Robust and automatic adjustment of display window width and center for MR images

• US 5,995,644 A
• Filed: 06/30/1997
• Issued: 11/30/1999
• Est. Priority Date: 06/30/1997
• Status: Expired due to Term

• Current Assignee: Siemens Medical Solutions USA Incorporated (Siemens AG)
• Original Assignee: Siemens Corporate Research Incorporated (Siemens AG)
• Inventors: Fang, Ming, Lai, Shang-Hong
• Primary Examiner(s): Johns, Andrew W.
• Application Number:
  US08/885,080
• Time in Patent Office: 883 Days
• Field of Search: 382/168, 382/132, 382/261, 382/239, 382/274, 382/131, 382/156, 382/169
• US Class Current: 382/131
• CPC Class Codes: G06T 2207/10088 Magnetic resonance imaging [MRI]
• CPC Class Codes: G06T 2207/20064 Wavelet transform [DWT]
• CPC Class Codes: G06T 2207/20084 Artificial neural networks [ANN]
• CPC Class Codes: G06T 2207/20221 Image fusion; Image merging
• CPC Class Codes: G06T 5/40 using histogram techniques
• CPC Class Codes: G06T 5/50 using two or more images, e.g. averaging or subtraction
• CPC Class Codes: G06T 5/60 using machine learning, e.g. neural networks
• CPC Class Codes: G06T 5/90 Dynamic range modification of images or parts thereof
• CPC Class Codes: G06V 10/25 Determination of region of interest [ROI] or a volume of interest [VOI]

Abstract

A system for deriving final display parameters for a wide range of MR images consists of a feature generator utilizing both histogram and spatial information computed from an input MR image, a wavelet transform within the feature generator for compressing the size of the feature vector, a competitive layer based neural network for clustering MR images into different subclasses, a bi-modal linear estimation network and a radial bases function network based non-linear estimator for each subclass, as well as a data fusion system using estimates from both estimators to compute the final display parameters.

Claims

1. A system for the automatic adjustment of display window width and center for MR images comprising:

a feature generator for receiving an input MR image;

hierarchical neural networks having a classifier connected to said feature generator for clustering said input MR image into a plurality of subclasses and a bi-modal linear estimator for each of said subclasses, said hierarchical neural networks provide window width and center estimates; and

data fusion connected to said hierarchical neural networks for combining said window width and center estimates for providing final display parameters.

2. A system for the automatic adjustment of display window width and center for MR images as claimed in claim 1 wherein said feature generator comprises:

compute and subsample histograms for receiving said input MR image and for providing a subsampled histogram;

range of interest connected to said compute and subsample histograms for receiving said subsampled histogram;

re-sample and normalize histogram connected to said range of interest;

wavelet transform connected to said re-sample and normalize histogram;

KL transform connected to said re-sample and normalize histogram;

wavelet histogram features connected to said wavelet transform; and

,

KL histogram features connected to said KL transform.

3. A system for the automatic adjustment of display window width and center for MR images as claimed in claim 2 wherein said range of interest comprises:

background thresholding means for finding a background thresholding value from said subsampled histogram, for setting a portion of said subsampled histogram that is below said background thresholding value to zero and for providing a new histogram;

compute means for computing a mean μ and

a standard deviation σ

of said input MR image from said new histogram; and

,

range compute means for taking a range of interest as an interval between max(μ

-3σ

,0) and min (μ

+3σ

,Imax).

4. A system for the automatic adjustment of display window width and center for MR images as claimed in claim 2 wherein:

said re-sample and normalize histogram smoothes and re-samples a histogram by using a pre-determined number of bins.

5. A system for the automatic adjustment of display window width and center for MR images as claimed in claim 2 wherein said wavelet transform and said KL transform comprise:

compression means for compressing histograms of said input MR image to a smaller size of a feature vector so that said clustering neural network can execute and train.

6. A system for the automatic adjustment of display window width and center for MR images as claimed in claim 1 wherein said feature generator comprises:

histogram means; and

,

spatial intensity distribution means;

wherein said histogram means and said spatial intensity distribution means resolve conflicting situations where window display settings are different although histograms of said MR images are similar.

7. A system for the automatic adjustment of display window width and center for MR images as claimed in claim I wherein said hierarchical neural networks further comprise

a radial bases function estimator for each of said different subclasses connected to said classifier.

8. A system for the automatic adjustment of display window width and center for MR images as claimed in claim 7 wherein:

said classifier is competitive layer neural network based.

9. A system for the automatic adjustment of display window width and center for MR images as claimed in claim 7 wherein said bi-modal linear estimator and said radial bases function estimator comprise:

training means for making said bi-modal linear estimator and said radial bases function estimator adaptable to new kinds of MR images simply by training said bi-modal linear estimator and said radial bases function estimator with said new kinds of MR images.

10. A system for the automatic adjustment of display window width and center for MR images as claimed in claim 7 wherein said bi-modal linear estimator comprises:

robust and reasonable estimation means for providing an estimate for a wide range of images that may not be included in a training data set.

11. A system for the automatic adjustment of display window width and center for MR images as claimed in claim 7 wherein said radial bases function estimator comprises:

accurate estimation means for providing an estimate for known (trained) MR image types.

12. A system for the automatic adjustment of display window width and center for MR images as claimed in claim 7 wherein:

said data fusion uses estimates from each of said bi-modal linear estimators and each of said radial bases function estimators for providing final display parameters.

13. A method for automatically adjusting display window width and center for MR images comprising the steps of:

utilizing a feature generator to receive an input MR image;

clustering said input MR image into different subclasses with a clustering neural network;

estimating each of said different subclasses with a bi-modal linear estimator;

estimating each of said different subclasses with a radial bases function estimator;

providing window width and center estimates from said clustering and estimating steps; and

fusing data including said window width and center estimates from said bi-modal linear estimator and said radial bases function estimator to provide final display parameters.

14. A method for automatically adjusting display window width and center for MR images as claimed in claim 13 wherein utilizing a feature generator comprises the steps of:

computing a histogram from said input MR image;

subsampling said histogram;

determining a range of interest;

re-sampling a new histogram

normalizing said new histogram;

performing a wavelet transform;

performing a KL transform;

providing wavelet histogram features; and

,

providing KL histogram features.

15. A method for automatically adjusting display window width and center for MR images as claimed in claim 14 wherein determining a range of interest comprises the steps of:

finding a background thresholding value from a subsampled histogram;

setting a portion of said subsampled histogram that is below said background thresholding value to zero;

providing a new histogram;

computing a mean μ and

a standard deviation σ

of said input MR image from said new histogram; and

,

taking a range of interest as an interval between max(μ

-3σ

,0) and min(μ

+3σ

,Imax).

16. A method for automatically adjusting display window width and center for MR images as claimed in claim 14 wherein performing a wavelet transform and performing a KL transform comprises the step of:

compressing histograms of said input MR image to a smaller size of a feature vector so that said clustering neural network can execute and train efficiently.

17. A method for automatically adjusting display window width and center for MR images as claimed in claim 13 wherein utilizing a feature generator comprises the steps of:

deriving a histogram; and

,

performing spatial intensity distribution;

wherein deriving a histogram and performing spatial intensity distribution resolves conflicting situations where window display settings are different although histograms of said MR images are similar.

18. A method for automatically adjusting display window width and center for MR images as claimed in claim 13 wherein estimating each of said different subclasses with a bi-modal linear estimator and estimating each of said different subclasses with a radial bases function estimator comprises the step of:

making said bi-modal linear estimator and said radial bases function estimator adaptable to new kinds of MR images simply by training said bi-modal linear estimator and said radial bases function estimator with said new kinds of MR images.

19. A method for automatically adjusting display window width and center for MR images as claimed in claim 13 wherein estimating each of said different subclasses with a bi-modal linear estimator comprises the step of:

providing a robust and reasonable estimate for a wide range of images that may not be included in a training data set.

20. A method for automatically adjusting display window width and center for MR images as claimed in claim 13 wherein estimating each of said different subclasses with a radial bases function estimator comprises the step of:

providing an accurate estimate for known MR image types.

21. A method for automatically adjusting display window width and center for MR images as claimed in claim 13 wherein fusing data comprises the steps of:

using estimates from said bi-modal linear estimator; and

,

using estimates from said radial bases function estimator.