Deep learning based processing of motion artifacts in magnetic resonance imaging data

Patent metadata

Field	Value
Publication number	US-11320508-B2
Application number	US-201816759778-A
Country	US
Kind code	B2
Filing date	Oct 22, 2018
Priority date	Oct 31, 2017
Publication date	May 3, 2022
Grant date	May 3, 2022

Abstract

The invention relates to a magnetic resonance imaging data processing system (126) for processing motion artifacts in magnetic resonance imaging data sets using a deep learning network (146, 502, 702) trained for the processing of motion artifacts in magnetic resonance imaging data sets. The magnetic resonance imaging data processing system (126) comprises a memory (134, 136) storing machine executable instructions (161, 164) and the trained deep learning network (146, 502, 702). Furthermore, the magnetic resonance imaging data processing system (126) comprises a processor (130) for controlling the magnetic resonance imaging data processing system. Execution of the machine executable instructions (161, 164) causes the processor (130) to control the magnetic resonance imaging data processing system (126) to: receive a magnetic resonance imaging data set (144, 500, 800), apply the received magnetic resonance imaging data set (144, 500, 800) as an input to the trained deep learning network (146, 502, 702), process one or more motion artifacts present in the received magnetic resonance imaging data set (144, 500, 800) using the trained deep learning network (146, 502, 702).

First claim

The invention claimed is:

1. A magnetic resonance imaging data processing system for processing motion artifacts in magnetic resonance imaging data sets using a deep learning network trained for the processing of motion artifacts in magnetic resonance imaging data sets, the magnetic resonance imaging data processing system comprising:
a memory storing machine executable instructions and the trained deep learning network,
a processor for controlling the magnetic resonance imaging data processing system, wherein an execution of the machine executable instructions causes the processor to control the magnetic resonance imaging data processing system to:
receive a magnetic resonance imaging data set,
apply the received magnetic resonance imaging data set as an input to the trained deep learning network, and
process one or more motion artifacts present in the received magnetic resonance imaging data set using the trained deep learning network,
wherein the deep learning network is further trained for filtering motion artifacts present in magnetic resonance imaging data sets, wherein the processing further comprises:
filtering the one or more motion artifacts present in the magnetic resonance imaging data set using the trained deep learning network; and
providing a motion-artifact-corrected magnetic resonance imaging data set using a result of the filtering;
wherein the deep learning network is further trained for detecting the presence of motion artifacts in magnetic resonance imaging data sets, wherein the processing comprises detecting the presence of the one or more motion artifacts in the received magnetic resonance imaging data set, and wherein the execution of the machine executable instructions further causes the processor to control the magnetic resonance imaging data processing system to:
indicate the presence of the one or more motion artifacts in the received magnetic resonance imaging data set;
wherein the deep learning network is further trained for determining a motion artifact level of magnetic resonance imaging data sets, the motion artifact level characterizing the number and/or degree of motion artifacts present in the respective magnetic resonance imaging data set,
wherein the processing further comprises:
determining the motion artifact level of the received magnetic resonance imaging data set based on the one or more motion artifacts detected to be present in the received magnetic resonance imaging data set using the trained deep learning network,
providing the motion artifact level as output from the trained deep learning network,
wherein the indicating comprises assigning a motion artifact level identifier to the received magnetic resonance imaging data set identifying the determined motion artifact level.

2. The magnetic resonance imaging data processing system of claim 1 , wherein the deep learning network is a deep convolutional neural network implementing deep learning.

3. The magnetic resonance imaging data processing system of claim 1 , wherein the deep learning network is a fully convolutional network implementing deep learning.

4. The magnetic resonance imaging data processing system of claim 1 , wherein the execution of the machine executable instructions further causes the processor to control the magnetic resonance imaging data processing system to train the deep learning network, wherein the training comprises:
providing a training set comprising a plurality of magnetic resonance imaging training data sets with and without motion artifacts.

5. The magnetic resonance imaging data processing system of claim 4 , wherein the providing of the training set comprises:
generating the magnetic resonance imaging training data sets with motion artifacts, wherein the generating of the magnetic resonance imaging training data sets comprises introducing varying numbers, degrees and/or types of artificially generated motion artifacts to magnetic resonance imaging data sets without motion artifacts.

6. The magnetic resonance imaging data processing system of claim 4 , wherein each of the magnetic resonance imaging training data sets is assigned with a motion artifact level identifier, wherein the training comprises training the deep learning network for determining motion artifact levels of magnetic resonance imaging data sets comprising:
applying the magnetic resonance imaging training data sets as input to the deep learning network,
determining for each of the magnetic resonance imaging training data sets a motion artifact level of the respective magnetic resonance imaging training data set using the trained deep learning network,
providing the motion artifact levels of the magnetic resonance imaging training data sets as output from the deep learning network,
comparing the output of the deep learning network with the motion artifact levels identified by the motion artifact level identifiers assigned to the input to the deep learning network,
adapting network parameters of the deep learning network in order to reduce differences between the output of the deep learning network and the motion artifact levels identified by the motion artifact level identifiers assigned to the input to the deep learning network.

7. The magnetic resonance imaging data processing system of claim 4 , wherein the training set further comprises for each of the magnetic resonance imaging training data sets a magnetic resonance imaging reference data set assigned to the respective magnetic resonance imaging training data set, wherein the magnetic resonance imaging reference data set is a motion-artifact-free version of the magnetic resonance imaging training data set to which it is assigned, wherein the training comprises training the deep learning network for filtering motion artifacts present in magnetic resonance imaging data sets comprising:
applying the magnetic resonance imaging training data sets as input to the deep learning network,
filtering motion artifacts of the magnetic resonance imaging training data sets using the trained deep learning network,
providing for each of the magnetic resonance imaging training data sets a motion-artifact-corrected magnetic resonance imaging data set using a result of the filtering,
comparing the motion-artifact-corrected magnetic resonance imaging data sets with the magnetic resonance imaging reference data sets,
adapting network parameters of the deep learning network in order to reduce differences between motion-artifact-corrected magnetic resonance imaging data sets and the magnetic resonance imaging reference data sets.

8. The magnetic resonance imaging data processing system of claim 7 , wherein the result of the filtering comprises the motion-artifact-corrected magnetic resonance imaging data sets which are provided as output from the deep learning network or
wherein the result of the filtering comprises motion-artifact-only magnetic resonance imaging data sets provided as output from the deep learning network and wherein the providing of the motion-artifact-corrected magnetic resonance imaging data sets comprises subtracting the motion-artifact-only magnetic resonance imaging data sets from the magnetic resonance imaging training data sets.

9. A magnetic resonance imaging system comprising the magnetic resonance imaging data processing system of claim 1 , wherein the magnetic resonance imaging system further comprises:
a main magnet for generating a main magnetic field within an imaging zone,
a magnetic field gradient system for generating a spatially dependent gradient magnetic field within the imaging zone,
a radio-frequency antenna system configured for acquiring magnetic resonance imaging data from the imaging zone,
wherein the memory further stores pulse seque

Assignees

• Koninklijke Philips Nv

Inventors

• Sommer Karsten
• Brosch Tom
• Harder Tim Philipp
• Keupp Jochen
• Graesslin Ingmar
• Wiemker Rafael
• Saalbach Axel

Classifications

• G06N3/045

  Combinations of networks · CPC title

• G06N3/0464

  Convolutional networks [CNN, ConvNet] · CPC title

• G06N3/09

  Supervised learning · CPC title

• G06T5/60

  using machine learning, e.g. neural networks · CPC title

• G01R33/56509 Primary

  due to motion, displacement or flow, e.g. gradient moment nulling (G01R33/567 takes precedence) · CPC title