3D general lesion segmentation in CT
View Patent ↗A general purpose method to segment any kind of lesions in 3D images is provided. Based on a click or a stroke inside the lesion from the user, a distribution of intensity level properties is learned. The random walker segmentation method combines multiple 2D segmentation results to produce the final 3D segmentation of the lesion.
1. A method for segmentation of an object from a background in three-dimensional (3D) image data, comprising:
initiating at least one seed in the object in a two-dimensional (2D) representation of the 3D image, the at least one seed defines the object;
determining a distribution of intensity levels of pixels in the 2D representation of the 3D image relative to an intensity level of the at least one seed;
segmenting the object from the background in the 2D representation of the 3D image based on the distribution of intensity levels into a first 2D segmentation contour;
establishing a 2D segmentation contour of the object in at least two additional 2D representations of the image, such that each 2D representation contains the at least one seed in the object; and
creating a 3D segmentation contour of the object by applying a random walker segmentation method to all pixels in the 3D image, using pixels inside the 2D segmentation contours as object seeds and pixels outside the 2D segmentation contours as background seeds;
determining a cost image C L associated with the at least one seed in the 2D representation of the object;
determining an object histogram H L associated with the cost image as the distribution of intensity levels of pixels:
computing a response image of the 2D representation of the object by applying an expression:
g
(
x
,
y
)
=
{
H
L
(
f
(
x
,
y
)
)
(
1
-
C
L
(
x
,
y
)
C
L
max
)
if
L
(
x
,
y
)
=
1
H
L
(
f
(
x
,
y
)
)
/
2
if
L
(
x
,
y
)
=
0.
2. The method as claimed in claim 1 , further comprising:
determining a background histogram H B for pixels not being part of the object in the 2D representation.
3. The method as claimed in claim 2 , wherein the establishing of a 2D segmentation contour of the object in at least two additional 2D representations of the 3D image data using a response image that can be computed from an expression:
g
(
x
,
y
)
=
{
g
(
x
,
y
)
if
H
L
(
f
(
x
,
y
)
)
>
H
B
(
f
(
x
,
y
)
)
else
3
H
L
(
f
(
x
,
y
)
)
4
(
1
-
C
L
(
x
,
y
)
C
L
max
)
if
L
(
x
,
y
)
=
1
H
L
(
f
(
x
,
y
)
)
/
4
if
L
(
x
,
y
)
=
0.
4. The method as claimed in claim 1 , wherein the 3D image data is provided by a CT scanner.
5. A method for segmentation of an object from a background in three-dimensional (3D) image data, comprising:
initiating at least one seed in the object in a two-dimensional (2D) representation of the 3D image, the at least one seed defines the object;
determining a distribution of intensity levels of pixels in the 2D representation of the 3D image relative to an intensity level of the at least one seed;
segmenting the object from the background in the 2D representation of the 3D image based on the distribution of intensity levels into a first 2D segmentation contour;
establishing a 2D segmentation contour of the object in at least two additional 2D representations of the image, each 2D representation containing the at least one seed in the object; and
creating a 3D segmentation contour of the object by applying a random walker segmentation method to all pixels in the 3D image, using pixels inside the 2D segmentation contours as object seeds and pixels outside the 2D segmentation contours as background seeds;
determining a cost image C L associated with the at least one seed in the 2D representation of the object;
determining an object histogram H L associated with the cost image as the distribution of intensity levels of pixels;
determining a background histogram H B for pixels not being part of the object in the 2D representation
wherein the establishing of a 2D segmentation contour of the object in at least two additional 2D representations of the 3D image data using a response image is computed from an expression:
g
(
x
,
y
)
=
{
g
(
x
,
y
)
if
H
L
(
f
(
x
,
y
)
)
>
H
B
(
f
(
x
,
y
)
)
else
3
H
L
(
f
(
x
,
y
)
)
4
(
1
-
C
L
(
x
,
y
)
C
L
max
)
if
L
(
x
,
y
)
=
1
H
L
(
f
(
x
,
y
)
)
/
4
if
L
(
x
,
y
)
=
0.
6. The method as claimed in claim 5 , wherein the 3D image data is provided by a CT scanner.