Method and system for reconstruction of CEST contrast image

Patent metadata

Field	Value
Publication number	US-11587270-B2
Application number	US-202017100114-A
Country	US
Kind code	B2
Filing date	Nov 20, 2020
Priority date	Nov 26, 2019
Publication date	Feb 21, 2023
Grant date	Feb 21, 2023

Abstract

Disclosed is a method for reconstruction of a Chemical Exchange Saturation Transfer (CEST) contrast image. The method includes: generating training samples for a deep neural network; training the deep neural network with the training samples to obtain a trained deep neural network; and reconstructing a CEST contrast image by using the trained deep neural network and PROPELLER undersampled CEST images. The method for reconstruction of a CEST contrast image can effectively shorten the experimental time of a CEST contrast imaging and can obtain a smoother and more accurate CEST contrast image. Further disclosed is a system for reconstruction of a CEST contrast image to implement the method for reconstruction.

First claim

What is claimed is:

1. A method for reconstruction of a Chemical Exchange Saturation Transfer (CEST) contrast image, comprising:
S 1 : generating training samples for a deep neural network, which specifically comprises the following steps:
S 101 : obtaining a simulated region;
S 102 : randomly generating in the simulated region a geometric figure that serves to simulate a part of an imaging object;
S 103 : setting parameters σ k , ω k , A k and δ of a Lorentzian line shape model of CEST effect in the geometric figure separately to obtain a geometric figure with parameter σ k , a geometric figure with parameter ω k , a geometric figure with parameter A k , and a geometric figure with parameter δ;
S 104 : adding filtered texture values and noise to the geometric figure with parameter σ k , the geometric figure with parameter ω k , the geometric figure with parameter A k , and the geometric figure with parameter δ, wherein texture values are used to simulate a texture of the imaging object, and noise is used to simulate a noise during magnetic resonance (MR) sampling;
S 105 : repeating S 102 to S 104 until random geometric figures overlay a whole simulated region, to obtain σ k , ω k , A k and δ parameter maps in the Lorentzian line shape model of CEST effect, respectively;
S 106 : obtaining a combination of ω values in CEST imaging, wherein the combination of ω values is composed of a plurality of ω values, comprising a relatively large value corresponding to an unsaturated CEST image which is used for signal normalization;
S 107 : generating CEST images for each ω value by using the σ k , ω k , A k and δ parameter maps in the Lorentzian line shape model of CEST effect;
S 108 : performing PROPELLER undersampling on the CEST images;
S 109 : combining the PROPELLER undersampled CEST images with an A k parameter map into a training sample; and
S 110 : repeating S 101 to S 109 to generate a preset quantity of training samples;
S 2 : training the deep neural network by using the training samples to obtain a trained deep neural network; and
S 3 : reconstructing a CEST contrast image by using the trained deep neural network and PROPELLER undersampled CEST images, which specifically comprises the following steps:
acquiring CEST image data from an actual imaging object with a magnetic resonance imaging (MRI) sequence to obtain PROPELLER undersampled k-space data at saturation frequency offsets corresponding to each ω value in the combination of ω values;
performing Fourier transformation on the PROPELLER undersampled k-space data of the actual imaging object to obtain PROPELLER undersampled CEST images corresponding to each ω value; and
obtaining a CEST contrast image with the trained deep neural network and the PROPELLER undersampled CEST images.

2. The method for reconstruction of a CEST contrast image according to claim 1 , wherein step S 107 specifically comprises: obtaining CEST images based on ω values and the σ k , ω k , A k and δ parameter maps by a formula of the Lorentzian line shape model of CEST effect.

3. The method for reconstruction of a CEST contrast image according to claim 1 , wherein step S 108 specifically comprises:
performing Fourier transformation on the CEST images corresponding to each ω value, to obtain corresponding k-space data;
wherein the k-space data of the CEST image corresponding to an ω value corresponds to a PROPELLER sampling blade of a particular sampling trajectory; each PROPELLER sampling blade collects several k-space lines of the corresponding CEST image; if there are M ω values, M sampling blades are present correspondingly; the angle of the first sampling blade is 0, and the difference between the angles of adjacent sampling blades is 180°/M, that is, 180°/M is taken as an increment for the angle of sampling blade; the angle of the sampling blade corresponding to the m th ω value is 180° (m−1)/M, and the sampling trajectory of M sampling blades spans a circle in the k-space.

4. A system for reconstruction of a CEST contrast image, comprising:
a training sample generating module configured to generate training samples for a deep neural network;
a network training module configured to train the deep neural network by using the training samples to obtain a trained deep neural network; and
an image reconstruction module configured to reconstruct a CEST contrast image by using the trained deep neural network and PROPELLER undersampled CEST images;
wherein the training sample generating module specifically comprises:
a simulated region obtaining unit configured to obtain a simulated region;
a geometric figure generating unit configured to randomly generate in the simulated region a geometric figure that serves to simulate a part of an imaging object;
a parameter setting unit configured to set parameters σ k , ω k , A k and δ of the Lorentzian line shape model of CEST effect in the geometric figure to obtain a geometric figure with parameter σ k , a geometric figure with parameter ω k , a geometric figure with parameter A k , and a geometric figure with parameter δ;
a texture and noise unit configured to add filtered texture values and noise to the geometric figure with parameter σ k , the geometric figure with parameter ω k , the geometric figure with parameter A k , and the geometric figure with parameter δ, respectively, wherein the texture values are used to simulate the texture of the imaging object, and the noise is used to simulate the noise during MR sampling;
a parameter map generating unit configured to overlay the whole simulated region with the random geometric figures, to obtain σ k , ω k , A k and δ parameter maps in the Lorentzian line shape model of CEST effect, respectively;
an ω value obtaining unit configured to obtain a combination of ω values in CEST imaging, wherein the combination of ω values is composed of a plurality of ω values, comprising a relatively large value corresponding to an unsaturated CEST image which is used for signal normalization;
a simulated CEST image generating unit configured to generate CEST images for each ω value by using the σ k , ω k , A k and δ parameter maps in the Lorentzian line shape model of CEST effect;
a PROPELLER undersampling unit configured to perform PROPELLER undersampling on the CEST images;
a single training sample combining unit configured to combine the PROPELLER undersampled CEST images with the A k parameter map into a training sample; and
a training sample generating unit configured to generate a preset quantity of training samples.

5. The system for reconstruction of a CEST contrast image according to claim 4 , wherein the image reconstruction module specifically comprises:
an actual data acquisition unit configured to acquire CEST image data from an actual imaging object with an MRI sequence to obtain PROPELLER undersampled k-space data at saturation frequency offsets corresponding to each ω value in the combination of ω values;
an actual CEST image determining unit configured to perform Fourier transformation on the PROPELLER undersampled k-space data of the actual imaging object to obtain PROPELLER undersampled CEST images corresponding to each ω value; and
a CEST contrast image determining unit configured to obtain a CEST contrast image from the PROPELLER undersampled CEST images with the trained deep neural network.

6. The system for reconstruction of a CEST contrast image according to claim 4 , wherein the simulated CEST image generating unit obtains CEST images based on ω values and the σ k , ω k , A k and δ parameter maps by a formula of the Lorentzian line shape model of CEST effect.

7. The system for reconstruction of a CEST contrast image according to claim 4 , wherein the PROPELLER undersampling unit performs Fourier

Assignees

• Univ Xiamen

Inventors

• Cai Shuhui
• Guo Chenlu
• Cai Congbo
• Wu Jian

Classifications

• G06T12/20

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

• G06T11/10

  Texturing; Colouring; Generation of textures or colours (retouching, inpainting or scratch removal G06T5/77) · CPC title

• G06T12/00 Primary

  Tomographic reconstruction from projections · CPC title

• G06N3/0464

  Convolutional networks [CNN, ConvNet] · CPC title

• G06N3/09

  Supervised learning · CPC title