# US 2019/0377047 A1 · Artifact Reduction by Image-to-Image Network in Magnetic Resonance Imaging

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

- Number: US 2019/0377047 A1
- Document ID: US20190377047A1
- Status: Published application
- Kind: A1
- Application number: 16/002,447
- Filed: 2018-06-07
- Published: 2019-12-12
- Earliest claimed: 2018-06-07
- Assignees: Siemens Healthcare GmbH
- Applicants: Siemens Healthcare GmbH
- Inventors: Xiao Chen; Boris Mailhe; Benjamin L. Odry; Pascal Ceccaldi; Mariappan S. Nadar
- Web page: https://patentgrep.com/patent/US20190377047A1

## 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 33/565: Correction of image distortions, e.g. due to magnetic field inhomogeneities
- 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 33/56509: {due to motion, displacement or flow, e.g. gradient moment nulling (G01R33/567 takes precedence)}
- G06T 12/30: Image post-processing, e.g. metal artefact correction
- G01R 33/56518: {due to eddy currents, e.g. caused by switching of the gradient magnetic field}
- G06T 2210/41: Medical
- G01R: MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES (indicating correct tuning of resonant circuits H03J3/12)
- G06T: IMAGE DATA PROCESSING OR GENERATION, IN GENERAL

## Abstract

For artifact reduction in a magnetic resonance imaging system, deep learning trains an image-to-image neural network to generate an image with reduced artifact from input, artifacted MR data. For application, the image-to-image network may be applied in real time with a lower computational burden than typical post-processing methods. To handle a range of different imaging situations, the image-to-image network may (a) use an auxiliary map as an input with the MR data from the patient, (b) use sequence metadata as a controller of the encoder of the image-to-image network, and/or (c) be trained to generate contrast invariant features in the encoder using a discriminator that receives encoder features.

## Claims

### Claim 1

A method for artifact reduction in a magnetic resonance imaging system, the method comprising:
scanning a patient by the magnetic resonance imaging system, the scanning providing at least first and second sets of magnetic resonance data, the first and second sets representing response to the scanning from first and second distributions of locations;
inputting the first and second sets of the magnetic resonance data to a neural network machine trained as an image-to-image network;
generating an image by the neural network in response to the inputting of the first and second sets; and
displaying the image.

### Claim 2 (depends on 1)

The method of claim 1 wherein scanning comprises acquiring the first set with a pre-scan of the patient and acquiring the second set as an imaging scan of the patient based on the pre-scan.

### Claim 3 (depends on 2)

The method of claim 2 wherein acquiring the first set comprises acquiring a coil correction pre-scan or a noise map.

### Claim 4 (depends on 1)

The method of claim 1 wherein scanning comprises scanning the patient and a background region around the patient, the first set being for the background region and the second set being for the patient.

### Claim 5 (depends on 1)

The method of claim 1 wherein inputting comprises inputting the second set as k-space data or as image data reconstructed from the k-space data, the second set including artifact information, and wherein the neural network was trained from third sets including artifacts and ground truth images with a reduced contribution from the artifacts.

### Claim 6 (depends on 1)

The method of claim 1 wherein inputting comprises inputting the first and second sets on separate input channels of the neural network, wherein the first and second distributions of locations comprise a same distribution of the same locations, and wherein inputting comprises inputting the first and second sets as a same size.

### Claim 7 (depends on 1)

The method of claim 1 wherein the first set comprises an auxiliary map, and the second set comprises the magnetic resonance data for representing the patient in imaging and including contribution from the artifact, and wherein generating the image comprises generating the image with an artifact level reduced relative to the contribution from the artifact.

### Claim 8 (depends on 1)

The method of claim 1 wherein scanning comprises scanning with a scan sequence, and wherein inputting comprises inputting information about the scan sequence to an encoder of the image-to-image network.

### Claim 9 (depends on 1)

The method of claim 1 wherein scanning comprises scanning with a scan sequence, and wherein the image-to-image network includes an encoder having been trained with a discriminator of scan sequences.

### Claim 10

A method for artifact reduction in a magnetic resonance imaging system, the method comprising:
scanning a patient by the magnetic resonance imaging system, the scanning providing a set of magnetic resonance data based on a scan sequence;
inputting the set of the magnetic resonance data to a neural network machine trained as an image-to-image network;
inputting information about the scan sequence to an encoder or decoder of the image-to-image network;
generating an image of the patient by the neural network in response to the inputting of the set and the information; and
displaying the image.

### Claim 11 (depends on 10)

The method of claim 10 wherein inputting the set comprises inputting the set as k-space data or as image data reconstructed from the k-space data, the set including artifact information, and wherein the neural network was trained from training sets of magnetic resonance data including artifacts and ground truth images with a reduced contribution from the artifacts.

### Claim 12 (depends on 10)

The method of claim 10 wherein inputting the information comprises inputting sequence metadata.

### Claim 13 (depends on 12)

The method of claim 12 wherein inputting the sequence metadata comprises inputting a sequence type and/or sequence protocol values.

### Claim 14 (depends on 10)

The method of claim 10 wherein inputting the information comprises inputting the information in a conditional matrix.

### Claim 15 (depends on 10)

The method of claim 10 wherein inputting the information comprises inputting parameters characterizing the sequence as a conditional matrix configured to tune features of the encoder.

### Claim 16 (depends on 10)

The method of claim 10 wherein the magnetic resonance data includes contribution from the artifact, and wherein generating the image comprises generating the image with an artifact level reduced relative to the contribution from the artifact.

### Claim 17 (depends on 10)

The method of claim 10 wherein inputting the set comprises inputting the set and an auxiliary map on separate input channels of the encoder or decoder.

### Claim 18 (depends on 10)

The method of claim 10 wherein the image-to-image network includes the encoder having been trained with a discriminator of scan sequences.

### Claim 19

A method of training a neural network for artifact reduction in a magnetic resonance imaging system, the method comprising:
obtaining sets of magnetic resonance data and ground truth images;
defining an image-to-image network with an encoder and a discriminator connected with the encoder, the discriminator configured to discriminate between scan sequences;
machine training the image-to-image network to generate a patient image from the sets and the ground truth images, the training using the discriminator; and
storing the image-to-image network as trained.

### Claim 20 (depends on 19)

The method of claim 19 wherein the sets include artifact information, the ground truth images have a level of artifact reduced relative to the artifact information, and wherein machine training comprises machine training the encoder of the image-to-image network to generate contrast invariant features.

### Claim 21 (depends on 19)

The method of claim 19 wherein defining comprises defining the discriminator to receive features from each pooling layer of the encoder.

### Claim 22 (depends on 19)

The method of claim 19 wherein defining comprises defining the encoder with a first input channel for the magnetic resonance data and a second input channel for an auxiliary map.

### Claim 23 (depends on 19)

The method of claim 19 wherein defining comprises defining the encoder to receive scan sequence metadata as controllers of the encoder.
