System and method for 2-D/3-D registration between 3-D volume and 2-D angiography
View Patent ↗A method for registering a 2-D DSA image to a 3-D image volume includes calculating a coarse similarity measure between a 2-D DRR of an aorta and a cardiac DSA image, and a 2-D DRR of a coronary artery and the cardiac DSA image, for a plurality of poses over a range of 2-D translations. Several DRR-pose combinations with largest similarity measures are selected as refinement candidates. The similarity measure is calculated between the refinement candidate DRRs and the DSA, for a plurality of poses over a range of 3-D translations and in-plane rotations. One or more DRR-pose combinations with largest similarity measures are selected as final candidates. The similarity measure between the final candidate DRRs the DSA are calculated for a plurality of poses over a range of 3D translations and 3D rotations, and a DRR-pose combination with a largest similarity measure is selected as a final registration result.
1. A method for registering a 2-dimensional digital subtraction angiography (DSA) image to a 3-dimensional image volume during a cardiac procedure, comprising the steps of:
calculating a coarse similarity measure between a 2-dimensional digitally reconstructed radiograph (DRR) of an aorta and a cardiac DSA image, and a 2-dimensional DRR of a coronary artery and the cardiac DSA image, for a plurality of poses over a range of 2-dimensional translations;
selecting one or more DRR-pose combinations for the aorta and the coronary artery with largest similarity measures as refinement candidates;
calculating the similarity measure between the refinement candidate DRRs of the aorta and the DSA, and between the refinement candidate DRRs of the coronary artery and the DSA, for a plurality of poses over a range of 3-dimensional translations and in-plane rotations;
selecting one or more DRR-pose combinations for the aorta and the coronary artery with largest similarity measures as final candidates;
calculating the similarity measure between the final candidate DRRs of the aorta and the DSA, and between the final candidate DRRs of the coronary artery and the DSA, for a plurality of poses over a range of 3-dimensional translations and 3-dimensional rotations; and
selecting a DRR-pose combination with a largest similarity measure as a final registration result.
2. The method of claim 1 , further comprising pre-processing the DSA image to subtract a background image, apply morphological operations, and remove artifacts of the morphological operations.
3. The method of claim 1 , wherein the DRR of an aorta and the DRR of a coronary artery are generated from a same 3-dimensional cardiac image volume.
4. The method of claim 3 , further comprising pre-processing the cardiac image volume to create 3-dimensional image masks for the aorta and the coronary artery, wherein said image masks are used to generate the DRR of the aorta and the DRR of the coronary artery.
5. The method of claim 3 , wherein the DRR of an aorta and the DRR of a coronary artery are generated for a plurality of poses at a plurality of depths in the cardiac image volume.
6. The method of claim 5 , wherein the DRR of an aorta and the DRR of a coronary artery are centered about an estimated center of the aortic root.
7. The method of claim 1 , wherein the coarse similarity measure is calculated between a downsampled 2-dimensional DRR of the aorta, a downsampled 2-dimensional DRR of the coronary artery, and a downsampled cardiac DSA image.
8. The method of claim 1 , wherein the similarity measure is defined by
SM=GC Aorta ( I Aorta ,I DSA ,M Aorta )+β· GC Coronary ( I Coronary ,I DSA ,M Coronary )
wherein I Aorta represents the aortic DRR image, I Coronary represents the coronary artery DRR image, I DSA represents the DSA image,
M Aorta represents a 2-dimensional aorta image mask defined as
M
Aorta
=
{
1
,
if
∇
I
Aorta
>
α
,
0
,
otherwise
,
,
where α is a threshold of the image gradient,
M Coronary represents a 2-dimensional coronary artery image mask defined as
M
Coronary
=
{
1
,
if
min
(
∇
I
Coronary
,
∇
I
DRR
)
>
α
,
0
,
otherwise
.
,
β is a heuristically determined weight for coronary ostia landmark features,
GC is a gradient correlation between images I 1 and I 2 with image mask M defined as
GC
(
I
1
,
I
2
,
M
)
=
NCC
(
∂
I
1
∂
x
,
∂
I
2
∂
x
,
M
)
+
NCC
(
∂
I
1
∂
y
,
∂
I
2
∂
y
,
M
)
,
and NCC denotes the normalized cross correlation of the masked images defined as
NCC
(
I
1
,
I
2
,
M
)
=
∑
M
(
x
,
y
)
=
1
I
1
(
x
,
y
)
-
I
_
1
(
x
,
y
)
I
2
(
x
,
y
)
-
I
_
2
(
x
,
y
)
(
∑
M
(
x
,
y
)
=
1
I
1
(
x
,
y
)
-
I
_
1
(
x
,
y
)
I
2
(
x
,
y
)
-
I
_
2
(
x
,
y
)
)
2
,
wherein Ī denotes the average intensity of pixels in the masked region of the image.
9. The method of claim 4 , wherein pre-processing the cardiac image volume to create 3-dimensional image masks for the aorta comprises:
segmenting the aorta in the cardiac image volume; and
defining the aortic mask by
VM
aorta
(
x
,
y
,
z
)
=
{
1
,
if
(
x
,
y
,
z
)
∈
aorta
,
0
,
otherwise
,
.
10. The method of claim 9 , wherein pre-processing the cardiac image volume to create 3-dimensional image masks for the coronary artery comprises:
detecting left and right coronary ostia in the cardiac image volume;
centering two spherical masks VM l and VM r around the detected coronary ostia;
forming a mask VM out from the union of VM l and VM r , excluding an area inside VM aorta and calculating an intensity distribution for voxels in the mask VM out ; and
choosing a lower bound of a volume transfer window associated with the coronary artery image mask as a given percentile of the intensity distribution, and an upper bound of the volume transfer window to be a highest voxel intensity of the volume.
11. A method for registering a 2-dimensional fluoroscopy image to a 3-dimensional image volume during a cardiac procedure, comprising the steps of:
providing a 3-dimensional cardiac image volume and a 2-D cardiac fluoroscopy image;
generating a plurality of 2-dimensional DRRs of an aorta and a plurality of 2-dimensional DRRs of a coronary artery at a plurality of depths from the cardiac image volume;
calculating the similarity measure between the aorta DRR and the DSA, and between the coronary artery DRR and the DSA, for a plurality of poses; and
selecting a DRR-pose combination with a largest similarity measure as a registration result,
wherein the similarity measure is defined by
SM=GC Aorta ( I Aorta ,I DSA ,M Aorta )+β· GC Coronary ( I Coronary ,I DSA ,M Coronary )
wherein I Aorta represents the aortic DRR image, I Coronary represents the coronary artery DRR image, I DSA represents the DSA image,
M Aorta represents a 2-dimensional aorta image mask defined as
M
Aorta
=
{
1
,
if
∇
I
Aorta
>
α
,
0
,
otherwise
,
,
where α is a threshold of the image gradient,
M Coronary represents a 2-dimensional coronary artery image mask defined as
M
Coronary
=
{
1
,
if
min
(
∇
I
Coronary
,
∇
I
DRR
)
>
α
,
0
,
otherwise
.
,
β is a heuristically determined weight for coronary ostia landmark features,
GC is a gradient correlation between images I 1 and I 2 with image mask M defined
as
GC
(
I
1
,
I
2
,
M
)
=
NCC
(
∂
I
1
∂
x
,
∂
I
2
∂
x
,
M
)
+
NCC
(
∂
I
1
∂
y
,
∂
I
2
∂
y
,
M
)
,
and NCC denotes the normalized cross correlation of the masked images defined as
NCC
(
I
1
,
I
2
,
M
)
=
∑
M
(
x
,
y
)
=
1
I
1
(
x
,
y
)
-
I
_
1
(
x
,
y
)
I
2
(
x
,
y
)
-
I
_
2
(
x
,
y
)
(
∑
M
(
x
,
y
)
=
1
I
1
(
x
,
y
)
-
I
_
1
(
x
,
y
)
I
2
(
x
,
y
)
-
I
_
2
(
x
,
y
)
)
2
,
wherein Ī denotes the average intensity of pixels in the masked region of the image.
12. The method of claim 11 , wherein calculating the similarity measure between the aorta DRR and the DSA, and between the coronary artery DRR and the DSA, for a plurality of poses comprises:
calculating the similarity measure between downsampled aortic DRRs and a downsampled cardiac DSA image, and downsampled coronary artery DRRs and a downsampled cardiac DSA image, for a plurality of poses over a range of 2-dimensional translations;
selecting one or more DRR-pose combinations for the aorta and the coronary artery with largest similarity measures as refinement candidates;
calculating the similarity measure between the refinement candidate aortic DRRs and the DSA, and between the refinement candidate coronary artery DRRs and the DSA, for a plurality of poses over a range of 3-dimensional translations and in-plane rotations;
selecting one or more DRR-pose combinations for the aorta and the coronary artery with largest similarity measures as final candidates;
calculating the similarity measure between the final candidate aortic DRRs and the DSA, and between the final candidate coronary artery DRRs, for a plurality of poses over a range of 3-dimensional translations and 3-dimensional rotations; and
selecting a DRR-pose combination with a largest similarity measure as a final registration result.
13. 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 registering a 2-dimensional digital subtraction angiography (DSA) image to a 3-dimensional image volume during a cardiac procedure, the method comprising the steps of:
calculating a coarse similarity measure between a 2-dimensional digitally reconstructed radiograph (DRR) of an aorta and a cardiac DSA image, and a 2-dimensional DRR of a coronary artery and the cardiac DSA image, for a plurality of poses over a range of 2-dimensional translations;
selecting one or more DRR-pose combinations for the aorta and the coronary artery with largest similarity measures as refinement candidates;
calculating the similarity measure between the refinement candidate DRRs of the aorta and the DSA, and between the refinement candidate DRRs of the coronary artery and the DSA, for a plurality of poses over a range of 3-dimensional translations and in-plane rotations;
selecting one or more DRR-pose combinations for the aorta and the coronary artery with largest similarity measures as final candidates;
calculating the similarity measure between the final candidate DRRs of the aorta and the DSA, and between the final candidate DRRs of the coronary artery and the DSA, for a plurality of poses over a range of 3-dimensional translations and 3-dimensional rotations; and
selecting a DRR-pose combination with a largest similarity measure as a final registration result.
14. The computer readable program storage device of claim 13 , the method further comprising pre-processing the DSA image to subtract a background image, apply morphological operations, and remove artifacts of the morphological operations.
15. The computer readable program storage device of claim 13 , wherein the DRR of an aorta and the DRR of a coronary artery are generated from a same 3-dimensional cardiac image volume.
16. The computer readable program storage device of claim 15 , the method further comprising pre-processing the cardiac image volume to create 3-dimensional image masks for the aorta and the coronary artery, wherein said image masks are used to generate the DRR of the aorta and the DRR of the coronary artery.
17. The computer readable program storage device of claim 15 , wherein the DRR of an aorta and the DRR of a coronary artery are generated for a plurality of poses at a plurality of depths in the cardiac image volume.
18. The computer readable program storage device of claim 17 , wherein the DRR of an aorta and the DRR of a coronary artery are centered about an estimated center of the aortic root.
19. The computer readable program storage device of claim 13 , wherein the coarse similarity measure is calculated between a downsampled 2-dimensional DRR of the aorta, a downsampled 2-dimensional DRR of the coronary artery, and a downsampled cardiac DSA image.
20. The computer readable program storage device of claim 13 , wherein the similarity measure is defined by
SM=GC Aorta ( I Aorta ,I DSA ,M Aorta )+β· GC Coronary ( I Coronary ,I DSA ,M Coronary )
wherein I Aorta represents the aortic DRR image, I Coronary represents the coronary artery DRR image, I DSA represents the DSA image,
M Aorta represents a aorta image mask defined as
M
Aorta
=
{
1
,
if
∇
I
Aorta
>
α
,
0
,
otherwise
,
,
where α is a threshold of the image gradient,
M Coronary represents a coronary artery image mask defined as
M
Coronary
=
{
1
,
if
min
(
∇
I
Coronary
,
∇
I
DRR
)
>
α
,
0
,
otherwise
.
,
β is a heuristically determined weight for coronary ostia landmark features,
GC is a gradient correlation between images I 1 and I 2 with image mask M defined
as
GC
(
I
1
,
I
2
,
M
)
=
NCC
(
∂
I
1
∂
x
,
∂
I
2
∂
x
,
M
)
+
NCC
(
∂
I
1
∂
y
,
∂
I
2
∂
y
,
M
)
,
and NCC denotes the normalized cross correlation of the masked images defined as
NCC
(
I
1
,
I
2
,
M
)
=
∑
M
(
x
,
y
)
=
1
I
1
(
x
,
y
)
-
I
_
1
(
x
,
y
)
I
2
(
x
,
y
)
-
I
_
2
(
x
,
y
)
(
∑
M
(
x
,
y
)
=
1
I
1
(
x
,
y
)
-
I
_
1
(
x
,
y
)
I
2
(
x
,
y
)
-
I
_
2
(
x
,
y
)
)
2
,
wherein Ī denotes the average intensity of pixels in the masked region of the image.
21. The computer readable program storage device of claim 16 , wherein pre-processing the cardiac image volume to create 3-dimensional image masks for the aorta comprises:
segmenting the aorta in the cardiac image volume; and
defining the aortic mask by
VM
aorta
(
x
,
y
,
z
)
=
{
1
,
if
(
x
,
y
,
z
)
∈
aorta
,
0
,
otherwise
,
.
22. The computer readable program storage device of claim 21 , wherein pre-processing the cardiac image volume to create 3-dimensional image masks for the coronary artery comprises:
detecting left and right coronary ostia in the cardiac image volume;
centering two spherical masks VM l and VM r around the detected coronary ostia;
forming a mask VM out from the union of VM l and VM r , excluding an area inside VM aorta , and calculating an intensity distribution for voxels in the mask VM out ; and
choosing a lower bound of a volume transfer window associated with the coronary artery image mask as a given percentile of the intensity distribution, and an upper bound of the volume transfer window to be a highest voxel intensity of the volume.