Yan Wu1, Yajun Ma2, Jiang Du2, and Lei Xing1
1Radiation Oncology, Stanford University, Stanford, CA, United States, 2Radiology, University of California San Diego, La Jolla, CA, United States
Synopsis
Inhomogeneity
of the radiofrequency field (B1) is one of the main problems in quantitative
MRI. Leveraging from the unique ability of
deep learning, we propose a data driven strategy to derive quantitative B1 map from a
single qualitative MR image without specific requirements on the weighting of
the input image. B1 estimation is accomplished using a self-attention deep
convolutional neural network, which makes efficient use of local and non-local
information. Without additional data acquisition, an accurate estimation of B1
map is achieved, which is useful for the compensation of field inhomogeneity in
T1 mapping as well as for other applications.
INTRODUCTION
In quantitative MRI, inhomogeneity
of the radiofrequency field (B1) is one of the main sources of error. Measurement
of B1 map effectively compensates for field inhomogeneity; however, it takes
addition scan time. In this study, we propose a deep learning-based strategy to
estimate B1 map from a single qualitative MR image without specific requirement on the weighting
of the input image. In this way, B1 can be estimated
from an MR image acquired in routine clinical practice or biomedical research without
involvement of extra data acquisition. This will lay a solid foundation for the
derivation of quantitative T1 map as well as for other applications. In this
study, the method is mainly validated in cartilage MRI.METHODS
To provide an end-to-end mapping from a single MR image to the
corresponding B1 map, a convolutional neural network is employed. In the training
of the network, input images are single images with a specific weighting (T1, T1r,
or T2) acquired using an ultra-short TE sequence 1-3; and ground truth
B1 maps are obtained using the widely adopted actual flip angle method 4. With the
difference between the predicted maps and the ground truth backpropagated,
network parameters are updated using the Adam algorithm. This iterative
procedure continues until convergence is reached, as
illustrated in Figure 1. For a test image acquired using the same imaging
protocol, B1 map is automatically generated by the established network model.
A special convolutional neural
network is constructed for B1 estimation 5.
The network has a hierarchical architecture, composed of an encoder and a
decoder. This enables feature extraction at various scales while
enlarging the receptive field at the same time. A unique shortcut pattern is
designed, where global shortcuts (that connect the
encoder path and the decoder path) compensate for details lost in
down-sampling, and local shortcuts (that forward the input to a hierarchical
level of a single path to all subsequent convolutional blocks) facilitate
residual learning.
Attention mechanism is
incorporated into the network to make efficient use of non-local information 6-8. Briefly, in self-attention, direct
interactions are established between
all voxels within a given image, and more attention is focused on regions that
contain similar spatial information. In every convolutional block, a
self-attention layer is integrated, where the self-attention map is derived by
attending to all the positions in the feature map obtained in the previous
convolutional layer. The value at a position of the attention map is determined
by two factors. One is the relevance between the signal at the current position and that at other positions, defined
by an embedded Gaussian function. The other is a representation of the feature
value at the other position, given by a linear function. Here, weight matrices
are identified by the model in training. The proposed network is shown in
Figure 2.
Deep neural networks are trained for B1 estimation, each
taking single input images with a specific weighting (T1, T1r, or T2). A total
of 1,224 slice images from 51 subjects (including healthy volunteers and
patients) are used for model training, and 120 images of 5 additional subjects
are employed for model testing. RESULTS
Using established models, B1 maps are predicted from test
images with various weightings. Figure 3 shows a representative case. The B1
maps estimated from different input images all show high fidelity to the ground
truth map displayed in the leftmost column. DISCUSSION
In conventional quantitative MRI
approaches or MR fingerprinting, B1 map was ever
estimated from variable contrast images without extra data acquisition. However, it is the first time that
B1 map is estimated from a single MR image. This
is accomplished by using deep learning to exploit the relationship between B1
map and MR image, which is inherently caused by the electrodynamic interaction
between the incident transmission radiofrequency field and patient anatomy.
Estimation of B1 map is practically
useful. It not only lays a solid foundation for accurate T1 mapping, but also
provides information for other applications (e.g. the derivation of
electrical property tomography).CONCLUSION
We present a deep learning-based strategy for the estimation
of B1 map. Using
a properly trained deep learning model, B1 map can
be predicted from a single MR
image with
high accuracy achieved.Acknowledgements
This research is partially supported by NIH/NCI (1R01 CA176553), NIH/NIAMS (1R01 AR068987), NIH/NINDS (1R01 NS092650).References
1.
Y. J. Ma, W. Zhao, L. Wan, T. Guo, A. Searleman,
H. Jang, et al., "Whole knee
joint T1 values measured in vivo at 3T by combined 3D ultrashort echo time
cones actual flip angle and variable flip angle methods," Magnetic resonance in medicine, vol. 81,
pp. 1634-1644, 2019.
2.
Y. J. Ma, M. Carl, A. Searleman, X. Lu, E. Chang,
and J. Du, "3D adiabatic prepared ultrashort echo time cones sequence
for whole knee imaging," Magnetic Resonance in Medicine, vol. 80,
pp. 1429-1439, 2018.
3.
Du,
J., Diaz E, Carl M, Bae W, Chung CB, Bydder GM, Ultrashort echo time imaging with bicomponent analysis. Magnetic
resonance in medicine, 2012. 67(3): p. 645-649.
4.
Yarnykh and V. Yarnykh, Actual flip-angle imaging in the pulsed steady state: A method for
rapid three-dimensional mapping of the transmitted radiofrequency field.
Magnetic Resonance in Medicine, 2007. 57(1):
p. 192-200.
5.
Y. Wu, Y. Ma, D. P. Capaldi, J. Liu, W. Zhao, J.
Du, et al., "Incorporating prior
knowledge via volumetric deep residual network to optimize the reconstruction
of sparsely sampled MRI," Magnetic resonance imaging, 2019.
6.
Wu, Y., Y. Ma, J. Liu, W. Zhao, J. Du et al., Self-attention convolutional neural network
for improved MR image reconstruction. Information Sciences, 2019. 490: p. 317-328.
7.
Vaswani, A., et al. Attention is all you need. in Advances
in Neural Information Processing Systems. 2017.
8.
Zhang, H., et al., Self-attention generative adversarial networks. arXiv preprint
arXiv:1805.08318, 2018.