Supervised machine learning technique for reduction of radiation dose in computed tomography imaging

Patent metadata

Field	Value
Publication number	US-9332953-B2
Application number	US-201314423997-A
Country	US
Kind code	B2
Filing date	Aug 30, 2013
Priority date	Aug 31, 2012
Publication date	May 10, 2016
Grant date	May 10, 2016

Abstract

Substantial reduction of the radiation dose in computed tomography (CT) imaging is shown using a machine-learning dose-reduction technique. Techniques are provided that (1) enhance low-radiation dosage images, beyond just reducing noise, and (2) may be combined with other approaches, such as adaptive exposure techniques and iterative reconstruction, for radiation dose reduction.

First claim

What is claimed:

1. A method, comprising:
obtaining, from a scanner, first computed tomography (CT) image information for scanning of a first article at a first radiation dosage level,
wherein the first CT image information exhibits one of a first noise level or a first image signal contrast;
applying, in a processor, signal processing to the first CT image information, the applied signal processing based on a machine-learning model trained using CT training images, wherein the CT training images correspond to scanning of one or more second articles at each of the first radiation dosage level and a second radiation dosage level higher than the first radiation dosage level; and
producing, from the processor as a result of the applied signal processing, second CT image information that exhibits one of a second noise level lower than the first noise level or a second image signal contrast higher than the first image signal contrast.

2. The method of claim 1 , further comprising:
displaying an image derived from the second CT image information.

3. The method of claim 1 , wherein the applied signal processing comprises at least one of:
noise suppression applied to portions of the first CT image information corresponding to a plurality of one of a sub-region or a sub-volume, the noise suppression determined at least in part based on the one or more CT training images corresponding to scanning of second articles at the second radiation dosage level; and
edge enhancement applied to the portions of the first CT image information corresponding to the plurality of one of a sub-region or a sub-volume, the edge enhancement determined at least in part based on the one or more CT training images corresponding to scanning of second articles at the second radiation dosage level.

4. The method of claim 1 , wherein the machine-learning model comprises at least one of a linear-output artificial neural network (ANN) regression model, a support vector regression model, or a nonlinear Gaussian process regression model.

5. The method of claim 4 , wherein the machine-learning model is the linear-output ANN regression model and produces a linear function as an activation function for continuous mapping of the first CT image information to the second CT image information.

6. The method of claim 1 , wherein the applied signal processing is based on a machine-learning classification model.

7. The method of claim 6 , wherein the machine-learning classification model is at least one of a multilayer perceptron, a support vector machine, a linear discriminant analysis machine, or a quadratic discriminant analysis machine.

8. The method of claim 1 , wherein the second CT image information has a signal-to-noise ratio of at least twice a signal-to-noise ratio of the first CT image information.

9. The method of claim 1 , wherein the first radiation dosage level is 0.1 milliseverts (mSv) or less.

10. The method of claim 9 , wherein the second radiation dosage level is 3 mSv.

11. The method of claim 1 , wherein the second CT image information corresponds to an image quality obtained by scanning the first article at a radiation dosage level of between 0.5 and 1.0 milliseverts (mSv).

12. The method of claim 1 , wherein the first radiation dosage level corresponds to a radiation reduction of at least 90% over the second radiation dosage level.

13. The method of claim 1 , wherein the second CT image information exhibits an average signal-to-noise ratio (SNR) improvement of at least 10 decibels (dB) over the first CT image information.

14. The method of claim 1 , wherein an estimated effective dosage range corresponds to a body region scanned to obtain the CT image information, and wherein the first radiation dosage level is within a lowest quarter of the estimated effective dosage range.

15. The method of claim 1 , wherein an estimated effective dosage range corresponds to a body region scanned to obtain the CT image information, and wherein the first radiation dosage level is within a lowest 10% of the estimated effective dosage range.

16. The method of claim 1 , wherein an estimated effective dosage range corresponds to a body region scanned to obtain the CT image information, and wherein the first radiation dosage level is within a lowest 5% of the estimated effective dosage range.

17. The method of claim 1 , wherein an estimated effective dosage range corresponds to a body region scanned to obtain the CT image information, and wherein the first radiation dosage level is within a lowest 1% of the estimated effective dosage range.

18. An apparatus, comprising:
a memory configured to store first computed tomography (CT) image information obtained by scanning of a first article at a first radiation dosage level, wherein the first CT image information exhibits one of a first noise level or a first image signal contrast; and
a processor coupled to the memory and configured to apply signal processing to the first CT image information, the applied signal processing based on a machine-learning model trained using CT training images, wherein the CT training images correspond to scanning of one or more second articles at each of the first radiation dosage level and a second radiation dosage level higher than the first radiation dosage level,
wherein the applied signal processing produces second CT image information that exhibits one of a second noise level lower than the first noise level or a second image signal contrast higher than the first image signal contrast.

19. The apparatus of claim 18 , further comprising:
a display configured to display an image derived from the second CT image information.

20. The apparatus of claim 18 , wherein the applied signal processing comprises at least one of:
noise suppression applied to portions of the first CT image information corresponding to a plurality of one of a sub-region or a sub-volume, the noise suppression determined at least in part based on the one or more CT training images corresponding to scanning of second articles at the second radiation dosage level; and
edge enhancement applied to the portions of the first CT image information corresponding to the plurality of one of a sub-region or a sub-volume, the edge enhancement determined at least in part based on the one or more CT training images corresponding to scanning of second articles at the second radiation dosage level.

21. The apparatus of claim 18 , wherein the applied signal processing is based upon
a determination of portions of first training CT image information corresponding to a plurality of one of a sub-region or a sub-volume each scanned at a third radiation dosage and a determination of one of pixels or voxels from counterpart portions of second training CT image information scanned at a fourth radiation dosage level higher than the third radiation dosage level, and
a correlation of the determined portions of the first training CT image information with the determined portions of the second training CT image information and an error analysis of the correlated, determined portions of the first and second training CT image information.

22. The apparatus of claim 21 , wherein the applied signal processing is based upon error correction parameters derived from the error analysis.

23. A method, comprising:
obtaining, from a scanner, first computed tomography (CT) image information for scanning of a first article at a first radiation dosage level, wherein the first CT image informa

Assignees

• Univ Chicago

Inventors

• Suzuki Kenji

Classifications

• G16H50/20

  for computer-aided diagnosis, e.g. based on medical expert systems · CPC title

• G06T12/30

  Image post-processing, e.g. metal artefact correction · CPC title

• A61B6/545

  involving automatic set-up of acquisition parameters · CPC title

• G06T3/4046

  using neural networks · CPC title

• A61B6/488

  involving pre-scan acquisition · CPC title