Alternative noise map estimation methods for CT images
View Patent ↗Noise map for CT images have been estimated by generalizing the prior art even-and-odd views approach. One example is to estimate a noise map from images reconstructed from three sets of independent views. A second example is to estimate a noise map from images reconstructed by using two sets of correlated views. A third example is to estimate a noise map from noise map from two images reconstructed from two sets of independent views while the number of views in each set is unequal. Physical phantom data were employed to validate our proposed noise map estimation methods. In comparison to the existing method, our alternative methods yield reasonably accurate noise map estimation.
1. A method of estimating a noise map from reconstructed images, comprising:
acquiring data for a predetermined first number of views;
grouping the views into two groups with some of the views that are correlated between the two groups, one of the two groups having a predetermined second number of the views, the other one of the two groups having a predetermined third number of the views, either one of the predetermined second number and the predetermined third number being less than the predetermined first number of the views;
reconstructing two images from the two groups of the views; and estimating variance based upon the two reconstructed images and a weight.
2. The method of estimating a noise map from reconstructed images according to claim 1 wherein the other one of the predetermined second number and the predetermined third number equals to the predetermined first number of the views.
3. The method of estimating a noise map from reconstructed images according to claim 1 wherein the variance is determined according to
Var{ F}=n/m−n Var{ F−X}
where F is a first reconstructed image based upon the predetermined first number m of the views, X is a second reconstructed image based upon the predetermined second number n of the views that is less than the predetermined first number m.
4. A method of estimating a noise map from reconstructed images, comprising:
acquiring data for a predetermined first number of views;
grouping the views into a predetermined second number of at least two groups with an unequal number of the views that are independent in each of the groups;
reconstructing an image for each of the groups; and
estimating variance based upon a sum of variance in intermediate images each generated by a difference between all combinations of two of the reconstructed images.
5. The method of estimating a noise map from reconstructed images according to claim 4 wherein the predetermined second number of the groups is at least three.
6. The method of estimating a noise map from reconstructed images according to claim 4 wherein the predetermined second number of the groups is two while the predetermined first number of views is m.
7. The method of estimating a noise map from reconstructed images according to claim 6 wherein the variance is determined according to
Var{ F }=( m−n ) n/m 2Var{ Y−X}
where F is a reconstructed image based upon two reconstructed images X and Y, which are respectively reconstructed from n images and m-n images, the n being smaller than the m.
8. A system for estimating a noise map from reconstructed images, comprising:
a scanner for acquiring data for a predetermined first number of views;
an image data processor connected to said scanner for grouping the views into two groups with some of the views that are correlated between the two groups, one of
the two groups having a predetermined second number of the views, the other one of the two groups having a predetermined third number of the views, either one of the predetermined second number and the predetermined third number being less than the predetermined first number of the views;
an image reconstructing unit connected to said image data processor for reconstructing two images from the two groups of the views; and
a variance estimation unit connected to said image reconstructing unit for estimating variance based upon the two reconstructed images and a weight.
9. The system for estimating a noise map from reconstructed images according to claim 8 wherein the other one of the predetermined second number and the predetermined third number equals to the predetermined first number of the views.
10. The system for estimating a noise map from reconstructed images according to claim 8 wherein the variance is determined according to
Var{ F}=n/m−n Var{ F−X}
where F is a first reconstructed image based upon the predetermined first number m of the views, X is a second reconstructed image based upon the predetermined second number n of the views that is less than the predetermined first number m.
11. A system for estimating a noise map from reconstructed images, comprising:
a scanner for acquiring data for a predetermined first number of views;
an image data processor connected to said scanner for grouping the views into a predetermined second number of at least two groups with an unequal number of the views that are independent in each of the groups;
an image reconstructing unit connected to said image data processor for reconstructing an image for each of the groups; and
a variance estimation unit connected to said image reconstructing unit for estimating variance based upon a sum of variance in intermediate images each generated by a difference between all combinations of two of the reconstructed images.
12. The system for estimating a noise map from reconstructed images according to claim 11 wherein the predetermined second number of the groups is at least three.
13. The system for estimating a noise map from reconstructed images according to claim 11 wherein the predetermined second number of the groups is two while the predetermined first number of views is m.
14. The system for estimating a noise map from reconstructed images according to claim 13 wherein the variance is determined according to
Var{ F }=( m−n ) n/m 2Var{Y−X}
where F is a reconstructed image based upon two reconstructed images X and Y, which are respectively reconstructed from n images and m−n images, the n being smaller than the m.