Combined segmentation and registration framework for parametric shapes
View Patent ↗Progress monitoring of lesions is done automatically by segmentation and registration of lesions in multi-phase medical images. A parametric level-set framework includes a model optimization for any number of lesions. The user specifies lesions in a baseline volume by clicking inside of them. The apparatus segments the lesions automatically in the baseline and follow-up volumes. The segmentation optimization compensates for lesion motion between baseline and follow-up volumes. 2D and 3D medical patient data can be processed by the methods.
1. A method for segmentation by a processor of N objects with N≧2 in a follow-up image I′ corresponding to a baseline image I containing N objects corresponding to the N objects in follow-up image I′, and simultaneously registering the N objects in follow-up image I′ to a position of the N corresponding objects in baseline image I, an image being represented by image data, comprising:
optimizing segmentation and registration of the N objects in image I′ by evolving a level set function maximizing a conditional probability expression that is a function of one or more contour parameters related to each object in image I′ under a condition of one or more contour parameters related to each object in image I; and
generating by the processor of the follow-up image for display with segmented and registered objects identified.
2. The method as claimed in claim 1 , wherein optimizing segmentation is achieved by adjusting a parametric level set function.
3. The method as claimed in claim 1 , wherein image data is generated by CT imaging and the objects are lesions.
4. The method as claimed in claim 1 , wherein the baseline image includes an image of a liver.
5. The method as claimed in claim 1 , further comprising:
marking each object i in image I;
generating and optimizing a segmentation contour for each object i of N objects in baseline image I, the contour of each object in I being determined by one or more contour parameters γ i ;
generating a contour for each object i′ of N objects in image I′, the contour of each object I′ being determined by one or more contour parameters γ i ′;
optimizing segmentation and registration of the i′ of N objects in image I′ by maximizing the conditional probability expression:
P(Γ′|Γ,I′)∝P(I′|Γ,Γ′)P(Γ|Γ′)P(Γ′), wherein:
Γ=(γ 1 , . . . ,γ N ) represents the one or more contour parameters of each of the N objects in the image I; and
Γ′=(γ′ 1 , . . . ,γ′ N ) represents the one or more parameters of each of the N objects in image I′; and
generating an image for display with segmented and registered objects identified.
6. The method as claimed in claim 1 , wherein the generating and optimizing a contour for an object i in baseline image I comprises:
optimizing the contour parameters γ i of the object i by maximizing the probability expression represented as P(γ|x 0 ,I)∝P(I|γ,x 0 )P(x 0 |γ)P(γ), wherein x 0 is a position inside the object i that was marked.
7. The method as claimed in claim 6 , wherein the optimizing of the contour for an object in baseline image I comprises the steps of:
initializing the contour parameters of the contour with a first set of parameters γ 0 ; and
evolving the contour by applying a gradient descent of a cost function.
8. The method as claimed in claim 7 , wherein the gradient descent of a cost function C is expressed as:
∂
C
∂
γ
(
γ
)
=
∑
x
∈
Ω
δ
(
ϕ
γ
(
x
)
)
(
P
out
(
I
(
x
)
)
-
P
i
n
(
I
(
x
)
)
)
∂
ϕ
γ
∂
γ
(
x
)
ⅆ
x
,
wherein
:
δ(x) is a Dirac delta function;
H(z) is a Heaviside step function;
φ γ denotes a level-set function representing a shape parameterized by a parameter γ;
P
i
n
(
z
)
=
∑
x
∈
Ω
H
(
ϕ
γ
(
x
)
)
δ
(
I
(
x
)
-
z
)
∑
x
∈
Ω
H
(
ϕ
γ
(
x
)
)
;
and
P
out
(
z
)
=
∑
x
∈
Ω
(
1
-
H
(
ϕ
γ
(
x
)
)
)
δ
(
I
(
x
)
-
z
)
∑
x
∈
Ω
(
1
-
H
(
ϕ
γ
(
x
)
)
)
.
9. The method as claimed in claim 5 , wherein generating the segmentation contour for each object i of N objects in baseline image I, comprises:
providing a click-point x 0 i to each object i of the N objects in image I to create N click-points X 0 =(x 0 1 , . . . ,x 0 N );
assigning a prior shape to each contour;
parameterizing a level-set function φ γ representing a shape, based on the prior shape; and
maximizing a conditional probability P(Γ|I,X 0 ).
10. The method as claimed in claim 9 , wherein the prior shape is an ellipsoid.
11. The method as claimed in claim 10 , further comprising optimizing segmentation and registration of the i′ of N objects in image I′ by finding a global rigid translation of all object shapes, and by letting each object translate itself locally.
12. The method as claimed in claim 11 , further comprising:
defining the level-set function as φ γ i (x):=φ(x,μ i +Δμ i +T,Σ i +ΔΣ i ) wherein Δμ i denotes a displacement of object i ellipsoid's mass center, T is a global rigid translation of the N objects, each object having an ellipsoid shape, and ΔΣ i describes a shape difference of object i in image I with corresponding object i′ in the follow-up image I′.
13. An apparatus for segmentation of N objects with N≧2 in a follow-up image I′ corresponding to a baseline image I containing N objects corresponding to the N objects in follow-up image I′, and simultaneously registering the N objects in follow-up image I′ to a position of the N corresponding objects in baseline image I, an image being represented by image data, comprising:
a memory;
a processor, retrieving instructions from the memory and executing the instructions to perform the steps of:
optimizing segmentation and registration of the N objects in image I′ by evolving a level set function maximizing a conditional probability expression that is a function of one or more contour parameters related to each object in image I′ under a condition of one or more contour parameters related to each object in image I; and
generating of the follow-up image for display with segmented and registered objects identified.
14. The apparatus as claimed in claim 13 , wherein optimizing segmentation is achieved by adjusting a parametric level set function.
15. The apparatus as claimed in claim 13 , wherein image data is generated by CT imaging and the objects are lesions.
16. The apparatus as claimed in claim 13 , wherein the baseline image includes an image of a liver.
17. The apparatus as claimed in claim 13 , further comprising instructions to perform the steps of:
marking each object i in image I;
generating and optimizing a segmentation contour for each object i of N objects in baseline image I, the contour of each object in I being determined by one or more contour parameters γ i ;
generating a contour for each object i′ of N objects in image I′, the contour of each object I′ being determined by one or more contour parameters γ′ i ;
optimizing segmentation and registration of the i′ of N objects in image I′ by maximizing the conditional probability expression:
P(Γ′|Γ,I′)∝P(I′|Γ,Γ′)P(Γ|Γ′)P(Γ′), wherein:
Γ=(γ 1 , . . . ,γ N ) represents the one or more contour parameters of each of the N objects in the image I; and
Γ′=(γ′ 1 , . . . ,γ′ N ) represents the one or more parameters of each of the N objects in image I′; and
generating an image for display with segmented and registered objects identified.
18. The apparatus as claimed in claim 17 , wherein generating the segmentation contour for each object i of N objects in baseline image I, comprises:
providing a click-point x 0 i to each object i of the N objects in image I to create N click-points X 0 =(x 0 1 , . . . ,x 0 N );
assigning a prior shape to each contour;
parameterizing a level-set function φ γ representing a shape, based on the prior shape; and
maximizing a conditional probability P(Γ|I,X 0 ).
19. The apparatus as claimed in claim 18 , wherein the prior shape is an ellipsoid.
20. The apparatus as claimed in claim 19 , further comprising:
defining the level-set function as φ γ i (x):=φ(x,μ i +Δμ i +T,Σ i +ΔΣ i ) wherein Δμ i denotes a displacement of object i ellipsoid's mass center T is a global rigid translation of the N objects, each object having an ellipsoid shape, and ΔΣ i describes a shape difference of object i in image I with corresponding object i′ in the follow-up image I′.