Denoising medical images

• US 8,605,970 B2
• Filed: 12/24/2009
• Issued: 12/10/2013
• Est. Priority Date: 12/25/2008
• Status: Active Grant

• Current Assignee: Medic Vision Imaging Solutions Ltd.
• Original Assignee: Medic Vision-Imaging Solutions Limited
• Inventors: Bar-Aviv, Ezer, Devir, Zvi, Dahan, Eliran, Kenig, Tal, Rosman, Guy
• Primary Examiner(s): Tabatabai, Abolfazl
• Application Number:
  US13/142,282
• Publication Number:
  US 20110268328A1
• Time in Patent Office: 1,447 Days
• Field of Search: 382128-134, 378/4, 378/8, 378 21- 27, 378/101, 378/901, 600/407, 600/410, 600/425, 600/427, 600/336, 600/411, 128/920, 128/922
• US Class Current: 382/128
• CPC Class Codes: G06T 2207/10081 Computed x-ray tomography [CT]
• CPC Class Codes: G06T 2207/20016 Hierarchical, coarse-to-fine, multiscale or multiresolution image processing; Pyramid transform
• CPC Class Codes: G06T 2207/20036 Morphological image processing
• CPC Class Codes: G06T 2207/20192 Edge enhancement; Edge preservation
• CPC Class Codes: G06T 2207/30004 Biomedical image processing
• CPC Class Codes: G06T 5/50 using two or more images, e.g. averaging or subtraction
• CPC Class Codes: G06T 5/70 Denoising; Smoothing
• CPC Class Codes: G06T 5/77 Retouching; Inpainting; Scratch removal

Abstract

A method of producing a denoised medical image at a specified resolution, comprising: a) denoising an original medical image acquired at a resolution higher than the specified resolution; and b) reducing the resolution of the image, after denoising, to the specified resolution, using a nonlinear resolution reduction procedure that preserves more fine detail in the image than a linear resolution reduction procedure would.

Claims

1. A method of producing a denoised medical image at a specified resolution, comprising:

a) denoising an original medical image acquired at a resolution higher than the specified resolution; and

b) reducing the resolution of the original medical image, after denoising, to the specified resolution, using a nonlinear resolution reduction procedure including the steps of;

(i) decomposing the thinner slices into band-pass components;

(ii) for at least some of the band-pass components, combining the thinner slice components into thicker slice components in a way that, at least sometimes, gives more weight to a thinner slice component for which the component has a higher amplitude; and

(iii) re-assembling the thicker slice band-pass components for each thicker slice, to produce the image at the specified resolution

wherein the original medical image is a three-dimensional image including thinner slices of an original thickness, and reducing the resolution includes fusing multiple thinner slices using the nonlinear resolution reduction procedure to obtain an image with thicker slices.

2. A method according to claim 1, wherein combining the thinner slice components into thicker slice components in a way that gives more weight to a thinner slice component for which the component has a higher amplitude, is done at least for band-pass components other than the component containing the lowest frequency content, when, for that band-pass component and among all of the thinner slices being combined into the same thicker slice, a thinner slice component with a greatest absolute value has an absolute value that is below a threshold value.

3. A method according to claim 1, wherein the original medical image is a three-dimensional image comprising slices arranged in a slice direction, and reducing the resolution comprises reducing the resolution in the slice direction, but not reducing the resolution in directions along each slice.

4. A method according to claim 1, wherein the nonlinear resolution reduction procedure comprises a band-pass decomposition procedure.

5. A method according to claim 1, wherein the band-pass decomposition procedure comprises a Laplacian pyramid procedure.

6. A method according to claim 1, wherein the band-pass decomposition procedure comprises a wavelet transform procedure.

7. A method according to claim 1, wherein denoising the original medical image comprises:

a) for each pixel being examined in the image, selecting a set of search pixels;

b) calculating a value of at least one feature of a neighborhood of each search pixel, and a value of corresponding features of a neighborhood of the pixel being examined;

c) calculating a value of at least one other feature of the same or a different neighborhood of each search pixel, and a value of corresponding feature of the same or a different neighborhood of the pixel being examined; and

d) calculating a reduced noise grey value for each pixel being examined, based on raw or transformed grey values of its search pixels, with greater sensitivity to those with one or more feature values similar to those of the pixel being examined.

8. A system for producing a denoised medical image at a specified resolution, comprising:

a) a medical imaging device for acquiring raw data for a 3-D medical image, capable of achieving at least a higher resolution than the specified resolution;

b) a controller programmed to reconstruct slices of the image from the raw data, at the higher resolution, to denoise the image, and to reduce the resolution of the denoised image to the specified resolution by fusing thinner slices to form thicker slices, using a nonlinear procedure that comprises;

i) decomposing the thinner slices into band-pass components;

ii) for at least some of the band-pass components, combining the thinner slice components into thicker slice components in a way that, at least sometimes, gives more weight to a thinner slice component for which the component has a higher amplitude; and

iii) re-assembling the thicker slice band-pass components for each thicker slice, to produce the image at the specified resolution.

9. A method of restoring removed structures to a denoised medical image, comprising:

a) denoising an original medical image to produce the denoised image;

b) finding a residual image that depends on differences between the denoised and original images;

c) applying a nonlocal smoothing filter to the residual image to obtain a filtered residual image which indicates a degree of structure as a function of position in the residual image; and

d) restoring structure by changing a grey value of each voxel in the denoised image by an amount which depends on the degree of structure at a corresponding location in the residual image, and on a grey value at a corresponding location in the residual image or the filtered residual image.

10. A method according to claim 9, wherein restoring structure comprises adding to the denoised image the weighted or unweighted filtered residual image.

11. A method according to claim 9, also including finding a measure of the degree of structure as a function of position in the residual image, from the filtered residual image, wherein restoring structure comprises adding to the denoised image the residual image or the filtered residual image weighted by an adaptive parameter that depends locally on the measure of the degree of structure as a function of position.

12. A method according to claim 11, wherein finding the measure of the degree of structure as a function of position in the residual image, comprises finding one or more eigenvalues of a structure tensor, a Hessian matrix, or both.

13. A method according to claim 9, wherein denoising the original medical image comprises:

a) for each pixel being examined in the image, selecting a set of search pixels;

b) calculating a value of at least one feature of a neighborhood of each search pixel, and a value of corresponding features of a neighborhood of the pixel being examined;

c) calculating a value of at least one other feature of the same or a different neighborhood of each search pixel, and a value of corresponding feature of the same or a different neighborhood of the pixel being examined; and

d) calculating a reduced noise grey value for each pixel being examined, based on raw or transformed grey values of its search pixels, with greater sensitivity to those with one or more feature values similar to those of the pixel being examined.

14. A method according to claim 9, wherein the nonlocal smoothing filter comprises a nonlinear edge-preserving filter.

15. A method according to claim 14, wherein the nonlinear edge-preserving filter comprises a nonlinear anisotropic diffusion filter.

16. A method according to claim 15, wherein the nonlinear anisotropic diffusion filter comprises a Beltrami flow filter.

17. A method according to claim 15, wherein the nonlinear anisotropic diffusion filter comprises a coherence enhancement diffusion (CED) filter.

18. A method according to claim 9, wherein, for most positions, the change in grey value is more sensitive to the grey value of the residual image or the filtered residual image or both, at positions where the degree of structure is greater, than it is at positions where the degree of structure is smaller.

19. A system for producing a denoised medical image with removed structure restored to it, comprising:

a) a medical imaging device for acquiring raw data for medical images;

b) a controller programmed to reconstruct a medical image from the raw data, to denoise the image, to find a residual image that depends on differences between the denoised and original images, to apply a nonlocal smoothing filter to the residual image to obtain a filtered residual image which indicates a degree of structure as a function of position in the residual image, and to restore structure by changing a grey value of each voxel in the denoised image by an amount which depends on the degree of structure at a corresponding location in the residual image, and on a grey value at a corresponding location in the residual image or the filtered residual image.

20. A method of producing a denoised medical image with a specified magnitude and distribution of noise, comprising:

a) removing noise from an original medical image to produce an initial denoised image; and

b) adding a portion of the removed noise to the initial denoised image to produced a denoised image with about the specified magnitude and spatial distribution of noise,

wherein adding a portion of the removed noise includes the steps of

(i) choosing a spatially varying weighting parameter which will achieve the specified magnitude and distribution of noise; and

(ii) finding a weighted sum of the original image and the initial denoised image, according to the weighting parameter.

21. A method according to claim 20, wherein choosing the spatially varying weighting parameter comprises calculating local standard deviations of the original image and the initial denoised image.

22. A method according to claim 20, wherein removing noise from the original medical image comprises restoring removed structures to the initial denoised image.

23. A method according to claim 20, wherein adding noise comprises adding the original image multiplied by a spatially varying parameter that depends on the specified magnitude and distribution of noise.

24. A system for producing a denoised medical image with a specified magnitude and distribution of noise, comprising:

a) a medical imaging device for acquiring raw data for medical images;

b) a controller programmed to reconstruct a medical image from the raw data, to denoise the image, and to add a portion of the removed noise to the initial denoised image to produced a denoised image with about the specified magnitude and spatial distribution of noise,

wherein said controller is further programmed to add a portion of the removed noise by;

(i) choosing a spatially varying weighting parameter which will achieve the specified magnitude and distribution of noise; and

(ii) finding a weighted sum of the original image and the initial denoised image, according to the weighting parameter.

25. A method of denoising a medical image, comprising:

a) for each of a plurality of selected voxels in the image, estimating a noise level in a large neighborhood surrounding the voxel, by;

i) for at least some voxels in the large neighborhood that fall within a chosen range of grey levels, finding a degree of variation in grey level in a small neighborhood around the voxel;

ii) finding a subset of the voxels for which the degree of variation in grey level is below a chosen quantile among the voxels for which the degree of variation was found; and

iii) setting the noise level equal to an average of the degrees of variation for the voxels in the subset;

b) finding noise level as function of position in the image, from the estimated noise levels for the selected voxels; and

c) denoising the image, using a denoising algorithm that makes use of the noise level as a function of position.

26. A method of denoising a medical image using a known noise level as a function of position in the image, comprising, for each voxel in the image:

a) choosing a set of comparison voxels for said voxel;

b) calculating a similarity measure between a neighborhood associated with said voxel and a neighborhood associated with each comparison voxel;

c) finding a weighted average of the grey levels of the comparison voxels, the weight for each comparison voxel depending on the similarity measure for its neighborhood, normalized to a value that depends on the noise level at the position of said voxel, the noise level at the position of the comparison voxel, or both; and

d) changing the grey level of said voxel according to the weighted average of the grey levels of the comparison voxels.

27. A method according to claim 26, wherein calculating the similarity measure comprises calculating a value of a feature of the neighborhood of said voxel and calculating a value of a corresponding feature of the neighborhood of the comparison voxel.