Registration system and method for tracking lung nodules in medical images
View Patent ↗A nodule registration system useful for tracking lung nodules in computed tomography (CT) scans is presented. The system registers a small sphere centered on a detected nodule in one scan with another scan under a rigid transformation assumption using a fast registration scheme. The registration scheme employs very fast simulated annealing (VFSA) with constraints to maximize a tunable cross-correlation (TCC) coefficient, enabling the system to register, with minimal registration error, all nodules within their maximum diameter. The system achieves an average registration time of 10 seconds or less on a 3.06 GHz computer programmed to implement the present invention.
1. A computer-readable medium storing a computer program implementing method steps for registering nodules in medical images, said method steps comprising:
a) preprocessing said medical images, comprising:
segmenting structures of interest into source and target volumes;
deriving three-dimensional (3D) bounding boxes from said volumes; and
estimating a location of a target nodule from said bounding boxes;
b) optimizing said location, comprising:
determining possible regions for said target nodule;
mapping 3D coordinates of points in each region to a 1D index; and
in the 1D index space, finding an optimized location that has the maximum Tunable Cross-Correlation (TCC) coefficient with a sphere containing a source nodule; and
c) optimizing said TCC coefficient, comprising:
using the negative of the TCC coefficient as an energy function for a Very Fast Simulated Annealing (VFSA) algorithm;
augmenting said VFSA algorithm with a re-annealing schedule and an anti-local-maximum procedure; and
controlling gradient of TCC changes with respect to shift, thereby improving registration accuracy.
2. The computer-readable medium according to claim 1 , wherein said TCC is defined as:
TCC
=
∑
(
i
,
j
,
k
)
∈
S
(
u
(
i
,
j
,
k
)
-
q
1
u
_
)
(
v
(
i
,
j
,
k
)
-
q
2
v
_
)
∑
(
i
,
j
,
k
)
∈
S
(
u
(
i
,
j
,
k
)
-
q
1
u
_
)
2
∑
i
,
j
,
k
(
v
(
i
,
j
,
k
)
-
q
2
v
_
)
2
where u(i,j,k) and v(i,j,k) denote the intensity of points in said source volume and said target volume, respectively, ū is the mean of u(i,j,k), v is the mean of v(i,j,k), S is the volume over which said TCC is calculated, and q 1 , q 2 are two real numbers.
3. The computer-readable medium according to claim 1 , wherein said medical images are computed tomography (CT) scans.
4. A computer system programmed to perform method steps, wherein said method steps comprising
a) preprocessing said medical images, comprising:
segmenting structures of interest into source and target volumes;
deriving three-dimensional (3D) bounding boxes from said volumes; and
estimating a location of a target nodule from said bounding boxes;
b) optimizing said location, comprising:
determining possible regions for said target nodule;
mapping 3D coordinates of points in each region to a 1D index; and
in the 1D index space, finding an optimized location that has the maximum Tunable Cross-Correlation (TCC) coefficient with a sphere containing a source nodule; and
c) optimizing said TCC coefficient, comprising:
using the negative of the TCC coefficient as an energy function for a Very Fast Simulated Annealing (VFSA) algorithm;
augmenting said VFSA algorithm with a re-annealing schedule and an anti-local-maximum procedure; and
controlling gradient of TCC changes with respect to shift, thereby improving registration accuracy.
5. The computer system according to claim 4 , wherein said TCC is defined as:
TCC
=
∑
(
i
,
j
,
k
)
∈
S
(
u
(
i
,
j
,
k
)
-
q
1
u
_
)
(
v
(
i
,
j
,
k
)
-
q
2
v
_
)
∑
(
i
,
j
,
k
)
∈
S
(
u
(
i
,
j
,
k
)
-
q
1
u
_
)
2
∑
i
,
j
,
k
(
v
(
i
,
j
,
k
)
-
q
2
v
_
)
2
where u(i,j,k) and v(i,j,k) denote the intensity of points in said source volume and said target volume, respectively, ū is the mean of u(i,j,k), v is the mean of v(i,j,k), S is the volume over which said TCC is calculated, and q 1 , q 2 are two real numbers.
6. The computer system according to claim 4 , wherein said medical images are computed tomography (CT) scans.