Sajith Rajamani1, Florintina C1, Rajagopalan Sundaresan1, Ashok Kumar P Reddy1, Arjun Narula2, Mayuri Limbachiya2, Suresh Joel1, and Ramesh Venkatesan1
1GE HealthCare, Bangalore, India, 2Narula Diagnostics, Rohtak, India
Synopsis
Motivation: To explore the feasibility of multi-echo susceptibility weighted imaging at 0.5T as lower field strength MRI scanners are cost effective.
Goal(s): Assess the effectiveness of swi at lower field strength as it is one of the important sequences in Brain MRI
Approach: This study involves acquisition of 3D multi-echo gradient echo sequence to generate SWI with combination of protocol optimization specifically for 0.5T and Denoising both real and imaginary images using deep learning algorithm to boost SNR
Results: Findings shows SWI at 0.5T can provide information about susceptibility variations in brain and helps improve diagnosis.
Impact: SWI at 0.5T using deep learning-based reconstruction will increase the quality of image by improving SNR,
reducing artefacts imparted by the noise in phase.
Introduction
Susceptibility weighted imaging (SWI) is an essential
sequence for brain to study micro hemorrhages, calcifications, and
intravascular thrombosis [1]. It is a 3D gradient echo sequence that leverages
T2* effects in combination with SWI processing by utilizing the phase
information to generate contrast based on the susceptibility variations due to
dia and para magnetic components [2-3]. Since susceptibility effects are lower
at low field strengths such as 0.5T, optimization of protocol parameters such
as TR, TE & Flip angle are required to generate adequate susceptibility
image contrast. Through simulation shown in Figure1, we found the optimal flip
angle at 0.5T for swi contrast is 45° to ensure that CSF is neither too
bright nor dark. This will be useful to detect edema which appear bright, and
hemorrhages that appear dark (signal void). Simulations were based on the T1
and T2* values of gray matter, white matter, and CSF at 0.5T obtained from literature
[4]. Another disadvantage in going to lower field strength is reduced SNR. This
loss in SNR can be compensated by separately denoising both real &
imaginary parts of SWI using our proposed deep learning method.Materials & Methods
GE 1.5T Signa Creator scanner ramped down to 0.5T and 14
channel receive only Head Neck Coil was used to image brain. 10 patients were
scanned at a diagnostic centre with conventional clinical MRI scan on the
commercial 1.5T scanner along with 3 to 6 clinically relevant series including
3D SWAN sequence on the 0.5T scanner after obtaining informed consent approved
by IRB. Protocol was optimized on 3 healthy volunteers and the scan parameters
used for 3D SWAN after optimization are FOV:24cm, Slice thickness: 3mm, TR:
149ms, TE: 108ms, Flip angle: 45°, No. of echoes:7, Receiver Bandwidth:
+/- 15.6kHz. ARC acceleration factor of 2 was applied to reduce scan time. The
scan duration was 5 minutes. Phase filtering was performed after channel
combination from all 14 channels. Echo combination was performed for magnitude
and phase separately and then combined in the complex plane. Data was then passed
through deep learning (DL) algorithms for denoising both magnitude and phase.
This denoised phase is used to create phase mask and multiplied 5 times with
magnitude to generate final SWI processed image.Results & Discussion
The image quality of SWI images was found to be adequate for
diagnosis in all 10 subjects. The image quality improved significantly by
optimizing the acquisition (Figure 2) when compared to acquisition used with
1.5T protocol without any optimization.
Figure3 shows a sample SWI image after DL Processing. This
patient has multiple cerebral micro bleeds, and it is preserved in the 0.5T
image. The same findings are appearing as signal voids in filtered phase image
because of the presence of deoxyhemoglobin which causes negative phase shift.
Since susceptibility effect is less in 0.5T, we need to use
longer echo time up to 108ms. with multiple echoes, the SNR of later echoes are
poor. Aggressive denoising of the low SNR echoes will result blurring. When the
blurred echoes are later combined, the final IQ is reduced. Hence, we propose
to combine the echoes first, by combining the magnitude and phase images
separately and brought back to complex domain, where the denoising and
super-resolution is performed. In SWI, as phase is raised to a high power and
multiplied with magnitude image to accentuate the bleeds and minerals, even
minimal noise added back to the phase will appear as hypointense spots. Hence
phase images are 100% denoised, but a percentage of residue can be added to
provide texture to magnitude image.Conclusion
Susceptibility weighted imaging at 0.5T is feasible and
provides adequate diagnostically useful information with protocol optimization.
Low SNR due to reduced field strength can be compensated by applying deep
learning algorithms for denoising both real and imaginary images to produce
final susceptibility weighted image.Acknowledgements
No acknowledgement found.References
- ACR–ASNR–SPR Practice parameter for the
performance and interpretation of magnetic resonance imaging (mri) of the brain
- Haacke, E. M., Xu, Y., Cheng, Y. C., and
Reichenbach, J. R. (2004). Susceptibility weighted imaging (SWI). Magn. Reson.
Med. 52, 612–618.
- Reichenbach JR, Venkatesan R, Schillinger DJ,
Kido DK, Haacke EM. Small vessels in the human brain: MR venography with
Deoxyhemoglobin as an intrinsic contrast agent. Radiology. 1997
Jul;204(1):272-7. doi: 10.1148/radiology.204.1.9205259.
- Adrienne
E. Campbell-Washburn et.al Opportunities in Interventional and Diagnostic
Imaging by Using High-Performance Low-Field-Strength MRI. Radiology
2019 293:2, 384-393.