US 9,317,927 B2 · Methods and systems for interactive 3D image segmentation

Abstract

Methods and systems for interactively segmenting 3D image data are provided. An initial segmentation of the 3D image data is obtained, and for each of a plurality of image regions, a segmentation uncertainty indicator for the initial image segmentation is associated with the image region and a strength is assigned to the segmentation uncertainty indicator. A low-confidence region in the 3D image data is identified based at least in part on proximity of the low-confidence region to the image regions and strengths of the corresponding segmentation uncertainty indicators. An optimization routine may be applied to an objective function whose value depends at least in part on proximity of the candidate region to the image regions and the strengths of the corresponding uncertainty indicators to identify the low-confidence region from among a plurality of candidate regions.

Claims40

Dependent claims

A computer-implemented method for improving an initial segmentation of 3D digital image data to obtain an updated segmentation of the 3D digital image data, the method comprising:

obtaining an initial segmentation of a 3D digital image comprising corresponding 3D image data;
storing the initial segmentation of the 3D digital image data in memory accessible to a computer;
for each of a plurality of image regions within the 3D digital image data, associating an uncertainty indicator for the initial image segmentation with the image region and assigning, by the computer, a strength to the uncertainty indicator;
identifying, by the computer, a low-confidence region in the 3D digital image data based at least in part on proximity of the low-confidence region to the image regions and strengths of the corresponding uncertainty indicators;
obtaining segmentation information pertaining to the low-confidence region and storing the segmentation information pertaining to the low-confidence region in the memory accessible to the computer, wherein the segmentation information pertaining to the low-confidence region is based at least in part on user input as to a segmentation of the low-confidence region;
updating, by the computer, the initial segmentation of the 3D digital image data to obtain an updated segmentation of the 3D digital image data, wherein updating the initial segmentation of the 3D digital image data to obtain the updated segmentation of the 3D digital image data is based on the segmentation information pertaining to the low-confidence region and thereby benefits from the user input to achieve improved accuracy of the updated segmentation of the 3D digital image data relative to the initial segmentation of the 3D digital image data;
and
storing the updated segmentation of the 3D digital image data in the memory accessible to the computer.

5 more elements

A computer-implemented method for improving an initial segmentation of 3D digital image data to obtain an updated segmentation of the 3D digital image data, the method comprising:

obtaining an initial segmentation of the 3D digital image data;
storing the initial segmentation of the 3D digital image data in memory accessible to a computer;
for each of a plurality of image regions within the 3D digital image data, associating an uncertainty indicator for the initial image segmentation with the image region and assigning, by the computer, a strength to the uncertainty indicator;
identifying, by the computer, a low-confidence region in the 3D digital image data based at least in part on proximity of the low-confidence region to the image regions and strengths of the corresponding uncertainty indicators;
obtaining segmentation information pertaining to the low-confidence region and storing the segmentation information pertaining to the low-confidence region in the memory accessible to the computer, wherein the segmentation information pertaining to the low-confidence region is based at least in part on user input as to a segmentation of the low-confidence region;
updating, by the computer, the initial segmentation of the 3D digital image data to obtain an updated segmentation of the 3D digital image data, wherein updating the initial segmentation of the 3D digital image data to obtain the updated segmentation of the 3D digital image data is based on the segmentation information pertaining to the low-confidence region and thereby benefits from the user input to achieve improved accuracy of the updated segmentation of the 3D digital image data relative to the initial segmentation of the 3D digital image data;
and
storing the updated segmentation of the 3D digital image data in the memory accessible to the computer;
wherein identifying, by the computer, the low-confidence region in the 3D image data comprises computing, for each of a plurality of candidate regions, a value of an objective function based at least in part on proximity of the candidate region to the image regions and the strengths of the corresponding uncertainty indicators;
and
wherein the value of the objective function for a candidate region having position R is proportional to a value of an expression
∑
i
∈
UI
⁢
s
i
⁢
L
⁡
(
p
i
,
R
)
or a mathematical equivalent thereto, where:
UI represents a set comprising the plurality of uncertainty indicators for the initial image segmentation corresponding to the plurality of image regions and indexed by an index variable i;
si is the strength of the ith uncertainty indicator;
pi is a position of the ith image region;
and
L is a proximity metric function whose value monotonically decreases with separation between R and pi.

30 more elements

A computer-implemented method for improving an initial segmentation of 3D digital image data to obtain an updated segmentation of the 3D digital image data, the method comprising:

obtaining an initial segmentation of the 3D digital image data;
storing the initial segmentation of the 3D digital image data in memory accessible to a computer;
for each of a plurality of image regions within the 3D digital image data, associating an uncertainty indicator for the initial image segmentation with the image region and assigning, by the computer, a strength to the uncertainty indicator;
identifying, by the computer, a low-confidence region in the 3D digital image data based at least in part on proximity of the low-confidence region to the image regions and strengths of the corresponding uncertainty indicators;
obtaining segmentation information pertaining to the low-confidence region and storing the segmentation information pertaining to the low-confidence region in the memory accessible to the computer, wherein the segmentation information pertaining to the low-confidence region is based at least in part on user input as to a segmentation of the low-confidence region;
updating, by the computer, the initial segmentation of the 3D digital image data to obtain an updated segmentation of the 3D digital image data, wherein updating the initial segmentation of the 3D digital image data to obtain the updated segmentation of the 3D digital image data is based on the segmentation information pertaining to the low-confidence region and thereby benefits from the user input to achieve improved accuracy of the updated segmentation of the 3D digital image data relative to the initial segmentation of the 3D digital image data;
and
storing the updated segmentation of the 3D digital image data in the memory accessible to the computer;
wherein identifying, by the computer, the low-confidence region in the 3D digital image data comprises computing, for each of a plurality of candidate regions, a value of an objective function based at least in part on proximity of the candidate region to the image regions and the strengths of the corresponding uncertainty indicators;
wherein the method further comprises, for each of the plurality of uncertainty indicators, assigning, by the computer, a corresponding unit normal vector to the uncertainty indicator;
wherein the value of the objective function for a planar candidate region P is proportional to a value of an expression
∑
i
∈
UI
⁢
s
i
⁡
[
q
p
⁡
(
i
)
+
(
1
-
q
P
⁡
(
i
)
)
⁢
L
⁡
(
d
⁡
(
p
i
,
P
)
)
]
or a mathematical equivalent thereto, where:
UI represents a set comprising the plurality of uncertainty indicators for the initial image segmentation corresponding to the plurality of image regions and indexed by an index variable i;
si is the strength of the ith uncertainty indicator;
pi is a position of the ith image region;
qP(i) is equal to (ni·nP)2, where nP represents a unit normal vector to the planar candidate region P, ni represents the unit normal vector of the ith uncertainty indicator and ni·nP is a dot product of ni and nP;
d(pi,P) represents a distance metric proportional to a distance between the position pi of the ith image region and the planar candidate region P;
and
L represents a logistic shaped function mapping [0,∞)→[0,1).

56 more elements

A computer-implemented method for improving an initial segmentation of 3D digital image data to obtain an updated segmentation of the 3D digital image data, the method comprising:

obtaining an initial segmentation of the 3D digital image data;
storing the initial segmentation of the 3D digital image data in memory accessible to a computer;
for each of a plurality of image regions within the 3D digital image data, associating an uncertainty indicator for the initial image segmentation with the image region and assigning, by the computer, a strength to the uncertainty indicator;
identifying, by the computer, a low-confidence region in the 3D digital image data based at least in part on proximity of the low-confidence region to the image regions and strengths of the corresponding uncertainty indicators;
obtaining segmentation information pertaining to the low-confidence region and storing the segmentation information pertaining to the low-confidence region in the memory accessible to the computer, wherein the segmentation information pertaining to the low-confidence region is based at least in part on user input as to a segmentation of the low-confidence region;
updating, by the computer, the initial segmentation of the 3D digital image data to obtain an updated segmentation of the 3D digital image data, wherein updating the initial segmentation of the 3D digital image data to obtain the updated segmentation of the 3D digital image data is based on the segmentation information pertaining to the low-confidence region and thereby benefits from the user input to achieve improved accuracy of the updated segmentation of the 3D digital image data relative to the initial segmentation of the 3D digital image data;
and
storing the updated segmentation of the 3D digital image data in the memory accessible to the computer;
wherein obtaining the initial segmentation of the 3D image data comprises:
obtaining at least one 2D image segmentation for the 3D image data;
placing seeds in an unvisited plane at intersections of the at least one 2D image segmentation with the unvisited plane;
ordering the seeds in the unvisited plane;
and
obtaining an automatic 2D image segmentation of the unvisited plane based on the ordered seeds.

11 more elements

A computer-implemented method for improving a first segmentation of 3D digital image data to obtain an updated segmentation of the 3D digital image data, the method comprising:

obtaining a 3D digital image comprising corresponding 3D image data and a first segmentation of the 3D image data;
storing the first segmentation of the 3D digital image data in a memory accessible to the computer
for each of a plurality of image regions within the 3D digital image data, associating an uncertainty indicator for the first image segmentation with the image region and assigning, by the computer, a strength to the uncertainty indicator;
identifying, by the computer, a low-confidence region in the 3D digital image data based at least in part on proximity of the low-confidence region to the image regions and strengths of the corresponding uncertainty indicators;
and
recommending the low-confidence region by outputting the low-confidence region to thereby indicate that the low-confidence region comprises the image region for which further user input would be desirable;
obtaining segmentation information pertaining to the low-confidence region and storing the segmentation information pertaining to the low-confidence region in the memory accessible to the computer, wherein the segmentation information pertaining to the low-confidence region is based at least in part on user input as to a segmentation of the low-confidence region;
updating, by the computer, the first segmentation of the 3D digital image data to obtain an updated segmentation of the 3D digital image data, wherein updating the first segmentation of the 3D digital image data to obtain the updated segmentation of the 3D digital image data is based on the segmentation information pertaining to the low-confidence region and thereby benefits from the user input to achieve improved accuracy of the updated segmentation of the 3D digital image data relative to the first segmentation of the 3D digital image data;
and
storing the updated segmentation of the 3D digital image data in the memory accessible to the computer.

7 more elements

Dependencies are inferred from the claim text.

Methods and systems for interactive 3D image segmentation

Life of this patent

Methods and systems for interactive 3D image segmentation

Life of this patent

Abstract

Claims40