# US 2022/0036512 A1 · Apparatus and method for image restoration of accelerated mri based on deep learning

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

- Number: US 2022/0036512 A1
- Document ID: US20220036512A1
- Status: Published application
- Kind: A1
- Application number: 17/505,159
- Filed: 2021-10-19
- Published: 2022-02-03
- Earliest claimed: 2020-01-28
- Assignees: Claripi Inc.; Seoul National University R&Db Foundation
- Applicants: ClariPI Inc.; Seoul National University R&DB Foundation
- Inventors: Jong Hyo KIM; Hyun Sook PARK; Tai Chul PARK; Chul Kyun AHN
- Web page: https://patentgrep.com/patent/US20220036512A1

## 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)}
- A61B 5/00: Measuring for diagnostic purposes (radiation diagnosis A61B6/00; diagnosis by ultrasonic, sonic or infrasonic waves A61B8/00); Identification of persons
- A61B 5/055: involving electronic [EMR] or nuclear [NMR] magnetic resonance, e.g. magnetic resonance imaging
- 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 33/56545: {caused by finite or discrete sampling, e.g. Gibbs ringing, truncation artefacts, phase aliasing artefacts}
- G06T 2207/10088: Magnetic resonance imaging [MRI]
- G06T 2207/20081: Training; Learning
- G16H 30/40: for processing medical images, e.g. editing
- G01R: MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES (indicating correct tuning of resonant circuits H03J3/12)
- A61B: DIAGNOSIS; SURGERY; IDENTIFICATION (usefulness limited to only animals A61D)
- G06T: IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G16H: HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA

## Abstract

Provided is a deep learning based accelerated MRI image quality restoring method. The deep learning based accelerated MRI image quality restoring method includes extracting test information from an input accelerated MRI image, selecting at least one deep learning model corresponding to the test information, among a plurality of previously trained deep learning models, and outputting an MRI image with a restored image quality for the input accelerated MRI image with the input accelerated MRI image as an input of at least one selected deep learning model.

## Claims

### Claim 1

A deep learning based accelerated MRI image quality restoring method, the method comprising:
extracting test information from an input accelerated MRI image;
selecting at least one deep learning model corresponding to the test information, among a plurality of previously trained deep learning models; and
outputting an MRI image with a restored image quality with respect to the input accelerated MRI image with the input accelerated MRI image as an input of at least one selected deep learning model.

### Claim 2 (depends on 1)

The restoring method according to claim 1, further comprising:
before the extracting,
generating a low quality of second MRI data set for training by applying an MRI image accelerated scanning simulator to a first MRI data set for training;
extracting test information from the second MRI data set for training and grouping the second MRI data set for training into a plurality of groups according to a predetermined rule; and
generating and training a plurality of deep learning models to be trained so as to correspond to each grouped second MRI data set for every group,
wherein in the selecting, the plurality of previously trained deep learning models is a plurality of deep learning models to be trained which is trained by the training.

### Claim 3 (depends on 2)

The restoring method according to claim 2, wherein in the generating step, the MRI image accelerated scanning simulator performs:
generating composite k-space data with an original image of the first MRI data set for training as an input;
generating first low quality composite k-space data by applying a predetermined level of sub sampling to the composite k-space data;
generating second low quality composite k-space data by adding a predetermined level of noise to the first low quality composite k-space data;
generating a composite low quality MRI image based on the second low quality composite k-space data; and
subtracting the original image from the generated composite low quality MRI image to generate a composite image quality degraded component MRI image.

### Claim 4 (depends on 2)

The restoring method according to claim 2, wherein the second MRI data set for training is formed of a pair of a composite low quality MRI image and a composite image quality degraded component MRI image obtained based on an original image of the first MRI data set for training.

### Claim 5 (depends on 2)

The restoring method according to claim 2, wherein in the training, in order to allow the plurality of deep learning models to be trained to have a function of extracting an image quality degraded component MRI image from the input accelerated MRI image input thereto, a composite accelerated MRI image for every MRI image is transmitted as an input of the deep learning model to be trained, in the grouped second MRI data set for training for every, and the deep learning model to be trained is repeatedly trained so as to minimize a difference between the composite image quality degraded component MRI image and the output of the deep learning model to be trained.

### Claim 6 (depends on 2)

The restoring method according to claim 2, wherein in the outputting, at least one selected deep learning model extracts an image quality degraded component MRI image from the input accelerated MRI image with the input accelerated MRI image as an input of at least one selected deep learning model and a predetermined value is multiplied with the extracted image quality degraded component MRI image to be subtracted from the input accelerated MRI image to output an MRI image with a restored image quality.

### Claim 7

A deep learning based accelerated MRI image quality restoring apparatus, the apparatus comprising:
an extracting unit which extracts test information from an input accelerated MRI image;
a selecting unit which selects at least one deep learning model corresponding to the test information, among a plurality of previously trained deep learning models; and
an output unit which outputs an MRI image with a restored image quality for the input accelerated MRI image with the input accelerated MRI image as an input of at least one selected deep learning model.

### Claim 8 (depends on 7)

The restoring apparatus according to claim 7, further comprising:
a training unit which generates and trains a plurality of deep learning models to be trained,
wherein the training unit generates a second MRI data set for training with a low quality by applying an MRI image accelerated scanning simulator to a first MRI data set for training, extracts test information from the second MRI data set for training, groups the second MRI data sets for training into a plurality of groups according to a predetermined rule, and generates and trains a plurality of deep learning models to be trained so as to correspond to each of the grouped second MRI data set for training for every group, and the plurality of previously trained deep learning models is the plurality of deep learning models to be trained which is trained by the training unit.

### Claim 9 (depends on 8)

The restoring apparatus according to claim 8, wherein the MRI image accelerated scanning simulator generates composite k-space data with an original image of the first MRI data set for training as an input, applies a predetermined level of sub sampling to the composite k-space data to generate first low quality composite k-space data, adds a predetermined level of noise to the first low quality composite k-space data to generate second low quality composite k-space data, generates a composite low quality MRI image based on the second low quality composite k-space data, and subtracts the original image from the generated composite low quality MRI image to generate a composite image quality degraded component MRI image.

### Claim 10 (depends on 8)

The restoring apparatus according to claim 8, wherein the second MRI data set for training is formed of a pair of a composite low quality MRI image and a composite image quality degraded component MRI image obtained based on an original image of the first MRI data set for training.

### Claim 11 (depends on 8)

The restoring apparatus according to claim 8, wherein in order to allow the plurality of deep learning models to be trained to have a function of extracting an image quality degraded component MRI image from the input MRI image input thereto, the training unit transmits a composite accelerated MRI image for every MRI image as an input of the deep learning model to be trained, in the grouped second MRI data set for training for every and repeatedly trains the deep learning model to be trained so as to minimize a difference between the composite image quality degraded component MRI image and the output of the deep learning model to be trained.

### Claim 12 (depends on 7)

The restoring apparatus according to claim 7, wherein the output unit allows at least one selected deep learning model to extract an image quality degraded component MRI image from the input accelerated MRI image with the input accelerated MRI image as an input of the at least one selected deep learning model and multiplies a predetermined value with the extracted image quality degraded component MRI image to be subtracted from the input accelerated MRI image to output an MRI image with a restored image quality.

### Claim 13 (depends on 1)

A non-transitory computer readable recording medium in which a program allowing a computer to execute the method according to claim 1 is recorded.

### Claim 14 (depends on 2)

A non-transitory computer readable recording medium in which a program allowing a computer to execute the method according to claim 2 is recorded.

### Claim 15 (depends on 3)

A non-transitory computer readable recording medium in which a program allowing a computer to execute the method according to claim 3 is recorded.

### Claim 16 (depends on 4)

A non-transitory computer readable recording medium in which a program allowing a computer to execute the method according to claim 4 is recorded.

### Claim 17 (depends on 5)

A non-transitory computer readable recording medium in which a program allowing a computer to execute the method according to claim 5 is recorded.

### Claim 18 (depends on 6)

A non-transitory computer readable recording medium in which a program allowing a computer to execute the method according to claim 6 is recorded.
