Fluid dynamics approach to image segmentation
View Patent ↗A method for segmenting image data within a data processing system includes acquiring an image. One or more seed points are established within the image. An advection vector field is computed based on image influences and user input. A dye concentration is determined at each of a plurality of portions of the image that results from a diffusion of dye within the computed advection field. The image is segmented into one or more regions based on the determined dye concentration for the corresponding dye.
1. A method for segmenting image data within a data processing system, comprising:
acquiring static image of a physical structure;
establishing one or more seed points within the image;
modeling the physical structure of the static image as a fluid model;
calculating a flow field for the fluid model by using image features of the physical structure of the image as influences of flow within the fluid model;
using the established one or more seed points as sources and sinks of at least one diffusive dye that diffuses and advects through the calculated flow field;
determining a dye concentration at each of a plurality of portions of the fluid model that results from the diffusion and advection of the dye within the calculated flow field; and
segmenting the physical structure of the image into one or more regions based on the determined dye concentration for the corresponding dye within the fluid model,
wherein segmenting the physical structure of the image based on the dye concentrations within the fluid model includes:
examining the fluid model as a set of pixels/voxels;
determining the concentration for each dye at each pixel/voxel;
segmenting the fluid model into continuous regions of pixels/voxels in accordance with a dominant concentration of a particular dye; and
carrying over the segmentation of the fluid model back into the physical structure of the image.
2. The method of claim 1 , additionally comprising establishing one or more advection jets within the fluid model and factoring in the effects of the established one or more advection jets in calculating the flow field for the fluid model.
3. The method of claim 1 , wherein there is one unique flow field for each diffusive dye.
4. The method of claim 1 , wherein the acquired image is a two-dimensional or three-dimensional image.
5. The method of claim 1 , wherein the acquired image is a three-dimensional medical image and either a CT image, an MR image, a PET image or an ultrasound image.
6. The method of claim 5 , wherein a flow field for the fluid model is obtained directly from an imaging modality used in acquiring the image of the physical structure.
7. The method of claim 1 , wherein the one or more seed points are established by a user identifying at least one location of the image that is known to be part of each of the one or more regions.
8. The method of claim 1 , wherein the one or more seed points are automatically established by identifying at least one location of the image that is known to be part of each of the one or more regions.
9. The method of claim 2 , wherein the one or more advection jets are established by a user identifying directions of flow corresponding to particular seed points.
10. The method of claim 2 , wherein the one or more advection jets are automatically established by identifying directions of flow corresponding to particular seed points.
11. The method of claim 2 , wherein:
the one or more seed points are established by a user identifying at least one location of the image that is known to be part of each of the one or more regions; and
the one or more advection jets are established by a user identifying directions of flow corresponding to particular seed points; wherein
the user identifies a seed point of the one or more seed points and an advection jet of the one or more advection jets substantially simultaneously by the use of a gesture-based input.
12. The method of claim 2 , wherein the flow field is calculated using Navier-Stokes equations adapted to scale diffusion and advection based on image features, viscosity and diffusion rates.
13. A method for segmenting image data within a data processing system, comprising:
acquiring static image of a physical structure;
establishing one or more seed points within the image;
modeling the physical structure of the static image as a fluid model;
calculating a flow field for the fluid model by using image features of the physical structure of the image as influences of flow within the fluid model;
using the established one or more seed points as sources and sinks of at least one diffusive dye that diffuses and advects through the calculated flow field;
determining a dye concentration at each of a plurality of portions of the fluid model that results from the diffusion and advection of the dye within the calculated flow field;
segmenting the physical structure of the image into one or more regions based on the determined dye concentration for the corresponding dye within the fluid model,
the method additionally comprising establishing one or more advection jets within the fluid model and factoring in the effects of the established one or more advection jets in calculating the flow field for the fluid model,
wherein the flow field is calculated using Navier-Stokes equations adapted to scale diffusion and advection based on image features, viscosity and diffusion rates,
wherein the adapted Navier-Stokes equations take the following form:
∂
u
∂
t
=
-
ⅇ
^
(
-
m
v
1
(
(
u
·
∇
)
I
)
2
)
(
u
·
∇
)
u
+
v
∇
·
[
∇
u
·
diag
(
ⅇ
^
(
-
m
v
2
(
∇
I
·
∇
I
)
)
)
]
+
f
∂
ρ
∂
t
=
-
ⅇ
^
(
-
m
d
1
(
(
u
·
∇
)
I
)
2
)
(
u
·
∇
)
ρ
+
k
∇
·
[
∇
ρ
·
ⅇ
^
(
-
m
d
2
(
∇
I
·
∇
I
)
)
)
]
+
S
wherein I represents the image, the vector u represents the flow field, ρ is a dye distribution through a region, v controls the viscosity of the fluid, k is the diffusion rate of the dye, the vector f represents external forces influencing the flow, the vector s represents external influences to the dye, the terms m v1 and m d1 control the influence of the image on advection of the fluid and dye, and the terms m v2 and m d2 control the viscosity and diffusion of the vector field and the diffusion of the dye.
14. The method of claim 1 , wherein each of the at least one diffusive dyes corresponds to a label and with respect to each label, the seed point(s) corresponding to a corresponding label are established as dye sources and seed point(s) of other labels are established as dye sinks and wherein this step is performed for each label.
15. The method of claim 1 , wherein the dye concentration is determined at a steady state, after a predetermined length of time, or at a point in which subsequent changes to the dye concentrations are insignificant.
16. The method of claim 1 , wherein the dye concentration is determined deterministically by a direct solution.
17. A method for segmenting image data within a data processing system, comprising:
acquiring static image of a physical structure;
modeling the physical structure of the static image as a fluid model;
calculating a flow field for the fluid model by using image features of the physical structure of the image;
modeling a diffusive dye that diffuses and advects through the calculated flow field;
segmenting the physical structure of the image into one or more regions based on a determined dye concentration for the diffusive dye within the fluid model,
wherein segmenting the physical structure of the image within the fluid model includes:
examining the fluid model as a set of pixels/voxels;
determining the concentration for the diffusive dye at each pixel/voxel;
segmenting the fluid model into continuous regions of pixels/voxels in accordance with a concentration of the diffusive dye; and
carrying over the segmentation of the fluid model back into the physical structure of the image.
18. A computer system comprising:
a processor; and
a program storage device readable by the computer system, embodying a program of instructions executable by the processor to perform method steps for segmenting image data within a data processing system, the method comprising:
acquiring a static image of a physical structure;
establishing one or more seed points within the image;
modeling the physical structure of the image as a fluid model;
computing an advection vector field within the fluid model based on image influences and user input within the image of the physical structure;
determining a dye concentration at each of a plurality of portions of the fluid model that results from a diffusion and advection of dye within the computed advection field; and
segmenting the physical structure of the image into one or more regions based on the determined dye concentration for the corresponding dye within the fluid model,
wherein segmenting the physical structure of the image based on the dye concentrations within the fluid model includes:
examining the fluid model as a set of pixels/voxels;
determining the concentration for each dye at each pixel/voxel;
segmenting the fluid model into continuous regions of pixels/voxels in accordance with a dominant concentration of a particular dye; and
carrying over the segmentation of the fluid model back into the physical structure of the image.
19. The computer system of claim 18 , wherein the method additionally comprises establishing one or more advection jets within the fluid model and factoring in the effects of the established one or more advection jets in determining the dye concentration at each portion of the fluid model.