Method for automated regularization of hybrid K-space combination using a noise adjustment scan
The present disclosure is generally directed to systems and methods for generating de-noised MR images that are reconstructed from a hybridization of two separate image reconstruction pipelines, at least one of which includes the use of a neural network. Further, the amount of influence that the neural network reconstruction has on the hybrid reconstructed image is controlled via a regularization parameter that is selected based on an estimated noise level associated with the initial image acquisition, which can be calculated from pre-scan data.
1 . A computer-implemented method for improving images captured via magnetic resonance imaging (MRI), the method comprising:
receiving, by a computer system, MRI data from an MRI machine, the MRI data comprising image data;
calculating, by the computer system, a noise level associated with the MRI data;
selecting, by the computer system, a regularization parameter corresponding to the calculated noise level;
processing, by the computer system, the image data through a first reconstruction pipeline to output a first reconstructed image;
processing, by the computer system, the image data through a second reconstruction pipeline to output a second reconstructed image, wherein the second reconstruction pipeline comprises a neural network; and
reconstructing, by the computer system, a hybrid image from the first reconstructed image and the second reconstructed image, wherein an amount that the second reconstructed image contributes to the hybrid image corresponds to the selected regularization parameter, wherein the regularization parameter is selected to cause the hybrid image to be de-noised relative to the image data,
wherein the hybrid image is reconstructed from a weighted combination of the first reconstructed image and the second reconstructed image and the reconstructing the hybrid image includes calculating, by the computer system, the hybrid image based on a measured k-space vector, a k-space sampling, a Fourier transform, a coil sensitivity corresponding to the MRI machine, a weighting matrix, the second reconstructed image, the image data, and the selected regularization parameter.
2 . The computer-implement method of claim 1 , wherein the first reconstruction pipeline comprises a sensitivity encoded (SENSE) reconstruction.
3 . The computer-implemented method of claim 1 , wherein selecting the regularization parameter comprises:
querying, by the computer system, a database comprising a plurality of regularization parameters indexed to a plurality of noise levels; and
selecting, by the computer system, the regularization parameter from the plurality of regularization parameters that corresponds to the calculated noise level.
4 . The computer-implemented method of claim 1 , wherein selecting the regularization parameter comprises:
calculating, by the computer system, the regularization parameter from a fitted parametric model relating the calculated noise level to the regularization parameter.
5 . The computer-implement method of claim 1 , wherein selecting the regularization parameter is subject to user-customized preferences.
6 . The computer-implement method of claim 1 , wherein reconstructing the hybrid image comprises
calculating, by the computer system, the hybrid image according to:
ρ
ˆ
=
argmin
ρ
d
-
Ω
FC
ρ
2
2
+
λ
WFC
(
ρ
n
e
t
-
ρ
)
2
2
wherein {circumflex over (p)} is the hybrid image, d is the measured k-space data vector, Ω is the k-space sampling, F is the Fourier transform, C is the coil sensitivity corresponding to the MRI machine, W is the weighting matrix, ρ net is the second reconstructed image, ρ is the image data, and λ is the selected regularization parameter.
7 . The computer-implement method of claim 1 , wherein the selected regularization parameter is varied spatially in the hybrid image.
8 . The computer-implement method of claim 1 , wherein the neural network comprises a deep neural network.
9 . The computer-implement method of claim 1 , wherein the neural network comprises a physics-informed network.
10 . The computer-implemented method of claim 1 , wherein:
the MRI data further comprises pre-scan data acquired without any generated MR signals; and
the noise level is calculated from the pre-scan data.
11 . The computer-implemented method of claim 1 , wherein the noise level comprises a signal-to-noise ratio.
12 . A magnetic resonance imaging (MRI) system for capturing images, the system comprising:
an MRI machine; and
a computer system coupled to the MRI machine, the computer system comprising a processor and a memory, the memory storing instructions that, when executed by the processor, cause the computer system to:
receive MRI data from the MRI machine, the MRI data comprising image data;
calculate a noise level associated with the MRI data;
select a regularization parameter corresponding to the calculated noise level;
process the image data through a first reconstruction pipeline to output a first reconstructed image;
process the image data through a second reconstruction pipeline to output a second reconstructed image, wherein the second reconstruction pipeline comprises a neural network; and
reconstruct a hybrid image from the first reconstructed image and the second reconstructed image, wherein an amount that the second reconstructed image contributes to the hybrid image corresponds to the selected regularization parameter, wherein the regularization parameter is selected to cause the hybrid image to be de-noised relative to the image data,
wherein the hybrid image is reconstructed from a weighted combination of the first reconstructed image and the second reconstructed image and the reconstructing the hybrid image includes calculating, by the computer system, the hybrid image based on a measured k-space vector, a k-space sampling, a Fourier transform, a coil sensitivity corresponding to the MRI machine, a weighting matrix, the second reconstructed image, the image data, and the selected regularization parameter.
13 . The MRI system of claim 12 , wherein the first reconstruction pipeline comprises a sensitivity encoded (SENSE) reconstruction.
14 . The MRI system of claim 12 , wherein the memory stores instructions that, when executed by the processor, cause the computer system to select the regularization parameter by:
querying a database comprising a plurality of regularization parameters indexed to a plurality of noise levels; and
selecting the regularization parameter from the plurality of regularization parameters that corresponds to the calculated noise level.
15 . The MRI system of claim 12 , wherein the memory stores instructions that, when executed by the processor, cause the computer system to select the regularization parameter by:
calculating, by the computer system, the regularization parameter from a fitted parametric model relating the calculated noise level to the regularization parameter.
16 . The MRI system of claim 12 , wherein the memory stores instructions that, when executed by the processor, cause the computer system to select the regularization parameter subject to user-customized preferences.
17 . The MRI system of claim 12 , wherein reconstructing the hybrid image comprises
calculating, by the computer system, the hybrid image according to:
ρ
ˆ
=
argmin
ρ
d
-
Ω
FC
ρ
2
2
+
λ
WFC
(
ρ
n
e
t
-
ρ
)
2
2
wherein {circumflex over (p)} is the hybrid image, d is the measured k-space data vector, Ω is the k-space sampling, F is the Fourier transform, C is the coil sensitivity corresponding to the MRI machine, W is the weighting matrix, ρ net is the second reconstructed image, ρ is the image data, and λ is the selected regularization parameter.
18 . The MRI system of claim 12 , wherein the selected regularization parameter is varied spatially in the hybrid image.
19 . The MRI system of claim 12 , wherein the neural network comprises a deep neural network.
20 . The MRI system of claim 12 , wherein the neural network comprises a physics-informed network.
21 . The MRI system of claim 12 , wherein:
the MRI data further comprises pre-scan data acquired without any generated MR signals; and
the noise level is calculated from the pre-scan data.
22 . The MRI system of claim 12 , wherein the noise level comprises a signal-to-noise ratio.