Motion artifact prediction during data acquisition

1. A magnetic resonance imaging system, the magnetic resonance imaging system comprising:

a non-transitory memory storing machine executable instructions, pulse sequence commands and a first machine learning model comprising a first deep learning network, wherein the pulse sequence commands are configured for controlling the magnetic resonance imaging system to acquire a set of magnetic resonance imaging data, wherein the first machine learning model comprises a first input and a first output; and

a processor, wherein execution of the machine executable instructions causes the processor to control the magnetic resonance imaging system to repeatedly perform an acquisition and analysis process, wherein for each current repetition, the acquisition and analysis process comprises:

acquiring a current dataset comprising a subset of the set of magnetic resonance imaging data from an imaging zone of the magnetic resonance imaging system according to the pulse sequence commands;

providing the current dataset to the first input of the first machine learning model in combination with at least a previous dataset acquired in a previous repetition preceding the current repetition;

receiving a prediction of a motion artifact level of the acquired magnetic resonance imaging data from the first output of the first machine learning model, the motion artifact level characterizing a number and/or extent of motion artifacts present in the acquired magnetic resonance imaging data;

when the prediction indicates an increased number and/or extent of motion artifacts present in the acquired magnetic resonance imaging data relative to a previous prediction received in the previous repetition preceding the current repetition, checking whether omitting the current repetition results in a next prediction received in a next repetition following the current repetition that also indicates an increased number and/or extent of motion artifacts present in the acquired magnetic resonance imaging data relative to the previous prediction of the previous repetition preceding the current repetition;

when omitting the current repetition does not result in the next prediction indicating an increased number and/or extent of motion artifacts, continuing the repeated performing of the acquisition and analysis process without the current dataset of the current repetition; and

when omitting the current repetition does result in the next prediction indicating an increased number and/or extent of motion artifacts, continuing the repeated performing of the acquisition and analysis process with the current dataset of the current repetition.

2. The magnetic resonance imaging system of claim 1 , wherein execution of the machine executable instructions further causes the processor to control the magnetic resonance imaging system to automatically abort the repeated performing of the acquisition and analysis process, when the prediction of the motion artifact level exceeds a first predefined threshold.

3. The magnetic resonance imaging system of claim 2 , wherein execution of the machine executable instructions further causes the processor to control the magnetic resonance imaging system to automatically restart the repeated performing of the acquisition and analysis process when the prediction of the motion artifact level exceeds the first predefined threshold.

4. The magnetic resonance imaging system of claim 2 , wherein the memory further stores a second machine learning model comprising a second deep learning network, wherein the second machine learning model comprises a second input and a second output,

wherein execution of the machine executable instructions further causes the processor to control the magnetic resonance imaging system to:

when the prediction of the motion artifact level exceeds a second predefined threshold, provide the current dataset to the second input of the second machine learning model,

provide a motion-artifact-corrected dataset as a replacement for the current dataset using a response received from the second output of the second machine learning model, and

continue the repeated performing of the acquisition and analysis process with the motion-artifact-corrected dataset.

5. The magnetic resonance imaging system of claim 4 , wherein the memory further stores a second learning algorithm for generating the second machine learning model, wherein execution of the machine executable instructions further causes the processor to control the magnetic resonance imaging system to:

receive second training sets, each second training set comprising a magnetic resonance imaging dataset and a motion-artifact-only magnetic resonance imaging dataset assigned to the magnetic resonance imaging dataset, and

execute the second learning algorithm on the received second training sets for generating the second machine learning model being trained with each of the second training sets to provide via the second output in response to receiving via the second input the magnetic resonance imaging dataset of a respective second training set, the motion-artifact-only magnetic resonance imaging dataset of the respective second training set enabling a providing of the motion-artifact-corrected dataset using the magnetic resonance imaging dataset provided to the second input by subtracting the motion-artifact-only magnetic resonance imaging dataset received from the second output.

6. The magnetic resonance imaging system of claim 1 , wherein the prediction of the motion artifact level depends on a location of the motion artifact relative to one or more anatomical structures of interest represented by the acquired magnetic resonance imaging data.

7. The magnetic resonance imaging system of claim 1 , wherein the current dataset comprises a magnetic resonance image reconstructed using the acquired magnetic resonance imaging data.

8. The magnetic resonance imaging system of claim 1 , wherein the current dataset has a common predefined size as other datasets, wherein the current dataset comprises magnetic resonance imaging data from sampling points distributed over k-space with a higher sampling rate at a center of the k-space relative to an outer portion of the k-space.

9. The magnetic resonance imaging system of claim 1 , wherein the current dataset has an arbitrary size being selected within a range defined by a predefined minimum size and a predetermined maximum size, wherein the current dataset comprises magnetic resonance imaging data from sampling points distributed over k-space with a higher sampling rate at a center of the k-space relative to an outer portion of the k-space.

10. The magnetic resonance imaging system of claim 1 , wherein the memory further stores a first learning algorithm for generating the first machine learning model, wherein execution of the machine executable instructions further causes the processor to control the magnetic resonance imaging system to:

receive first training sets, each first training set comprising a magnetic resonance imaging dataset and an artifact level identifier identifying an artifact level assigned to the respective magnetic resonance imaging dataset,

execute the first learning algorithm on the received first training sets for generating the first machine learning model.

11. A non-transitory computer readable medium storing machine executable instructions for execution by a processor controlling a magnetic resonance imaging system using pulse sequence commands and a first machine learning model comprising a first deep learning network, wherein the pulse sequence commands are configured for controlling the magnetic resonance imaging system to acquire a set of magnetic resonance imaging data, wherein the first machine learning model comprises a first input and a first output,

wherein execution of the machine executable instructions causes the processor to repeatedly perform an acquisition and analysis process, wherein for each current repetition, the acquisition and analysis process comprises:

acquiring a current dataset comprising a subset of the set of magnetic resonance imaging data from an imaging zone of the magnetic resonance imaging system according to the pulse sequence commands;

providing the current dataset to the first input of the first machine learning model in combination with at least a previous dataset acquired in a previous repetition preceding the current repetition;

receiving a prediction of a motion artifact level of the acquired magnetic resonance imaging data from the first output of the first machine learning model, the motion artifact level characterizing a number and/or extent of motion artifacts present in the acquired magnetic resonance imaging data;

when the prediction indicates an increased number and/or extent of motion artifacts present in the acquired magnetic resonance imaging data relative to a previous prediction received in the previous repetition preceding the current repetition, checking whether omitting the current repetition results in a next prediction received in a next repetition following the current repetition that also indicates an increased number and/or extent of motion artifacts present in the acquired magnetic resonance imaging data relative to the previous prediction of the previous repetition preceding the current repetition;

when omitting the current repetition does not result in the next prediction indicating an increased number and/or extent of motion artifacts, continuing the repeated performing of the acquisition and analysis process without the current dataset; and

when omitting the current repetition does result in the next prediction indicating an increased number and/or extent of motion artifacts, continuing the repeated performing with the current dataset.

12. The non-transitory computer readable medium of claim 11 , wherein execution of the machine executable instructions further causes the processor to control the magnetic resonance imaging system to automatically abort the repeated performing of the acquisition and analysis process, when the prediction of the motion artifact level exceeds a first predefined threshold.

13. The non-transitory computer readable medium of claim 12 , wherein execution of the machine executable instructions further causes the processor to control the magnetic resonance imaging system to automatically restart the repeated performing of the acquisition and analysis process when the prediction of the motion artifact level exceeds the first predefined threshold.

14. The non-transitory computer readable medium of claim 12 , further storing a second machine learning model comprising a second deep learning network, wherein the second machine learning model comprises a second input and a second output, wherein execution of the machine executable instructions further causes the processor to control the magnetic resonance imaging system to:

provide the current dataset to the second input of the second machine learning model when the prediction of the motion artifact level exceeds a second predefined threshold,

provide a motion-artifact-corrected dataset as a replacement for the current dataset using a response received from the second output of the second machine learning model, and

continue the repeated performing of the acquisition and analysis process with the motion-artifact-corrected dataset.

15. The non-transitory computer readable medium of claim 11 , wherein the prediction of the motion artifact level depends on a location of the motion artifact relative to one or more anatomical structures of interest represented by the acquired magnetic resonance imaging data.

16. The non-transitory computer readable medium of claim 11 , wherein the current dataset comprises a magnetic resonance image reconstructed using the acquired magnetic resonance imaging data.

17. A method of operating a magnetic resonance imaging system, the method comprising:

providing pulse sequence commands for controlling the magnetic resonance imaging system to acquire a set of magnetic resonance imaging data;

providing a first machine learning model comprising a first deep learning network, wherein the first machine learning model comprises a first input and a first output; and

repeatedly performing an acquisition and analysis process, wherein for each current repetition, the acquisition and analysis process comprises:

acquiring a current dataset comprising a subset of the set of magnetic resonance imaging data from an imaging zone of the magnetic resonance imaging system according to the pulse sequence commands;

providing the current dataset to the first input of the first machine learning model in combination with at least a previous dataset acquired in a previous repetition preceding the current repetition;

receiving a prediction of a motion artifact level of the acquired magnetic resonance imaging data from the first output of the first machine learning model, the motion artifact level characterizing a number and/or extent of motion artifacts present in the acquired magnetic resonance imaging data;

when the prediction indicates an increased number and/or extent of motion artifacts present in the acquired magnetic resonance imaging data relative to a previous prediction received in the previous repetition preceding the current repetition, checking whether omitting the current repetition results in a next prediction received in in a next repetition following the current repetition that also indicates an increased number and/or extent of motion artifacts present in the acquired magnetic resonance imaging data relative to the prediction of the previous repetition preceding the current repetition;

when omitting the current repetition does not result in the next prediction indicating an increased number and/or extent of motion artifacts, continuing the repeated performing of the acquisition and analysis process without the current dataset; and

when omitting the current repetition does not in the next prediction indicating an increased number and/or extent of motion artifacts, continuing the repeated performing of the acquisition and analysis process with the current dataset.

18. The method of claim 17 , further comprising:

automatically aborting the repeated performing of the acquisition and analysis process when the prediction of the motion artifact level exceeds a first predefined threshold.

19. The method of claim 18 , further comprising:

automatically restarting the repeated performing of the acquisition and analysis process when the prediction of the motion artifact level exceeds the first predefined threshold.

20. The method of claim 18 , further comprising:

providing a second machine learning model comprising a second deep learning network, wherein the second machine learning model comprises a second input and a second output;

providing the current dataset to the second input of the second machine learning model when the prediction of the motion artifact level exceeds a second predefined threshold;

providing a motion-artifact-corrected dataset as a replacement for the current dataset using a response received from the second output of the second machine learning model; and

continuing the repeated performing of the acquisition and analysis process with the motion-artifact-corrected dataset.