure BDA0001839972120000054
and B, step B: combining a Denoising Automatic Encoder (DAE) network with image prior information with a traditional PET iterative reconstruction method, and alternately iterating the Denoising Automatic Encoder (DAE) network and the traditional PET iterative reconstruction method to obtain a reconstructed image.
Further, the step a specifically includes:
(1) The method comprises the following specific steps of establishing a DAE network model:
as shown in fig. 2, the DAE network consists of 20 convolutional layers, with the exception of layer 1 and layer 20, between which batch normalization is performed, with the exception of layer 20, activation using a linear rectification function (ReLU), the size of the convolutional kernel being 3 × 3, the number of input and output channels being 3 (RGB), and the remaining layers being 64;
the output of the optimal DAE network is the local mean of the true data density, and the denoised auto-encoder (DAE) error (the difference between the output and the input of the trained auto-encoder) is the average shift vector, the size of this average shift vector is used as the negative log-likelihood of the image prior, for image reconstruction, the probability of using the gradient descent is maximized by back-propagating the auto-encoder error, a ση (I) Representing the DAE network, the input image is I, the output of the DAE network is A ση (I),
Figure BDA0001839972120000061
The DAE is trained by minimizing a cost function:
Figure BDA0001839972120000062
wherein the Gaussian noise η has a variance
Figure BDA0001839972120000063
A ση Indicating DAE as noise variance
Figure BDA0001839972120000064
Training, noise variance
Figure BDA0001839972120000065
And the variance of the degraded noise
Figure BDA0001839972120000066
Irrelevant, it is a specified parameter;
(2) Obtaining prior information of an image by using a DAE network, wherein L (I) = | | | A ση (I)-I|| 2 Measuring image I and local average value A thereof in true data density ση (I) The proximity of (a).
Further, the step B specifically includes:
(1) In the step B, the conventional iterative reconstruction method takes a maximum likelihood method (MLEM) as an example, and the corresponding steps are as follows:
and E: get the desired X through ML, construct alternate iterations:
Figure BDA0001839972120000067
and step M: will find Z ij Carry-in, let the partial derivative equal 0:
Figure BDA0001839972120000068
(2) Combining a Denoising Automatic Encoder (DAE) network with image prior information and an MLEM reconstruction method, and alternately iterating the two methods to obtain a reconstructed image;
Figure BDA0001839972120000069
wherein P ∈ R MxN Is a detection probability matrix, x is an unknown image, and gamma is a relative influence used for weighting the data item and the prior information; the DAE network iteration process is as follows: the first step is as follows: computing data items relative to an image
Figure BDA0001839972120000071
A gradient of (a); the second step: calculating the gradient of prior information, and averaging the moving vector | | | A ση (I)-I|| 2 The gradient of (2) requires an automatic encoder A ση (I) The gradient of (c), back-propagating the computation through the network; the last step uses two laddersThe weighted sum of the degree terms updates the image I.
Algorithm PSNR SSIM
Traditional-FBP 24.4681 0.6755
Traditional-OSEM 22.6343 0.7961
Traditional-MLEM 24.8795 0.8320
E-ML-INC-EM-1 25.3645 0.8225
E-PL-OS-EMDP 21.8457 0.7873
The invention 30.3197 0.9275
The experimental results in the table show that the two indexes of the peak signal-to-noise ratio (PSNR) and the Structural Similarity (SSIM) of the image reconstructed by the optimized algorithm are higher than those of the image reconstructed by the traditional algorithm, so that a satisfactory effect is achieved.
The foregoing merely represents preferred embodiments of the invention, which are described in some detail and detail, and therefore should not be construed as limiting the scope of the invention. It should be noted that, for those skilled in the art, various changes, modifications and substitutions can be made without departing from the spirit of the present invention, and these are all within the scope of the present invention. Therefore, the protection scope of the present patent should be subject to the appended claims.

Claims (1)

1. A Positron Emission Tomography (PET) reconstruction method based on an automatic encoder network is characterized by comprising the following steps: the method comprises the following steps:
step A: establishing a Denoising Automatic Encoder (DAE) network model on the basis of a Positron Emission Tomography (PET) image, and acquiring prior information of the image by using a trained Denoising Automatic Encoder (DAE);
PET reconstructed image mathematical model: measurement data
Figure 96646DEST_PATH_IMAGE002
Can be modeled as a set of independent Poisson random variables, transformed by affine
Figure 420311DEST_PATH_IMAGE004
And unknown images
Figure 731206DEST_PATH_IMAGE006
The relationship of (a) to (b) is as follows:
Figure DEST_PATH_IMAGE008
wherein,
Figure DEST_PATH_IMAGE010
is a matrix of the probability of detection,
Figure DEST_PATH_IMAGE012
is represented by a detector
Figure DEST_PATH_IMAGE014
Detecting origin of voxels
Figure DEST_PATH_IMAGE016
The probability of a photon of (a) is,
Figure DEST_PATH_IMAGE018
and
Figure DEST_PATH_IMAGE020
respectively representing scatter coincidence and random coincidence data,
Figure DEST_PATH_IMAGE022
is the number of lines of response (LOR),
Figure DEST_PATH_IMAGE024
is the number of pixels in image space, and the log-likelihood function is:
Figure DEST_PATH_IMAGE026
unknown image
Figure DEST_PATH_IMAGE028
The maximum likelihood estimate of (c) is:
Figure DEST_PATH_IMAGE030
and B: combining a Denoising Automatic Encoder (DAE) network with image prior information with a traditional PET iterative reconstruction method, and alternately iterating the Denoising Automatic Encoder (DAE) network and the traditional PET iterative reconstruction method to obtain a reconstructed image;
the step A specifically comprises the following steps:
(1) The method comprises the following specific steps of establishing a DAE network model:
the DAE network consists of 20 convolutional layers, with the exception of layer 1 and layer 20, between which batch normalization is performed, with the exception of layer 20, which is activated using a linear rectification function (ReLU), with the size of the convolutional kernel being
Figure DEST_PATH_IMAGE032
The number of input and output channels is 3 (RGB), and the remaining layers are 64;
the output of the optimal DAE network is the local mean of the true data density and the denoised auto-encoder (DAE) error is the mean shift vector, the size of this mean shift vector is used as the negative log-likelihood of the image prior, for image reconstruction, the probability of using the gradient descent is maximized by back-propagating the auto-encoder error,
Figure DEST_PATH_IMAGE034
representing a DAE network, the input image is I, the output of the DAE network is
Figure DEST_PATH_IMAGE036
Figure DEST_PATH_IMAGE038
Then the DAE is trained by minimizing a cost function:
Figure DEST_PATH_IMAGE040
wherein, gaussian noise
Figure DEST_PATH_IMAGE042
Having a variance
Figure DEST_PATH_IMAGE044
Figure DEST_PATH_IMAGE046
Indicating DAE as noise variance
Figure DEST_PATH_IMAGE048
Training, noise variance
Figure DEST_PATH_IMAGE050
And the variance of the degraded noise
Figure DEST_PATH_IMAGE052
Irrelevant, it is a specified parameter;
(2) Obtaining a priori information about the image using the DAE network
Figure DEST_PATH_IMAGE054
Measuring image I and its local average value in real data density
Figure DEST_PATH_IMAGE056
The proximity of (a);
the step B specifically comprises the following steps:
(1) In the step B, the conventional iterative reconstruction method takes a maximum likelihood method (MLEM) as an example, and the corresponding steps are as follows:
and E: get the desired X through ML, construct alternate iterations:
Figure DEST_PATH_IMAGE058
and step M: will find out
Figure DEST_PATH_IMAGE060
Carry-in, let the partial derivative equal 0:
Figure DEST_PATH_IMAGE062
(2) Combining a Denoising Automatic Encoder (DAE) network with image prior information and an MLEM reconstruction method, and alternately iterating the two methods to obtain a reconstructed image;
Figure DEST_PATH_IMAGE064
wherein,
Figure DEST_PATH_IMAGE066
is a matrix of the probability of detection,
Figure DEST_PATH_IMAGE068
in order to be an unknown image,
Figure DEST_PATH_IMAGE070
is used to weight the relative influence of the data item and the prior information; the DAE network iteration process is as follows: the first step is as follows: computing data items against an image
Figure DEST_PATH_IMAGE072
A gradient of (a); second step, calculating the gradient of prior information and averaging the moving vector
Figure DEST_PATH_IMAGE074
Gradient of (2) requires an automatic encoder
Figure DEST_PATH_IMAGE076
The gradient of (c), back-propagating the computation through the network; the final step updates the image I using a weighted sum of the two gradient terms.
CN201811247856.3A 2018-10-24 2018-10-24 Positron Emission Tomography (PET) reconstruction method based on automatic encoder network Active CN109584324B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201811247856.3A CN109584324B (en) 2018-10-24 2018-10-24 Positron Emission Tomography (PET) reconstruction method based on automatic encoder network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201811247856.3A CN109584324B (en) 2018-10-24 2018-10-24 Positron Emission Tomography (PET) reconstruction method based on automatic encoder network

Publications (2)

Publication Number Publication Date
CN109584324A CN109584324A (en) 2019-04-05
CN109584324B true CN109584324B (en) 2022-12-06

Family

ID=65920560

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201811247856.3A Active CN109584324B (en) 2018-10-24 2018-10-24 Positron Emission Tomography (PET) reconstruction method based on automatic encoder network

Country Status (1)

Country Link
CN (1) CN109584324B (en)

Families Citing this family (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110246094B (en) * 2019-05-13 2021-09-28 南昌大学 6-dimensional embedded denoising self-coding prior information algorithm for color image super-resolution reconstruction
CN111627082B (en) * 2020-05-21 2022-06-21 浙江大学 PET image reconstruction method based on filtering back projection algorithm and neural network
CN111709897B (en) * 2020-06-18 2023-10-24 深圳先进技术研究院 Domain transformation-based positron emission tomography image reconstruction method
WO2021253326A1 (en) * 2020-06-18 2021-12-23 深圳先进技术研究院 Domain transform-based method for reconstructing positron emission tomography image
CN112258414B (en) * 2020-10-26 2023-11-10 北京航星机器制造有限公司 Noise suppression method for security inspection CT ray signals
CN112862915B (en) * 2021-03-02 2024-05-28 明峰医疗系统股份有限公司 PET image reconstruction method, system and computer readable storage medium based on non-training neural network
CN113469915A (en) * 2021-07-08 2021-10-01 深圳高性能医疗器械国家研究院有限公司 PET reconstruction method based on denoising and scoring matching network
CN114332287B (en) * 2022-03-11 2022-07-15 之江实验室 Method, device, equipment and medium for reconstructing PET (positron emission tomography) image based on transformer feature sharing

Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105573121A (en) * 2016-01-18 2016-05-11 南昌大学 Autonomous adjusting deadweight balance mechanism control algorithm for force feedback equipment

Family Cites Families (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CA2495215A1 (en) * 2002-08-12 2004-02-19 Naviscan Pet Systems, Inc. Mission-specific positron emission tomography
CN105678821B (en) * 2016-01-12 2018-08-07 浙江大学 A kind of dynamic PET images method for reconstructing based on self-encoding encoder image co-registration
CN109863512B (en) * 2016-09-01 2023-10-20 通用医疗公司 System and method for automatic transformation by manifold approximation
US10074038B2 (en) * 2016-11-23 2018-09-11 General Electric Company Deep learning medical systems and methods for image reconstruction and quality evaluation
WO2018106805A1 (en) * 2016-12-09 2018-06-14 William Marsh Rice University Signal recovery via deep convolutional networks
US10475214B2 (en) * 2017-04-05 2019-11-12 General Electric Company Tomographic reconstruction based on deep learning

Patent Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105573121A (en) * 2016-01-18 2016-05-11 南昌大学 Autonomous adjusting deadweight balance mechanism control algorithm for force feedback equipment

Also Published As

Publication number Publication date
CN109584324A (en) 2019-04-05

Similar Documents

Publication Publication Date Title
CN109584324B (en) Positron Emission Tomography (PET) reconstruction method based on automatic encoder network
EP2156408B1 (en) Pet local tomography
CN102831627A (en) PET (positron emission tomography) image reconstruction method based on GPU (graphics processing unit) multi-core parallel processing
CN106846427B (en) A kind of limited angle CT method for reconstructing based on the weighting full variation of anisotropy again
CN102184559B (en) Particle filtering-based method of reconstructing static PET (Positron Emission Tomograph) images
CN109712213B (en) PET image reconstruction method, system, readable storage medium and apparatus
Evangelista et al. A new bayesian Poisson denoising algorithm based on nonlocal means and stochastic distances
Liu et al. Singular value decomposition-based 2D image reconstruction for computed tomography
CN111161182A (en) MR structure information constrained non-local mean guided PET image partial volume correction method
Zhang et al. Weighted tensor low-rankness and learnable analysis sparse representation model for texture preserving low-dose CT reconstruction
Huang et al. Gapfill-recon net: a cascade network for simultaneously pet gap filling and image reconstruction
Nuyts On estimating the variance of smoothed MLEM images
WO2011100575A2 (en) Systems, methods and computer readable storage mediums storing instructions for applying multiscale bilateral filtering to magnetic resonance (mr) images
Pfister Tomographic reconstruction with adaptive sparsifying transforms
Lee et al. A modified OSEM algorithm for PET reconstruction using wavelet processing
Kamasak et al. Quality and precision of parametric images created from PET sinogram data by direct reconstruction: proof of concept
Song et al. 4D reconstruction of cardiac SPECT using a robust spatialtemporal prior
Rajith et al. Edge preserved de-noising method for medical x-ray images using wavelet packet transformation
Li et al. Robust frame based X-ray CT reconstruction
Gopi et al. Iterative computed tomography reconstruction from sparse-view data
Boudjelal et al. PDEs on graphs for image reconstruction on positron emission tomography
Gopi et al. Regularization based CT image reconstruction using Algebraic techniques
Bai et al. PET image reconstruction: methodology and quantitative accuracy
US20190206094A1 (en) Fast iterative image reconstruction method for emission tomography
Kaur et al. Complex diffusion regularisation-based low dose 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
GR01 Patent grant
GR01 Patent grant