Identifying image artifacts by means of machine learning

Patent metadata

Field	Value
Publication number	US-11010897-B2
Application number	US-201816142076-A
Country	US
Kind code	B2
Filing date	Sep 26, 2018
Priority date	Oct 2, 2017
Publication date	May 18, 2021
Grant date	May 18, 2021

Abstract

A method is for producing an identification unit for identifying image artifacts automatically. In an embodiment, the method includes providing a learning processing apparatus; providing an initial identification unit; providing a first image data library including artifact reference acquisitions containing image artifacts; and training the identification unit using the image artifacts. An identification method is for identifying image artifacts automatically in an image acquisition. In an embodiment, the identification method includes: providing a trained identification unit; providing an image acquisition produced via a medical imaging system; inspecting the image acquisition for image artifacts by the identification unit; and labeling the ascertained image artifacts.

First claim

What is claimed is:

1. A method for producing a trained identification unit for identifying image artifacts in an image acquisition via a medical imaging system, the method comprising:
training, by a learning processing apparatus, an initial identification unit according to a machine learning principle to obtain a trained identification unit, the learning processing apparatus configured to recognize graphical elements in image acquisitions, the training being based on recognizing image artifacts of artifact reference acquisitions included in a first image data library, the artifact reference acquisitions being produced via a medical imaging system; and
generating a new image in response to identifying, by the trained identification unit, an image artifact in an acquired image.

2. The method of claim 1 , wherein
the training the initial identification unit is additionally based on recognizing image components of clear reference acquisitions included in a second image data library, the clear reference acquisitions being produced via a medical imaging system, and the clear reference acquisitions including substantially no image artifacts; and
image components contained in the artifact reference acquisitions and not in the clear reference acquisitions are tagged as image artifacts.

3. The method of claim 2 , wherein the artifact reference acquisitions of the first image data library and the clear reference acquisitions of the second image data library include at least one of:
acquisitions of an object at two different acquisition energies; or
acquisitions decomposed according to material.

4. The method of claim 2 , wherein the artifact reference acquisitions of the first image data library and the clear reference acquisitions of the second image data library include labeled image objects, the artifact reference acquisitions including image artifacts labeled as artifact objects, and the clear reference acquisitions including image objects having no artifacts labeled as non-artifact objects.

5. The method of claim 1 , further comprising:
creating a first base-element library from a set of the artifact reference acquisitions, the first base-element library including artifact objects as base elements; and
creating a second base-element library using clear reference acquisitions, the second base-element library including non-artifact objects as base elements.

6. An identification unit for identifying image artifacts in an image acquisition from a medical imaging system, produced by the method of claim 1 .

7. A learning processing apparatus, comprising:
a data storage device storing computer-readable instructions; and
a processor configured to execute the computer-readable instructions to capture reference acquisitions,
recognize image objects in the reference acquisitions,
identify image artifacts in the image objects as artifact objects, and
train an identification unit according to a machine learning principle based on recognizing the image artifacts of the reference acquisitions.

8. An identification method for identifying image artifacts in an image acquisition produced via a medical imaging system, the identification method comprising:
training an identification unit based on recognizing image artifacts of artifact reference acquisitions included in a first image data library;
inspecting, via the identification unit, an image acquisition for image artifacts, the image acquisition being produced via a medical imaging system; and
processing the image acquisition to generate a new image, the processing including at least one of
labeling the image artifacts ascertained based on the inspecting,
reducing the image artifacts ascertained based on the inspecting, or
eliminating all of the image artifacts ascertained based on the inspecting.

9. The identification method of claim 8 , wherein the processing comprises generating at least one combination image based on a combination of first base elements of a first base-element library, the first base elements being artifact objects.

10. The identification method of claim 9 , wherein the at least one combination image includes at least one of: a calcium image, an iodine image, a virtual non-contrast image, a water image, a soft-tissue image, an iron image, a contrast-agent image or a mono-energy image.

11. The identification method of claim 8 , further comprising:
providing a control data library including control datasets for a medical imaging system, the control datasets corresponding to artifact objects; and
selecting a control dataset from among the control datasets according to at least one identified image artifact ascertained in the image acquisition,
wherein the processing includes using the control dataset for re-acquisition of a subject of the image acquisition.

12. A control device, for controlling a medical imaging system, comprising:
at least one processor configured to execute computer-readable instructions to inspect an acquired image for image artifacts;
process the acquired image by at least one of
labeling the image artifacts ascertained based on the inspection,
reducing the image artifacts ascertained based on the inspection, and
eliminating all of the image artifacts ascertained based on the inspection; and

generate at least one combination image based on the acquired image as processed and a combination of base elements of a base-element library, the base elements being artifact objects.

13. A medical imaging system, comprising:
the control device of claim 12 .

14. A non-transitory computer-readable medium storing a computer program, the computer program including program segments that, when executed by at least one processor, cause the at least one processor to perform the identification method of claim 8 .

15. A non-transitory computer-readable medium, storing program segments that, when executed by at least one processor, cause the at least one processor to perform the method of claim 1 .

16. The method of claim 2 , wherein the generating the new image includes tagging image components contained in the clear reference acquisitions and not in the artifact reference acquisitions as artifact-free image objects.

17. The method of claim 1 , wherein the artifact reference acquisitions of the first image data library include at least one of:
acquisitions of an object at two different acquisition energies; or
acquisitions decomposed according to material.

18. The method of claim 1 , wherein the artifact reference acquisitions of the first image data library include reference image artifacts labeled as artifact objects.

19. The method of claim 1 , further comprising:
creating a first base-element library from a set of artifact reference acquisitions, the first base-element library including artifact objects as base elements.

20. The identification method of claim 9 , wherein the generating that at least one combination image generates the at least one combination image based on a combination of the first base elements and second base elements of a second base-element library, the second base elements being non-artifact objects.

21. The identification method of claim 9 , wherein the at least one combination image includes a first combination image composed of the image artifacts and a complementary combination image containing fewer artifacts than the first combination image.

22. The identification method of claim 20 , wherein the at least one combination image includes a first co

Assignees

• Siemens Healthcare Gmbh

Inventors

• Flohr Thomas
• Schmidt Bernhard

Classifications

• G06V10/774 Primary

  Generating sets of training patterns; Bootstrap methods, e.g. bagging or boosting · CPC title

• G06T7/0014 Primary

  using an image reference approach · CPC title

• G06F18/214

  Generating training patterns; Bootstrap methods, e.g. bagging or boosting · CPC title

• G06V2201/031

  of internal organs · CPC title

• G06T2207/30061

  Lung · CPC title