# US 12,761,303 B2 · Imaging method and device using a mobile magnetic resonance apparatus, storage medium, and terminal

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 12,761,303 B2
- Document ID: US12761303B2
- Status: Granted
- Kind: B2
- Application number: 18/443,443
- Filed: 2024-02-16
- Published: 2024-08-22
- Granted: 2026-10-13
- Earliest claimed: 2023-02-22
- Assignees: Beijing Chao-Yang Hospital, Capital Medical University
- Applicants: Beijing Chao-Yang Hospital, Capital Medical University
- Inventors: Qi Yang; Yueyan Bian; Xiuqin Jia; Xiaoxu Yang
- Web page: https://patentgrep.com/patent/US12761303B2

## CPC

- G01R 33/5608: {Data processing and visualization specially adapted for MR, e.g. for feature analysis and pattern recognition on the basis of measured MR data, segmentation of measured MR data, edge contour detection on the basis of measured MR data, for enhancing measured MR data in terms of signal-to-noise ratio by means of noise filtering or apodization, for enhancing measured MR data in terms of resolution by means for deblurring, windowing, zero filling, or generation of gray-scaled images, colour-coded images or images displaying vectors instead of pixels (image data processing or generation, in general G06T)}
- G01R 33/0047: {Housings or packaging of magnetic sensors (packaging of semiconductor devices H10W99/00); Holders}
- G01R 33/4818: {MR characterised by data acquisition along a specific k-space trajectory or by the temporal order of k-space coverage, e.g. centric or segmented coverage of k-space}
- G01R: MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES (indicating correct tuning of resonant circuits H03J3/12)

## Abstract

The present disclosure provides an imaging method using a mobile magnetic resonance apparatus, which includes: randomly sampling an object-to-be-scanned using a mobile magnetic resonance apparatus to acquire encoding of the frequency and phase of the tissue of the object-to-be-scanned, and obtain target K-space data; inputting the target K-space data and pre-generated denoising reconstruction network parameters into a pre-trained denoising reconstruction network; where the pre-trained denoising reconstruction network is generated by training based on fully sampled training data; and outputting a magnetic resonance image corresponding to the object-to-be-scanned. The fully sampled training data is constructed by increasing the scanning time in the present application, the pre-trained denoising reconstruction network is obtained through model training based on the training data, real-time data is obtained through random sampling in a short period of time in practical applications, and high-quality images are output in combination with the pre-trained denoising reconstruction network.

## Claims

### Claim 1

An imaging method using a mobile magnetic resonance apparatus, comprising:
randomly sampling an object-to-be-scanned using a mobile magnetic resonance apparatus to acquire encoding of the frequency and phase of the tissue of the object-to-be-scanned, and obtain target K-space data;
inputting the target K-space data and pre-generated denoising reconstruction network parameters into a pre-trained denoising reconstruction network; wherein the pre-trained denoising reconstruction network is generated by training based on fully sampled training data; and
outputting a magnetic resonance image corresponding to the object-to-be-scanned,
wherein the pre-trained denoising reconstruction network is generated according to the following steps, which comprise:
using the mobile magnetic resonance apparatus to perform full sampling on multiple scanning objects to acquire the encoding of the frequency and phase of the tissue of each scanning object, and obtain multiple pieces of K-space training data;
constructing fully sampled training data based on the multiple pieces of K-space training data;
constructing a target denoising reconstruction network, inputting the fully sampled training data into the target denoising reconstruction network, and outputting a network cost value; and
generating a pre-trained denoising reconstruction network and pre-generated denoising reconstruction network parameters when the network cost value reaches its minimum,
wherein the constructing fully sampled training data based on the multiple pieces of K-space training data comprises:
performing image reconstruction based on each K-space data to obtain the magnetic resonance image of each K-space data;
associating each K-space data with its corresponding magnetic resonance image to obtain a K-space image dataset; and
randomly dividing the K-space image dataset into equal parts, and determining a preset number of parts of the K-space image dataset as the fully sampled training data.

### Claim 2 (depends on 1)

The method according to claim 1, wherein the constructing a target denoising reconstruction network comprises:
constructing a denoising reconstruction network using neural networks;
constructing a cost function of the denoising reconstruction network; and
mapping the cost function to the denoising reconstruction network to obtain a target denoising reconstruction network;
wherein the cost function is:
F
⁡
(
x
)
=
min
x
,
D
,
{
γ
i
}
∑
i
(
(

R
i
⁢
x
-
D
⁢
γ
i

)
+
α
⁢

Y
-
FFTx

2
2
+
β
⁢

R
i
⁢
x
-
FFTx

2
2
;
where Ri is a feature coefficient vector, x is a feature vector, D is the fully sampled training data, γi is the parameter of the denoising reconstruction network, FFT is the discrete Fourier transform, Y is a true value of the reconstructed image, and α and β are constants.

### Claim 3 (depends on 1)

The method according to claim 1, wherein before the randomly sampling an object-to-be-scanned using a mobile magnetic resonance apparatus, the method further comprises:
using a random sampling function to construct a random sampling block of a preset size to obtain a data random collection layer; and
setting data collection parameters of the mobile magnetic resonance apparatus as the data random collection layer.

### Claim 4 (depends on 3)

The method according to claim 3, wherein the function of the random sampling block is:
K
S
(
i
,
j
)
=
{
Rand
S
(
i
,
j
)
≥
0.5
Rand
S
⁢
(
i
,
j
)
<
0.5
,
0
≤
i
≤
S
,
0
≤
j
≤
S
;
where Ks is the random sampling block, i, j are element coordinates of the random sampling block, S is the size of the random sampling block, and Rands is the random sampling function.

### Claim 5 (depends on 1)

The method according to claim 1, wherein the generating a pre-trained denoising reconstruction network and pre-generated denoising reconstruction network parameters when the network cost value reaches its minimum comprises:
when the network cost value does not reach its minimum, backpropagating the network cost value to update the network parameters of the denoising reconstruction network, and continuing the execution of the step of “inputting the fully sampled training data into the target denoising reconstruction network and outputting a network cost value” until the network cost value reaches its minimum and the number of network training reaches a preset number to generate the pre-trained denoising reconstruction network and pre-generated denoising reconstruction network parameters.

### Claim 6

An imaging device using a mobile magnetic resonance apparatus, comprising:
a target K-space data generation module, which is configured to randomly sample an object-to-be-scanned using a mobile magnetic resonance apparatus to acquire encoding of the frequency and phase of the tissue of the object-to-be-scanned, and obtain target K-space data;
a data input module, which is configured to input the target K-space data and pre-generated denoising reconstruction network parameters into a pre-trained denoising reconstruction network; wherein the pre-trained denoising reconstruction network is generated based on fully sampled training data; and
a magnetic resonance image output module, which is configured to output a magnetic resonance image corresponding to the object-to-be-scanned,
wherein the pre-trained denoising reconstruction network is generated according to the following steps, which comprise:
using the mobile magnetic resonance apparatus to perform full sampling on multiple scanning objects to acquire the encoding of the frequency and phase of the tissue of each scanning object, and obtain multiple pieces of K-space training data;
constructing fully sampled training data based on the multiple pieces of K-space training data;
constructing a target denoising reconstruction network, inputting the fully sampled training data into the target denoising reconstruction network, and outputting a network cost value; and
generating a pre-trained denoising reconstruction network and pre-generated denoising reconstruction network parameters when the network cost value reaches its minimum;
wherein the constructing fully sampled training data based on the multiple pieces of K- space training data comprises:
performing image reconstruction based on each K-space data to obtain the magnetic resonance image of each K-space data;
associating each K-space data with its corresponding magnetic resonance image to obtain a K-space image dataset; and
randomly dividing the K-space image dataset into equal parts, and determining a preset number of parts of the K-space image dataset as the fully sampled training data.

### Claim 7 (depends on 1)

A computer-readable storage medium, storing multiple instructions suitable for being loaded by a processor to execute the method according to claim 1, wherein the storage medium includes: a magnetic disk, an optical disc, or a read-only storage memory.

### Claim 8 (depends on 1)

A terminal comprising a processor and a memory; wherein computer programs are stored in the memory, and the computer programs are suitable for being loaded by the processor to execute the method according to claim 1.