Yujiao Zhao1,2, Jiahao Hu1,2, Vick Lau1,2, Linfang Xiao1,2, Alex T. L. Leong1,2, and Ed X. Wu1,2
1Laboratory of Biomedical Imaging and Signal Processing, The University of Hong Kong, Hong Kong, China, 2Department of Electrical and Electronic Engineering, The University of Hong Kong, Hong Kong, China
Synopsis
Keywords: System Imperfections: Measurement & Correction, New Signal Preparation Schemes
MRI scans are commonly performed
inside a fully-enclosed RF shielding room, posing stringent installation
requirement and unnecessary patient discomfort. This study develops a strategy
of active EMI sensing and deep learning MR signal prediction using residual U-Net
for RF shielding-free MRI. We implemented it on an ultra-low-field 0.055T head MRI
scanner. Our experimental results demonstrated that this strategy could directly
and accurately predict EMI-free MRI signals from the signals acquired by MRI
receive coil and EMI sensing coils. It worked robustly with strong and
dynamically varying EMI sources, yielding significantly improved brain image
quality.
Introduction
Clinical MRI scanners all require
bulky and enclosed RF shielding cage to prevent EMI during MRI scanning.
Several methods have been recently proposed to remove this shielding
requirement for portable ultra-low-field (ULF) MRI1-7. An analytical approach was proposed to estimate
EMI signal in MRI receive coil from EMI signals detected by EMI sensing coils
based on frequency domain transfer functions (TFs) among coils1,2, and further extended for time domain
implementation with an adaptive procedure4. Our recent study proposed a deep learning method
using a CNN model to better characterize the relationships among EMI signals
detected by EMI sensing coils and MRI receive coils, producing improved EMI reduction for shielding-free MRI
at 0.055T5. We also demonstrated its applicability at 1.5T6,7. However, EMI environments can be extremely complex
in practice. For example, EMI sources can be very strong, and vary temporally
and spatially. Patient body position change during scanning may also alter EMI behaviours
since body can act as an antenna8,9. Such practical issues can degrade the performance
of these newly developed EMI removal methods. For shielding-free MRI to become a
reality, it is imperative to develop more robust strategies.
In this study, a new EMI elimination strategy is developed for RF shielding-free MRI. We propose direct prediction of EMI-free MRI signals
from the signals detected by MRI receive coil and EMI sensing coils through a
residual U-Net based deep learning. We demonstrated its robust performance for
0.055T brain MRI under complex EMI conditions.Methods
Active EMI Sensing and Direct Deep
Learning MRI Signal Prediction
Fig. 1 illustrates the proposed
EMI elimination strategy. EMI sensing coils are strategically placed around and
within scanner to actively detect radiative EMI signals from both external
environments and internal electronics (Fig. 1A). Within each TR, MRI
receive coil and EMI sensing coils are used to simultaneously sample data
within two windows, one for conventional MRI signal acquisition, and the other for
acquiring EMI characterization data (i.e., EMI signals only) (Fig. 1B).
After each scan, we train a
residual U-Net model to directly predict the 1D temporal MRI signals (i.e., frequency
encoding or FE lines) from the signals acquired by both MRI receive coil and
EMI sensing coils during
EMI characterization window (Fig. 1C). Specifically, EMI-free k-space
MRI signals from 3T brain MRI data (or any publicly available data) are first added
into EMI-only signals received by MRI receive coil. Such synthesized
EMI-contaminated MRI signals and signals received by EMI sensing coils
are used as model inputs, while EMI-free MRI signals serve as model target. The
trained model is then applied to predict EMI-free MRI signals from the signals collected
during MRI signal acquisition window, creating EMI-free k-space data
prior to any averaging or/and image reconstruction.
Evaluation and Model
Implementation
Brain imaging experiments were conducted
on a shielding-free 0.055T MRI scanner with one MRI receive coil and ten
EMI sensing coils. T1W datasets were acquired using 3D GRE with BW = 6.25kHz (acquisition
matrix = 128×128×32 and NEX = 2). T2W and FLAIR-like datasets were acquired using
3D FSE with BW = 10kHz (T2W: acquisition matrix = 128×126×32 and NEX = 2; FLAIR: acquisition
matrix = 128×117×32 and NEX = 4).
A residual U-Net architecture10 containing 4 scales was employed.
Four successive residual blocks were used in each scale (Fig. 1C). 1D
temporal signals detected by MRI receive coil and EMI sensing coils were
treated as separate channels, and the number of channels in each layer from the
first to the fourth scale was 12, 32, 64, and 128, respectively. For each
dataset, a model was trained by minimizing L1 loss using Adam optimizer11 with batch size = 64 for 40
epochs and initial learning rate = 0.0002. Typical training time for each model
(i.e., each dataset) was around 8 mins on a Quadro RTX 8000 GPU and Intel Core
i9-10900X CPU. Results
Fig. 2 presents the performance of the proposed EMI
elimination method for simulated brain datasets, indicating nearly complete
removal of EMI signals. Moreover, no pseudo-structures were observed in error
maps, supporting that the MRI signals were accurately predicted. Figs. 3 to 5 show
EMI elimination results for human brain datasets acquired on a shielding-free 0.055T
MRI platform. The proposed strategy reliably eliminated all undesirable EMI signals
in all cases as seen in both images and spectral analyses, yielding
significantly improved image quality. Figs. 4 and 5 demonstrate the robust performance of the proposed method when EMI
sources were extremely strong and dynamically varied during scanning. In
contrast, the existing transfer function (TF) method2 produced incomplete EMI signal removal, leading to
residual EMI artifacts and increased noise level in the images. Discussion and Conclusions
This study presents a new deep
learning EMI elimination strategy for RF shielding-free MRI. It exploits the coupling
relationships among MRI receive coil and EMI sensing coils. In contrast to direct
EMI signal prediction as in recently developed methods1-7, our new method directly predicts MRI signals from
acquired MRI receive coil and EMI sensing coil signals, leading to minimal
noise or bias propagation. The experimental results demonstrate its robust
performance even in presence of strong and complex EMI sources.Acknowledgements
This work was supported in part by Hong Kong
Research Grant Council (R7003-19F, HKU17112120, HKU17127121 and HKU17127022 to
E.X.W., and HKU17103819, HKU17104020 and HKU17127021 to A.T.L.L.), Lam Woo
Foundation, and Guangdong Key Technologies for AD Diagnostic and Treatment of
Brain (2018B030336001) to E.X.W..References
[1] Rearick
T, Charvat GL, Rosen MS, Rothberg JM; Noise suppression methods and apparatus
patent US Patent No. 9,797,971. 2017.
[2] Dyvorne
H, Rearick T, Poole M, Lazarus C, Weiss P, Sacolick L, Jordan J, Hugon C,
Mileski W, Chen G, O'Halloran R, McNulty C, Lowthert J, Nelson A, Lorenzo A,
Zwart N, Kundu P, Martin S, Loutchouk A, Welch EB, By S, Cahn B, Yuen M, Mazurek
M, Pranhat A, Rosen M, Sheth K, Rothberg J. Freeing MRI from its Faraday cage
with Interference Rejection. In: Proceedings of the 29th Annual Meeting of
ISMRM, 2021, p 0749.
[3] SA;
S, CZ; C, JP; S, PC; M, LL. W. Retrospective electromagnetic interference
mitigation in a portable low field MRI system. In: Proceedings of the 28th
Annual Meeting of ISMRM, 2020, p 1269.
[4] Srinivas
SA, Cauley SF, Stockmann JP, Sappo CR, Vaughn CE, Wald LL, Grissom WA, Cooley
CZ. External Dynamic InTerference Estimation and Removal (EDITER) for low field
MRI. Magn Reson Med 2022;87(2):614-628.
[5] Liu
Y, Leong ATL, Zhao Y, Xiao L, Mak HKF, Tsang A, Lau GKK, Leung GKK, Wu EW. A
Low-cost and shielding-free ultra-low-field brain MRI scanner. Nat Commun
2021;12(1):1-14.
[6] Zhao
Y, Xiao L, Liu Y, Leong ATL, Wu EX. Deep Learning Driven EMI Prediction and
Elimination for RF Shielding-Free MRI at 0.055T and 1.5T. In: Proceedings of
the 31st Annual Meeting of ISMRM, 2022, p 3864.
[7] Zhao
Y, Xiao L, Lau V, Liu Y, Leong AT, Wu EX. Robust Electromagnetic Interference
(EMI) Elimination via Simultaneous Sensing and Deep Learning Prediction for RF
Shielding-free MRI. arXiv preprint arXiv:221006730 2022.
[8] Kibret
B, Teshome AK, Lai DT. Analysis of the human body as an antenna for wireless
implant communication. IEEE Transactions on antennas and propagation
2016;64(4):1466-1476.
[9] Sen
S, Maity S, Das D. The body is the network: to safeguard sensitive data, turn
flesh and tissue into a secure wireless channel. IEEE Spectrum 2020;57(12):44-49.
[10] Zhang
K, Li Y, Zuo W, Zhang L, Van Gool L, Timofte R. Plug-and-play image restoration
with deep denoiser prior. IEEE Trans Pattern Anal Mach Intell
2022;44(10):6360-6376.
[11] Kingma DP, Ba J. Adam: A method for
stochastic optimization. In: 3rd International Conference on Learning
Representations ICLR, San Diego, CA, USA, 2015.