Parameter-free denoising of complex MR images by iterative multi-wavelet thresholding

Patent metadata

Field	Value
Publication number	US-9569843-B1
Application number	US-201514849391-A
Country	US
Kind code	B1
Filing date	Sep 9, 2015
Priority date	Sep 9, 2015
Publication date	Feb 14, 2017
Grant date	Feb 14, 2017

Abstract

A method for denoising Magnetic Resonance Imaging (MRI) data includes receiving a noisy image acquired using an MRI imaging device and determining a noise model comprising a non-diagonal covariance matrix based on the noisy image and calibration characteristics of the MRI imaging device. The noisy image is designated as the current best image. Then, an iterative denoising process is performed to remove noise from the noisy image. Each iteration of the iterative denoising process comprises (i) applying a bank of heterogeneous denoisers to the current best image to generate a plurality of filter outputs, (ii) creating an image matrix comprising the noisy image, the current best image, and the plurality of filter outputs, (iii) finding a linear combination of elements of the image matrix which minimizes a Stein Unbiased Risk Estimation (SURE) value for the linear combination and the noise model, (iv) designating the linear combination as the current best image, and (v) updating each respective denoiser in the bank of heterogeneous denoisers based on the SURE value. Following the iterative denoising process, the current best image is designated as a final denoised image.

First claim

We claim:

1. A method for denoising Magnetic Resonance Imaging (MRI) data, the method comprising:
receiving a noisy image acquired using an MRI imaging device;
determining a noise model comprising a non-diagonal covariance matrix based on the noisy image and calibration characteristics of the MRI imaging device;
designating the noisy image as a current best image;
performing an iterative denoising process to remove noise from the noisy image, each iteration of the iterative denoising process comprising:
applying a bank of heterogeneous denoisers to the current best image to generate a plurality of filter outputs,
creating an image matrix comprising the noisy image, the current best image, and the plurality of filter outputs,
finding a linear combination of elements of the image matrix which minimizes a Stein Unbiased Risk Estimation (SURE) value for the linear combination and the noise model,
designating the linear combination as the current best image,
updating each respective denoiser in the bank of heterogeneous denoisers based on the SURE value; and

following the iterative denoising process, designating the current best image as a final denoised image.

2. The method of claim 1 , wherein each respective denoiser in the bank of heterogeneous denoisers corresponds to a wavelet filter.

3. The method of claim 2 , wherein each respective denoiser in the bank of heterogeneous denoisers corresponds to a distinct Daubechies wavelet.

4. The method of claim 2 , wherein the iterative denoising process further comprising:
performing a sequential cycle spinning operation to shift transforms performed by each respective wavelet filter by a predetermined offset value.

5. The method of claim 4 , wherein the predetermined offset value varies during each iteration of the iterative denoising process.

6. The method of claim 1 , wherein each respective denoiser in the bank of heterogeneous denoisers uses a thresholding operator corresponding to an empirical Wiener filter.

7. The method of claim 6 , wherein the thresholding operator applies non-negative garrote thresholding.

8. The method of claim 6 , wherein updating each respective denoiser in the bank of heterogeneous denoisers based on the SURE value comprises:
updating the thresholding operator used by each respective denoiser based on the SURE value.

9. The method of claim 1 , further comprising:
acquiring k-space data using the MRI imaging device;
determining a spatial modulation measurement operator corresponding to the MRI imaging device based on the k-space data;
applying one or more filters to the k-space data to yield filtered k-space data and a spectral measurement operator;
reconstructing the noisy image based on the filtered k-space data;
generating the noise model based on the spatial modulation measurement operator and the spectral measurement operator.

10. The method of claim 1 , wherein one or more denoisers included in the bank of heterogeneous denoisers is configured based on the noise model.

11. The method of claim 1 , wherein the iterative denoising process further comprises:
selecting a subset of the plurality of filter outputs, wherein each respective output in the subset corresponds to a recombination weight in the linear combination below a predetermined threshold value;
identifying one or more low-weighted denoisers in the bank of heterogeneous denoisers corresponding to the plurality of filter outputs; and
removing the one or more low-weighted denoisers from the bank of heterogeneous denoisers.

12. An article of manufacture for denoising Magnetic Resonance Imaging (MRI) data, the article of manufacture comprising a non-transitory, tangible computer-readable medium holding computer-executable instructions for performing a method comprising:
receiving a noisy image acquired using an MRI imaging device;
determining a noise model comprising a non-diagonal covariance matrix based on the noisy image and calibration characteristics of the MRI imaging device;
designating the noisy image as a current best image;
performing an iterative denoising process to remove noise from the noisy image, each iteration of the iterative denoising process comprising:
applying a bank of heterogeneous denoisers to the current best image to generate a plurality of filter outputs,
creating an image matrix comprising the noisy image, the current best image, and the plurality of filter outputs,
finding a linear combination of elements of the image matrix which minimizes a Stein Unbiased Risk Estimation (SURE) value for the linear combination and the noise model,
designating the linear combination as the current best image,
updating each respective denoiser in the bank of heterogeneous denoisers based on the SURE value; and

following the iterative denoising process, designating the current best image as a final denoised image.

13. The article of manufacture of claim 12 , wherein each respective denoiser in the bank of heterogeneous denoisers corresponds to a wavelet filter.

14. The article of manufacture of claim 13 , wherein the iterative denoising process further comprising:
performing a sequential cycle spinning operation to shift transforms performed by each respective wavelet filter by a predetermined offset value that varies during each iteration of the iterative denoising process.

15. The article of manufacture of claim 12 , wherein each respective denoiser in the bank of heterogeneous denoisers uses a thresholding operator corresponding to an empirical Wiener filter.

16. The article of manufacture of claim 15 , wherein the thresholding operator applies non-negative garrote thresholding.

17. The article of manufacture of claim 15 , wherein updating each respective denoiser in the bank of heterogeneous denoisers based on the SURE value comprises:
updating the thresholding operator used by each respective denoiser based on the SURE value.

18. The article of manufacture of claim 12 , wherein the method further comprises:
receiving k-space data acquired using the MRI imaging device;
determining a spatial modulation measurement operator corresponding to the MRI imaging device based on the k-space data;
applying one or more filters to the k-space data to yield filtered k-space data and a spectral measurement operator;
reconstructing the noisy image based on the filtered k-space data;
generating the noise model based on the spatial modulation measurement operator and the spectral measurement operator.

19. The article of manufacture of claim 12 , wherein one or more denoisers included in the bank of heterogeneous denoisers is configured based on the noise model.

20. A system for denoising Magnetic Resonance Imaging (MRI) data, the system comprising:
an MRI imaging device configured to acquire k-space data;
a computer comprising an image data processor configured to:
reconstruct a noisy image based on the k-space data,
determine a noise model comprising a non-diagonal covariance matrix based on the noisy image and calibration characteristics of the MRI imaging device;
designate the noisy image as a current best image;
perform an iterative denoising process to remove noise from the noisy image, each iteration of the iterative denoising process comprising:
(i) applying a bank of heterogeneous denoisers to the current best image to generate a plurality of filter outputs,
(ii) creating an image matrix comprising the noisy image, the current best image, and the plurality of filter outputs,
(iii) finding a linear combination o

Assignees

• Siemens Healthcare Gmbh

Inventors

• Mailhe Boris
• Nadar Mariappan S
• Kannengiesser Stephan

Classifications

• G06T5/10 Primary

  using non-spatial domain filtering · CPC title

• G06T5/20

  using local operators · CPC title

• G06T2207/10088

  Magnetic resonance imaging [MRI] · CPC title

• G01R33/5611

  Parallel magnetic resonance imaging, e.g. sensitivity encoding [SENSE], simultaneous acquisition of spatial harmonics [SMASH], unaliasing by Fourier encoding of the overlaps using the temporal dimension [UNFOLD], k-t-broad-use linear acquisition speed-up technique [k-t-BLAST], k-t-SENSE (structural details of arrays of sub-coils G01R33/3415) · CPC title

• G06T7/0012 Primary

  Biomedical image inspection · CPC title