Lorenzo Di Sopra1, Jérôme Yerly1,2, Christopher W. Roy1, Juerg Schwitter3, and Matthias Stuber1,2
1Department of Diagnostic and Interventional Radiology, Lausanne University Hospital (CHUV) and University of Lausanne (UNIL), Lausanne, Switzerland, 2Center for Biomedical Imaging (CIBM), Lausanne, Switzerland, 3Division of Cardiology and Cardiac MR Center, Lausanne University Hospital (CHUV) and University of Lausanne (UNIL), Lausanne, Switzerland
Synopsis
Cardiac self-gating (SG) approaches for
free-running MR imaging of the heart have proven to be a robust alternative to ECG-gating.
However, for both the ECG and SG signals, one feature is typically extracted
(ECG: R-wave, SG: zero-crossing) and used for triggering, gating, or data
binning. While this often works for constant heart rates, periods of quiescence
cannot easily be predicted when heart rates change. Therefore, we have
developed an algorithm that, without any prior knowledge, identifies periods of
cardiac quiescence from SG signals, and demonstrated that resultant
motion-suppressed image quality matches that from conventional approaches.
Introduction
Cardiac
self-gating (SG) approaches for free-running MRI of the heart[1,2] have
proven to be a robust and precise alternative to standard ECG gating that is
based on R-wave detection.[3] However, such SG methods typically only generate
temporal triggers for binning the k-space data throughout the cardiac cycle, in contrast to respiratory SG
signals, whose amplitude’s temporal evolution is exploited to inform binning
for motion-resolved or motion-registered reconstructions.[1-5] In this
study, we investigated the hypothesis that the intervals of least variability of
cardiac SG signals correspond to the resting period of the cardiac cycle during
diastole. To this end, free-running data sets were acquired in healthy
volunteers and patients, and two methods for retrospective reconstruction of 3D
static images were investigated. First, conventional mid-diastolic data subset
selection and reconstruction was performed using constant window width and subject-specific
time delay after the trigger point. Second, using an adaptive, non-constant
window width and time delay determined by the intervals of least variability of
the cardiac SG signal for each individual cardiac cycle. Timing of selected
data and quality of motion-suppression in reconstructed images were then quantitatively
compared for both techniques.Methods
This IRB
approved study included 9 healthy volunteers (6F, 27.4±2.4y) and 5 patients (5M, 58.2±6.1y) who provided written informed
consent. All examinations were performed on a clinical 1.5T MRI scanner
(MAGNETOM Aera, Siemens Healthcare, Erlangen, Germany) with a prototype bSSFP free-running
(non-ECG-triggered) 3D golden-angle radial sequence with (1.15mm)3
isotropic spatial resolution.[1,4] Both respiratory
and cardiac motion signals were extracted from the imaging data with a
previously reported SG technique,[1] and reconstructed as cardiac and
respiratory motion-resolved 5D images using compressed sensing[3] in
order to visually confirm periods of quiescent cardiac motion. 3D static images
were then reconstructed with 2 different approaches, both aiming at selecting
about 12% (=30%×40% along the two
independent cardiac and respiratory dimensions, respectively) of all acquired readouts, corresponding to 25% of the radial
Nyquist limit, similar to previous 3D radial cardiac
imaging techniques.[4,6] For the first
approach, a temporal window for readout selection was set with a fixed delay
from the SG cardiac triggers, using the knowledge obtained from the 5D images, to reconstruct a diastolic 3D static image IFIXED. For the second approach, without prior
knowledge of the temporal localization of the diastolic phase within the
acquired data, the defined fraction of all readouts was selected from the most
quiescent portion of each cardiac cycle to reconstruct a 3D IADAPT
image. To select the intervals of least variability of the cardiac SG signal $$$c$$$, we defined the following weighting function $$$w_{c}$$$:
$$w_{c}\left(i\right)=\sqrt{\left(\frac{c\left(i\right)-\bar{c}}{SD\left(c\left(i\right)-\bar{c}\right)}\right )^{2}+\left(\frac{{c}'\left(i\right)}{SD\left({c}'\left(i\right)\right)}\right)^{2}+\left(\frac{{c}''\left(i\right)}{SD\left({c}''\left(i\right)\right)}\right)^{2}}$$
where $$$i$$$
is the readout index, $$$\bar{c}$$$
is the mode of $$$c$$$
(i.e. most frequent value of $$$c$$$), $$${c}'$$$ and $$${c}''$$$ are the first and second derivatives of $$$c$$$, respectively, and $$$SD$$$
is the standard deviation. The 30% of readouts with the lowest weight
were then selected (Fig.1). As for
respiration, for both IFIXED and IADAPT, the same 40% of
end-expiratory data were selected based on the respiratory SG signal amplitude
as previously reported.[1-3,5] To test for
consistency between the period of cardiac quiescence selected using a priori
knowledge in IFIXED and the automatic and adaptive selection in IADAPT,
average delay and standard deviation from the corresponding cardiac SG triggers
are reported. Then, image sharpness was measured for both reconstruction
methods across all subjects using the slope of a sigmoid function fitted to multiple
manually selected points at the left ventricular blood-myocardium interface and was statistically
compared using a t-test with p<0.05 considered significant.[7]Results
On average and
across all subjects, the temporal windows with a fixed delay for IFIXED
reconstructions were retrospectively selected at 604±74 ms
after the cardiac SG triggers, while the readouts selected for IADAPT
reconstructions were found at 594±94 ms (p=0.29) (Fig.2). Conversely, the readouts
selected by the fixed window showed a significantly smaller deviation from the
average trigger delay than the corresponding selection from the adaptive
approach (85±9 ms vs 113±14 ms, p<0.0001) (Fig.3). The left ventricular
blood-myocardium interface sharpness measurements showed highly consistent results
with IFIXED =2.71±0.31 and IADAPT =2.71±0.34 (p=0.99)
(Fig.4-5).Discussion
The proposed data
selection procedure was able to robustly identify the resting phase of the
cardiac cycle from the cardiac SG signal without image-based or ECG-derived a
priori knowledge. The average, dynamic trigger delay that was automatically
extracted for IADAPT matched that from IFIXED very
closely. The only difference was found in the delay variability of selected
data, which was significantly higher for IADAPT due to the automated
selection procedure that dynamically adapts to the variability of individual cardiac
cycle durations throughout the entire acquisition. Blood-myocardium sharpness measurements
also confirmed the effectiveness of the adaptive data selection as it perfectly
corroborated the values measured for the fixed-window reconstructions.Conclusion
This study
suggests that there is currently unused information in cardiac SG signals that
could be exploited to maximize the quality of motion-resolution in free-running
cardiac images. This may help inform an adaptive, patient-specific binning strategy
throughout the cardiac cycle that aims at optimizing cardiac motion suppression
for all reconstructed phases, which may be of particular benefit for clinical
cohorts displaying increased heart-rate variability.Acknowledgements
No acknowledgement found.References
1. Di Sopra L, Piccini D, Coppo
S, Stuber M, Yerly J. An automated approach to fully self-gated free-running
cardiac and respiratory motion-resolved 5D whole-heart MRI. Magnetic resonance
in medicine 2019;82(6):2118-2132.
2. Pang J, Sharif B, Fan Z, Bi X, Arsanjani R,
Berman DS, Li D. ECG and navigator-free four-dimensional whole-heart coronary
MRA for simultaneous visualization of cardiac anatomy and function. Magnetic
resonance in medicine 2014;72(5):1208-1217.
3. Feng L, Axel L, Chandarana H, Block KT,
Sodickson DK, Otazo R. XD-GRASP: Golden-angle radial MRI with reconstruction of
extra motion-state dimensions using compressed sensing. Magnetic resonance in
medicine 2016;75(2):775-788.
4. Coppo S, Piccini D, Bonanno G, Chaptinel J,
Vincenti G, Feliciano H, van Heeswijk RB, Schwitter J, Stuber M. Free-running
4D whole-heart self-navigated golden angle MRI: Initial results. Magnetic
resonance in medicine 2015;74(5):1306-1316.
5. Feng L, Coppo S, Piccini D, Yerly J, Lim RP,
Masci PG, Stuber M, Sodickson DK, Otazo R. 5D whole-heart sparse MRI. Magnetic
resonance in medicine 2017.
6. Piccini D, Littmann A,
Nielles-Vallespin S, Zenge MO. Respiratory
self-navigation for whole-heart bright-blood coronary MRI: methods for robust
isolation and automatic segmentation of the blood pool. Magnetic resonance in
medicine 2012;68(2):571-579.
7. Ahmad R, Ding Y,
Simonetti OP. Edge
Sharpness Assessment by Parametric Modeling: Application to Magnetic Resonance
Imaging. Concepts Magn Reson Part A Bridg Educ Res 2015;44(3):138-149.