Image quality assessment system and method

• US 10,713,785 B2
• Filed: 02/09/2018
• Issued: 07/14/2020
• Est. Priority Date: 02/13/2017
• Status: Active Grant

• Current Assignee: Siemens Healthineers AG (Siemens AG)
• Original Assignee: Siemens Healthcare GMBH (Siemens AG)
• Inventors: Braun, Sandro, Lu, Xiaoguang, Mailhe, Boris, Odry, Benjamin L., Chen, Xiao, Nadar, Mariappan S.
• Primary Examiner(s): Garcia, Santiago
• Application Number:
  US15/892,746
• Publication Number:
  US 20180232878A1
• Time in Patent Office: 886 Days
• CPC Class Codes: G06T 2207/10088 Magnetic resonance imaging [MRI]
• CPC Class Codes: G06T 2207/20081 Training; Learning
• CPC Class Codes: G06T 2207/20084 Artificial neural networks [ANN]
• CPC Class Codes: G06T 2207/30016 Brain
• CPC Class Codes: G06T 2207/30168 Image quality inspection
• CPC Class Codes: G06T 7/0002 Inspection of images, e.g. flaw detection
• CPC Class Codes: G06T 7/0012 Biomedical image inspection
• CPC Class Codes: G06T 7/194 involving foreground-background segmentation
• CPC Class Codes: G06T 7/20 Analysis of motion motion estimation for coding, decoding, compressing or decompressing digital video signals H04N19/43, H04N19/51

Abstract

A system and method includes generation of one or more motion-corrupted images based on each of a plurality of reference images, and training of a regression network to determine a motion score, where training of the regression network includes input of a generated motion-corrupted image to the regression network, reception of a first motion score output by the regression network in response to the input image, and determination of a loss by comparison of the first motion score to a target motion score, the target motion score calculated based on the input motion-corrupted image and a reference image based on which the motion-corrupted image was generated.

Claims

1. A system comprising:

a memory storing processor-executable program code; and

a processor to execute the processor-executable program code to cause the system to;

acquire a plurality of motion free reference images;

generate one or more motion-corrupted images by modifying each of the plurality of motion free reference images to include effects of one or more motion types; and

train a regression network to determine a motion score, where training of the regression network comprises;

input of a generated motion-corrupted image to the regression network;

reception of a first motion score output by the regression network in response to the input image; and

determination of a loss by comparison of the first motion score to a target motion score, the target motion score calculated based on the input motion-corrupted image and a reference image that was modified to generate the motion-corrupted image.

2. A system according to claim 1, the processor to further execute the processor-executable program code to cause the system to:

input an image to the trained regression network;

receive a second motion score output by the trained regression network;

compare the second motion score to a threshold motion score; and

determine whether the image is of acceptable quality based on the comparison between the second motion score and the threshold motion score.

3. A system according to claim 2, the processor to further execute the processor-executable program code to cause the system to determine the threshold motion score by:

inputting a plurality of images indicated as acceptable and a plurality of images indicated as unacceptable to the trained regression network;

determining a motion score for each of the plurality of images indicated as acceptable and the plurality of images indicated as unacceptable; and

determining the threshold motion score based on the motion scores for each of the plurality of images indicated as acceptable and the motion scores for each of the plurality of images indicated as unacceptable.

4. A system according to claim 2, further comprising:

an imaging system to acquire the image.

5. A system according to claim 1, wherein generation of the one or more motion-corrupted images comprises:

determination of k-space data of a first reference image;

phase-shifting of the k-space data to introduce first motion in the k-space data; and

reconstruction of a first motion-corrupted image from the phase-shifted k-space data.

6. A system according to claim 5, wherein generation of the one or more motion-corrupted images comprises:

phase-shifting of the k-space data to introduce second motion in the k-space data; and

reconstruction of a second motion-corrupted image from the k-space data phase-shifted to introduce second motion.

7. A system according to claim 1, wherein input of a generated motion-corrupted image to the regression network comprises:

segmentation of the input image into two or more segmented images; and

input of the segmented images into the regression network.

8. A computer-implemented method comprising:

generating one or more motion-corrupted images by modifying each of a plurality of reference images to include one or more motion effects previously not exhibited by a respective reference image; and

training a regression network to determine a motion score, where training of the regression network comprises;

inputting of a generated motion-corrupted image to the regression network;

receiving a first motion score output by the regression network in response to the input image; and

determining a loss by comparing the first motion score to a target motion score, the target motion score calculated based on the input motion-corrupted image and a reference image that was modified to generate the motion-corrupted image.

9. A method according to claim 8, further comprising:

inputting an image to the trained regression network;

receiving a second motion score output by the trained regression network;

comparing the second motion score to a threshold motion score; and

determining whether the image is of acceptable quality based on the comparison between the second motion score and the threshold motion score.

10. A method according to claim 9, wherein determining the threshold motion score comprises:

inputting a plurality of images indicated as acceptable and a plurality of images indicated as unacceptable to the trained regression network;

determining a motion score for each of the plurality of images indicated as acceptable and the plurality of images indicated as unacceptable; and

determining the threshold motion score based on the motion scores for each of the plurality of images indicated as acceptable and the motion scores for each of the plurality of images indicated as unacceptable.

11. A method according to claim 8, wherein generating the one or more motion-corrupted images comprises:

determining k-space data of a first reference image;

phase-shifting the k-space data to introduce first motion in the k-space data; and

reconstructing a first motion-corrupted image from the phase-shifted k-space data.

12. A method according to claim 11, wherein generating the one or more motion-corrupted images comprises:

phase-shifting the k-space data to introduce second motion in the k-space data; and

reconstructing a second motion-corrupted image from the k-space data phase-shifted to introduce second motion.

13. A method according to claim 8, wherein inputting a generated motion-corrupted image to the regression network comprises:

segmenting the input image into two or more segmented images; and

inputting the segmented images into the regression network.

14. A system comprising:

a memory storing processor-executable program code; and

a processor to execute the processor-executable program code to cause the system to;

train a generative network to learn a feature distribution of reference images;

input a plurality of acceptable images and a plurality of motion corrupted images to the trained generative network to generate a respective plurality of output images constrained to the feature distribution of acceptable reference images; and

train a discriminator network to discriminate between reference images and unacceptable images based on differences between each of the reference images and the plurality of motion corrupted images and a respective one of the plurality of output images.

15. A system according to claim 14, wherein training of the generative network to learn a feature distribution of reference images comprises:

inputting a first reference image to the generative network;

receiving a first output image from the generative network; and

determining a loss based on a comparison between the first reference image and the first output image.

16. A system according to claim 15, the generative network comprises an encoder-decoder.

17. A system according to claim 14, wherein training of the discriminator network comprises:

inputting a first reference image to the generative network;

receiving a first output image from the generative network;

determining a first residual based on a difference between the first reference image and the first output image;

inputting the first residual to the discriminator network;

receiving a first quality label from the discriminator network; and

determining a first loss based on whether or not the first quality label indicates a a reference image.

18. A system according to claim 17, wherein training of the discriminator network comprises:

inputting a first motion corrupted image to the generative network;

receiving a second output image from the generative network;

determining a second residual based on a difference between the first motion corrupted image and the second output image;

inputting the second residual to the discriminator network;

receiving a second quality label from the discriminator network; and

determining a second loss based on whether or not the second quality label indicates a motion corrupted image.