Image processing method using convolutional neural network, image processing device and storage medium

Patent metadata

Field	Value
Publication number	US-10984565-B2
Application number	US-201816235882-A
Country	US
Kind code	B2
Filing date	Dec 28, 2018
Priority date	Dec 29, 2017
Publication date	Apr 20, 2021
Grant date	Apr 20, 2021

Abstract

An image processing method, comprising: acquiring, by a CT scanning system, projection data of an object; and processing, by using a convolutional neural network, the projection data, to acquire an estimated image of the object. The convolutional neural network comprises: a projection domain network for processing input projection data to obtain estimated projection data; an analytical reconstruction network layer for performing analytical reconstruction to obtain a reconstructed image; an image domain network for processing the reconstructed image to obtain an estimated image, a projection layer for performing a projection operation by using a system projection matrix of the CT scanning system, to obtain a projection result of the estimated image; and a statistical model layer for determining consistency among the input projection data, the estimated projection data, and the projection result of the estimated image based on a statistical model.

First claim

We claim:

1. An image processing method, comprising:
acquiring, by a Computerized-Tomography (CT) scanning system, projection data of an object; and
processing, by using a convolutional neural network, the projection data, to acquire an estimated image of the object;
wherein the convolutional neural network comprises:
a projection domain network for processing input projection data to obtain estimated projection data;
an analytical reconstruction network layer for performing analytical reconstruction on the estimated projection data to obtain a reconstructed image;
an image domain network for processing the reconstructed image to obtain an estimated image,
a projection layer for performing a projection operation on the estimated image by using a system projection matrix of the CT scanning system, to obtain a projection result of the estimated image; and
a statistical model layer for determining consistency among the input projection data, the estimated projection data, and the projection result of the estimated image based on a statistical model;
wherein the image processing method comprises training the convolutional neural network by:
adjusting parameters of convolutional kernels of the image domain network and the projection domain network by using a consistency cost function of a data model based on the input projection data, the estimated projection data, and the projection result of the estimated image.

2. The method according to claim 1 , wherein training the neural network further comprises:
constructing a cost function consistent with the projection using the projection layer, constructing a likelihood relation cost function using the statistical model layer, and forming the consistency cost function of the data model using at least one of the cost function consistent with the projection and the likelihood relation cost function.

3. The method according to claim 1 , wherein the convolutional neural network further comprises at least one priori model layer for adjusting the image domain network by using a priori model cost function based on the estimated image, and performing back propagation of a gradient through the analytical reconstruction network layer to adjust parameters of a convolutional kernel of the projection domain network.

4. The method according to claim 1 , wherein a forward propagation process of the projection domain network, the analytical reconstruction network layer and the image domain network comprise:
expressing input projection data of the projection domain network as g={g 1 , g 2 , . . . , g M }, expressing estimated projection data output by the projection domain network as {tilde over (g)}={{tilde over (g)} 1 , {tilde over (g)} 2 , . . . , {tilde over (g)} M′ }, wherein M′≥M, after the estimated projection data is weighted, obtaining Diag(W){tilde over (g)}={W 1 {tilde over (g)} 1 , W 2 {tilde over (g)} 2 , . . . , W M′ }, after the weighted projection data passes through a ramp filtering layer, obtaining h⊗Diag(W){tilde over (g)}, after the filtered data is back-projected, obtaining an output of the analytical reconstruction layer as {tilde over (f)}=H R T h⊗Diag(W){tilde over (g)}, and assuming that φ N represents a processing function of the image domain network, obtaining the estimated image output by the image domain network as {circumflex over (f)}=φ N ({tilde over (f)}),
wherein a superscript T represents transposition of a matrix, h is a discrete ramp filtering operator, H R is a system matrix for M′×N dimensional reconstruction, N is a total number of pixels of the reconstructed image, and W 1 , W 2 , . . . , W M represent weighting coefficients.

5. The method according to claim 4 , wherein the consistency cost function of the data model is expressed as Ψ({tilde over (g)}; H{circumflex over (f)}, g)=L(g; {tilde over (g)})+β∥{tilde over (g)}−H{circumflex over (f)}∥ 2 , and error transfer relations from the consistency of the data model are

∂
Ψ

∂

g
~

=

∂

L
⁡

(

g
;

g
~

)

∂

g
~

+

2
⁢

β
⁡

(

g
~

-

H
⁢

f
^

)

⁢

⁢
and
⁢

⁢

∂
Ψ

∂

f
^

=

2
⁢
β
⁢

⁢

H
T

⁡

(

H
⁢

f
^

-

g
~

)

,

wherein L(g; {tilde over (g)}) is a likelihood relation cost function, the smaller the L(g; {tilde over (g)}) becomes, the more consistent the pro

Assignees

• Univ Tsinghua
• Nuctech Co Ltd

Inventors

• Xing Yuxiang
• Liang Kaichao
• Shen Le
• Zhang Li
• Yang Hongkai
• Kang Kejun
• Chen Zhiqiang
• Li Jianmin
• Liu Yinong

Classifications

• G06T12/00 Primary

  Tomographic reconstruction from projections · CPC title

• G06T12/20 Primary

  Inverse problem, i.e. transformations from projection space into object space · CPC title

• G06T12/10

  Image preprocessing, e.g. calibration, positioning of sources or scatter correction · CPC title

• G06T12/30

  Image post-processing, e.g. metal artefact correction · CPC title

• A61B6/03 Primary

  Computed tomography [CT] · CPC title