High spatial and temporal resolution dynamic contrast-enhanced magnetic resonance imaging
View Patent ↗A method for high spatial and temporal resolution dynamic contrast enhanced magnetic resonance imaging using a random subsampled Cartesian k-space using a Poisson-disk random pattern acquisition strategy and a compressed sensing reconstruction algorithm incorporating magnitude image subtraction is presented. One reconstruction uses a split-Bregman minimization of the sum of the L1 norm of the pixel-wise magnitude difference between two successive temporal frames, a fidelity term and a total variation (TV) sparsity term.
1. A method of contrast enhanced magnetic resonance angiography, the method comprising:
(a) acquiring a first series of magnetic resonance imaging data sets of a subject from a scanner without contrast by randomly under-sampling Cartesian k-space;
(b) acquiring a second series of magnetic resonance imaging data sets of a subject from a scanner in the presence of contrast by randomly under-sampling Cartesian k-space; and
(c) reconstructing images from the acquired data sets with a split-Bregman minimization of the sum of the L1 norm of the pixel-wise magnitude difference between two successive temporal frames, a fidelity term and a total variation (TV) sparsity term.
2. A method as recited in claim 1 , further comprising performing a preliminary low definition reconstruction to assess timing before performing a high definition reconstruction to produce final magnetic resonance images.
3. A method as recited in claim 2 , wherein said first series and said second series of magnetic resonance data sets are acquired using an acquisition technique selected from the group consisting of time-resolved imaging with stochastic trajectories (TWIST), time-resolved imaging of contrast kinetics (TRICKS) and time resolved angiography using keyhole (TRAK).
4. A method as recited in claim 1 , further comprising optimizing weighting parameters by retrospectively under-sampling the k-space and minimizing root mean square error between original k-space images and reconstructed images.
5. A method as recited in claim 4 , further comprising normalizing acquired data by maximum image intensity prior to image reconstruction.
6. A method as recited in claim 1 , wherein said random under-sampling of Cartesian k-space uses variable density Poisson-disk random patterns.
7. A method as recited in claim 1 , wherein said reconstruction further comprises minimizing cost function for a volume by dividing a cost function into multiple Bregman distances and updating the distances according to a norm and a threshold.
8. A method as recited in claim 1 , wherein said reconstruction from said datasets comprises solving:
(
I
1
,
I
2
)
=
argmin
{
U
1
F
(
I
1
)
-
K
1
2
2
+
λ
TV
(
I
1
)
+
μ
I
2
e
i
ϕ
1
-
I
1
1
U
2
F
(
I
2
)
-
K
2
2
2
+
λ
TV
(
I
2
)
+
μ
I
2
-
I
1
e
i
ϕ
2
1
}
wherein fidelity terms ∥U 1 F(I 1 )−K 1 ∥ 2 2 and ∥U 2 F(I 2 )−K 2 ∥ 2 2 compare the Fourier transform (F) of the estimated images (I 1 and I 2 ) to measured samples (K 1 and K 2 ) using sampling masks (U 1 and U 2 );
wherein the total variation terms λTV(I 1 ) and λTV(I 2 ) are the L1-norm of the discrete gradients of I 1 and I 2 ; and
wherein the ∥I 2 |e iφ 1 −I 1 | 1 and |I 2 −|I 1 |e iφ 2 | 1 terms are the L1 norm of the pixel-wise magnitude subtracted image: ∥I 2 |−|I 1 ∥ 1 =∥I 2 |e iφ 1 −I 1 | 1 =|I 2 −I 1 |e iφ 2 | 1 .
9. A method of contrast enhanced magnetic resonance angiography, the method comprising:
(a) acquiring a first series of magnetic resonance imaging data sets of a subject from a scanner without contrast by randomly under-sampling Cartesian k-space using variable density Poisson-disk random patterns;
(b) acquiring a second series of magnetic resonance imaging data sets of a subject from a scanner in the presence of contrast by randomly under-sampling Cartesian k-space using variable density Poisson-disk random patterns; and
(c) reconstructing images from the acquired data sets by solving:
(
I
1
,
I
2
)
=
argmin
{
U
1
F
(
I
1
)
-
K
1
2
2
+
λ
TV
(
I
1
)
+
μ
I
2
e
i
ϕ
1
-
I
1
1
U
2
F
(
I
2
)
-
K
2
2
2
+
λ
TV
(
I
2
)
+
μ
I
2
-
I
1
e
i
ϕ
2
1
}
wherein fidelity terms ∥U 1 F(I 1 )−K 1 ∥ 2 2 and ∥U 2 F(I 2 )−K 2 ∥ 2 2 compare the Fourier transform (F) of the estimated images (I 1 and I 2 ) to measured samples (K 1 and K 2 ) using sampling masks (U 1 and U 2 );
wherein the total variation terms λTV(I 1 ) and λTV(12) are the L1-norm of the discrete gradients of I 1 and I 2 ; and
wherein the ∥I 2 |e iφ 1 −I 1 | 1 and |I 2 −|I 1 |e iφ 2 | 1 terms are the L1 norm of the pixel-wise magnitude subtracted image: ∥I 2 |−|I 1 ∥ 1 =∥I 2 |e iφ 1 −I 1 | 1 =|I 2 −I 1 |e iφ 2 | 1 .
10. A method as recited in claim 9 , further comprising optimizing weighting parameters by retrospectively under-sampling the k-space and minimizing root mean square error between original k-space images and reconstructed images.
11. A method as recited in claim 10 , further comprising normalizing acquired data by maximum image intensity prior to image reconstruction.
12. A method as recited in claim 10 , further comprising: plotting a sampling trajectory on a ky-kz plane in k-space; and sampling k-space at the beginning of each temporal frame to motion sensitivity and eddy currents that can affect imaging.
13. A method as recited in claim 12 , wherein said sampling trajectory is a spiral pattern centered at the center of k-space.