Efficient kernel density estimation of shape and intensity priors for level set segmentation
View Patent ↗Methods and systems for image segmentation are disclosed. A nonlinear statistical shape model of an image is integrated with a non-parametric intensity model to estimate characteristics of an image and create segmentations of an image based on Bayesian inference from characteristics of prior learned images based on the same models.
1. A method of segmenting an object in a set of image data using one or more prior instances of the object, comprising:
determining a nonparametric estimate of a statistical shape distribution from the one or more prior instances of the object in a subspace spanned by the one or more prior instances of the object by a kernel density estimator;
determining a nonparametric estimate of a statistical intensity distribution from one or more prior instances of the object by a kernel density estimator;
combining the kernel density estimator of the statistical shape distribution with the kernel density estimator of the statistical intensity distribution in a Bayesian expression conditioned on the set of image data, wherein the expression is provided in accordance with:
E
(
α
,
h
,
θ
)
=
-
∫
Ω
(
H
ϕ
log
p
in
(
I
)
+
(
1
-
H
ϕ
)
log
p
out
(
I
)
)
ⅆ
x
-
log
(
1
N
σ
∑
i
=
1
N
K
(
α
-
α
i
σ
)
)
;
selecting a segmentation of the object in the set of image data by executing a level set method by a processor which optimizes the Bayesian expression; and
the processor generating an image of the segmented object on a display.
2. The method of claim 1 , wherein a shape vector of the one or more prior instances of the object is based on an average value of a shape vector associated with each of the one or more prior instances of the object.
3. The method of claim 1 , wherein a shape vector of the one or more prior instances of the object corresponds to a signed distance function.
4. The method of claim 1 , wherein the expression is an energy function.
5. The method of claim 1 , wherein the statistical shape distribution is translation and rotation invariant.
6. The method of claim 1 , wherein the kernel density estimator related to the nonparametric estimate of the statistical intensity distribution is a smoothed intensity histogram.
7. The method as claimed in claim 1 , wherein the set of image data is ultra-sound image data.
8. The method as claimed in claim 1 , wherein the set of image data is CT image data.
9. The method as claimed in claim 1 , wherein the object is a prostate.
10. A system for segmenting an object in a set of image data using one or more prior instances of the object, comprising:
a processor;
computer software operable on the processor, the computer software being capable of:
determining a nonparametric estimate of a statistical shape distribution from the one or more prior instances of the object in a subspace spanned by the one or more prior instances of the object by a kernel density estimator;
determining a nonparametric estimate of a statistical intensity distribution from the one or more prior instances of the object by a kernel density estimator;
combining the kernel density estimator of the statistical shape distribution with the kernel density estimator of the statistical intensity distribution in a Bayesian expression conditioned on the set of image data, wherein the expression is provided in accordance with:
E
(
α
,
h
,
θ
)
=
-
∫
Ω
(
H
ϕ
log
p
in
(
I
)
+
(
1
-
H
ϕ
)
log
p
out
(
I
)
)
ⅆ
x
-
log
(
1
N
σ
∑
i
=
1
N
K
(
α
-
α
i
σ
)
)
;
and
selecting a segmentation of the object in the set of image data by executing a level set method which optimizes the Bayesian expression.
11. The system of claim 10 , wherein a shape vector of the one or more prior instances of the object is based on an average value of a shape vector associated with each of the one or more prior instances of the object.
12. The system of claim 10 , wherein a shape vector of the one or more prior instances of the object corresponds to a signed distance function.
13. The system of claim 10 , wherein the expression is an energy function.
14. The system of claim 10 , wherein the statistical shape distribution is translation and rotation invariant.
15. The system of claim 10 , wherein the kernel density estimator related to the nonparametric estimate of the statistical intensity distribution is a smoothed intensity histogram.
16. The system as claimed in claim 10 , wherein the set of image data is ultra-sound image data.
17. The system as claimed in claim 10 , wherein the set of image data is CT image data.
18. The system as claimed in claim 10 , wherein the object is a prostate.
19. The system as claimed in claim 10 , wherein the set of image data is image data of a heart.
20. A system for segmenting an object in a set of image data using one or more prior instances of the object, comprising:
a processor;
computer software operable on the processor, the computer software being capable of:
determining a nonparametric estimate of a statistical shape distribution from the one or more prior instances of the object in a subspace spanned by the one or more prior instances of the object;
imposing the nonparametric estimate of the statistical shape distribution in a Bayesian expression conditioned on the set of image data, wherein the expression is provided as an energy expression in accordance with:
E
(
α
,
h
,
θ
)
=
-
∫
Ω
(
H
ϕ
log
p
in
(
I
)
+
(
1
-
H
ϕ
)
log
p
out
(
I
)
)
ⅆ
x
-
log
(
1
N
σ
∑
i
=
1
N
K
(
α
-
α
i
σ
)
)
;
and
selecting a segmentation of the object in the set of image data by executing a level set method which optimizes the Bayesian expression.