Paul Sprenger1,2, Kosuke Morita3, Takeshi Nakaura4, Masami Yoneyama5, Shuo Zhang2,6, Maike Bode2, Christiane K. Kuhl2, and Nils A. Kraemer2
1Faculty of Medicine, RWTH Aachen University, Aachen, Germany, 2Diagnostic and Interventional Radiology, University Hospital RWTH Aachen, Aachen, Germany, 3Kumamoto University Hospital, Kumamoto, Japan, 4Graduate School of Medical Sciences, Kumamoto University, Kumamoto, Japan, 5Philips Japan, Tokyo, Japan, 6Philips GmbH DACH, Hamburg, Germany
Synopsis
Compressed sensing technologies have recently
become commercially available that serve the purpose to shorten the image
acquisition time for morphological imaging. Here we explore
the reconstruction algorithm that
allows the combination of wavelet transformation of compressed sensing with
coil information of sensitivity encoding (SENSE) in body diffusion MRI based on
2D single-shot EPI and uniform undersampling, and report our results in both
healthy volunteers and patients on both 3T and 1.5T, in comparison to the
conventional SENSE parallel imaging technique.
Introduction
Compressed sensing is a new
and promising technique to accelerate the imaging process by reconstruction of
a magnetic resonance (MR) image from non-uniformly undersampled k-space data 1.
The clinical benefit in morphological imaging with combination of sensitivity
encoding coil information has been recently described 2. We aimed to investigate the
technical feasibility and clinical utility of employing compressed sensing reconstruction
in two-dimensional (2D) single-shot echo-planar imaging (EPI) with a uniform
k-space undersampling strategy in body diffusion MR imaging on both 3T and
1.5T, with comparison to the conventional parallel imaging reconstruction.Methods
All human subjects underwent MRI on a 3T and 1.5T systems (Ingenia,
Philips, Best, the Netherlands). Body diffusion MRI was performed using 2D
multi-slice single-shot echo-planer imaging (EPI) sequences in the liver,
prostate and breast. Conventional parallel imaging was done using sensitivity
encoding (SENSE). For compressed sensing, uniform k-space undersampling was
applied and the subsampled data were reconstructed
with a combination of wavelet transformation and SENSE coil information (compressed
SENSE or C-SENSE) 2. While spatial resolution and b-value
were adapted for each anatomy, most imaging parameters were kept the same
between conventional SENSE and compressed SENSE. Detailed imaging parameters
were summarized in Table 1.
For image analysis, visual assessment was performed
for all anatomies. In addition, quantitative assessment was done for comparison
of the signal-to-noise ratio (SNR) and apparent
diffusion coefficient (ADC) in liver diffusion-weighted EPI (DW-EPI) of healthy
volunteers on 1.5T. For SNR measurement additional noise maps were acquired
without radio frequency (RF) excitation for each DW-EPI scan. Regions of interest (ROIs) were selected in the
liver parenchyma with uniform signal without vessels or artifacts as well as in
the corresponding noise map. The same criteria of ROI selection also applied to
ADC quantification. Values were reported in mean and standard deviation (SD). For
statistics paired t-test was used and
a p value <0.001 was considered
significant.Results and Discussion
12 healthy volunteers (32 ± 7 years
[mean ± SD], 6 female) and 10 patients (54 ± 14 years, 3 female) were included. In
general, reduced background noise and clearer structural delineation can be
visualized with C-SENSE reconstruction, compared to conventional SENSE. This
applied to all three anatomies on both field strengths. Selected examples are presented
in Figure 1 to 3. In particular, C-SENSE
clearly reduced the noise in the center of the SENSE-based DWI (arrows in Figure 1 and 2),
which was associated with g-factor penalties and thus obscured an accurate lesion
detection and cancer interpretation. This improvement most likely benefits from the iterative de-noising
process in the wavelet domain, in which the noise patterns are sparsely
represented.
For liver DW-EPI,
imaging with different C-SENSE acceleration factors was conducted. Typical
results with a C-SENSE factor of 3 were shown in Figure 4A. In comparison to conventional SENSE with a factor of 2, SNR
measurement showed almost a two-fold benefit for both b values of 0 and 800 s/mm2. On
the other hand, while ADC quantification revealed similar values between
C-SENSE and conventional SENSE (p =
0.006), the former presented a significantly lower SD than that of the latter (p < 0.001) (Figure 4B). This lower variability may again attribute to a higher
signal in the C-SENSE images and may indicate a higher precision and reproducibility
in ADC assessment. Such quantitative findings were in consistency with the qualitative
observation above and were also in
accordance to the previous reports 3,4. While further studies
of its clinical performance are warranted, a systematic investigation in
different undersampling variants, noise behaviors and image quality
characteristics may help to identify standardized approaches for determining
optimal imaging parameters in clinical routine exams 5.Conclusion
2D multi-slice single-shot
EPI can be combined with the recently introduced compressed SENSE technology for
body diffusion MRI. Though uniformly undersampled, the results have shown that
an iterative de-noising process in the
wavelet domain in combination with coil sensitivity may help not only in
shortening image acquisition but also in image quality improvement, compared
to the conventional SENSE reconstruction. The achieved better SNR and less
variable ADC may benefit lesion detection in various applications, and this
needs to be further investigated in clinical studies with larger cohorts.Acknowledgements
The authors thank Stephanie Tackenberg, Lucia Noël und
Dega Abdi for their helps in image acquisition.References
-
Lustig M, et al. Magn Reson Med (2007) 58:1182
- Sartoretti E et al. Plos One (2019) 14:e0214887.
- Yoneyama M, et al. ISMRM, Montreal (2019) 27:1634.
- Morita K, et al. ISMRM, Montreal (2019) 27:3355.
- Zijlstra F et al. Invest Radiol (2016) 51:410.