System and method for verifying registration accuracy in digital medical images
View Patent ↗A method of verifying a registration of a digital image includes receiving a reference image and a moving image registered to the reference image representing the same object, selecting one or more landmarks in the images, defining a neighborhood in the domain of each image near each selected landmark, extracting feature sets from the neighborhoods for the reference image and the registered moving image, calculating a total quality measure of the registration of the moving image from a distance metric applied to the feature sets extracted from the neighborhoods of the reference image and the registered moving image, comparing the total quality measure to a threshold determined from a reference atlas for the object represented in the image, and determining whether to accept or reject the registration of the moving image based on the comparison result.
1. A method of verifying a registration of a digital image, said method comprising the steps of:
receiving a reference image and a moving image that has been registered to said reference image, wherein said reference image and said moving image are representations of a same object, wherein each said image comprises a plurality of intensities defined for a three-dimensional (3D) domain of points;
selecting a symmetry axis in the object represented in the image;
selecting one or more landmarks in said images, and defining a neighborhood in the domain of each image near each selected landmark, said neighborhoods defined to be symmetric about said symmetry axis;
extracting feature sets from the neighborhoods for the reference image and the moving image;
calculating a total quality measure of the registration of the moving image from a distance metric applied to the feature sets extracted from the neighborhoods of the reference image and the registered moving image, wherein said total quality measure is a weighed sum of neighborhood quality measures calculated by applying the distance metric to said feature sets in each neighborhood defined as
q
i
=
d
(
S
ref
+
(
i
)
,
S
mov
+
(
i
)
,
S
ref
-
(
i
)
,
S
mov
-
(
i
)
)
,
wherein d is the distance metric for comparing two sets of data, S mov + (i) and S mov − (i) express the features extracted from each landmark i in their respective neighborhoods of said registered moving image on both sides of the symmetry axis, and S ref + (i) and S ref − (i) express the features extracted from each landmark i in their respective neighborhoods of said reference image on both sides of the symmetry axis;
comparing the total quality measure to a threshold determined from a reference atlas for the object represented in the image; and
determining whether to accept or reject the registration of the moving image based on the comparison result.
2. The method of claim 1 , wherein said distance metric is a cross correlation calculated over respective neighborhoods of the registered moving image and the reference image.
3. The method of claim 1 , wherein said distance metric is a sum of squared intensity differences calculated over respective neighborhoods of the registered moving image and the reference image.
4. The method of claim 1 , wherein said total quality measure is defined as
QM
=
∑
i
n
w
i
·
d
(
S
ref
(
i
)
,
S
mov
(
i
)
)
+
∑
i
,
j
w
ij
2
·
d
(
S
mov
,
j
+
(
x
i
,
y
i
,
z
i
)
,
S
mov
,
j
-
(
x
i
,
y
i
,
z
i
)
)
,
wherein d is said distance metric, the first sum is over the neighborhoods, w i are weights for the neighborhoods, the second sum is over the neighborhoods i for different orientations j about said symmetry axis, w ij are weights for the neighborhoods at orientation j, and sets S mov,j + and S mov,j − are subsets of set S mov (i) at orientation j on respective sides of the symmetry axis.
5. The method of claim 1 , wherein said object is a brain, and said symmetry axis is a mid-saggital plane.
6. A non-transitory program storage device readable by a computer, tangibly embodying a program of instructions executable by the computer to perform the method steps for verifying a registration of a digital image, said method comprising the steps of:
receiving a reference image and a moving image that has been registered to said reference image, wherein said reference image and said moving image are representations of a same object, wherein each said image comprises a plurality of intensities defined for a three-dimensional (3D) domain of points;
selecting a symmetry axis in the object represented in the image;
selecting one or more landmarks in said images, and defining a neighborhood in the domain of each image near each selected landmark, said neighborhoods defined to be symmetric about said symmetry axis;
extracting feature sets from the neighborhoods for the reference image and the moving image;
calculating a total quality measure of the registration of the moving image from a distance metric applied to the feature sets extracted from the neighborhoods of the reference image and the registered moving image, wherein said total quality measure is a weighed sum of neighborhood quality measures calculated by applying the distance metric to said feature sets in each neighborhood defined as
q
i
=
d
(
S
ref
+
(
i
)
,
S
mov
+
(
i
)
,
S
ref
-
(
i
)
,
S
mov
-
(
i
)
)
,
wherein d is the distance metric for comparing two sets of data, S mov + (i) and S mov − (i) express the features extracted from each landmark i in their respective neighborhoods of said registered moving image on both sides of the symmetry axis, and S ref + (i) and S ref − (i) express the features extracted from each landmark i in their respective neighborhoods of said reference image on both sides of the symmetry axis;
comparing the total quality measure to a threshold determined from a reference atlas for the object represented in the image; and
determining whether to accept or reject the registration of the moving image based on the comparison result.
7. The computer readable program storage device of claim 6 , wherein said distance metric is a cross correlation calculated over respective neighborhoods of the registered moving image and the reference image.
8. The computer readable program storage device of claim 6 , wherein said distance metric is a sum of squared intensity differences calculated over respective neighborhoods of the registered moving image and the reference image.
9. The computer readable program storage device of claim 6 , wherein said total quality measure is defined as
QM
=
∑
i
n
w
i
·
d
(
S
ref
(
i
)
,
S
mov
(
i
)
)
+
∑
i
,
j
w
ij
2
·
d
(
S
mov
,
j
+
(
x
i
,
y
i
,
z
i
)
,
S
mov
,
j
-
(
x
i
,
y
i
,
z
i
)
)
,
wherein d is said distance metric, the first sum is over the neighborhoods, w i are weights for the neighborhoods, the second sum is over the neighborhoods i for different orientations j about said symmetry axis, w ij are weights for the neighborhoods at orientation j, and sets S mov,j + and S mov,j − are subsets of set S mov (i) at orientation j on respective sides of the symmetry axis.
10. The computer readable program storage device of claim 6 wherein said object is a brain, and said symmetry axis is a mid-saggital plane.