Pierre Daudé1, Rajiv Ramasawmy1, Ahsan Javed1, Robert J Lederman2, Kelvin Chow3, and Adrienne Campbell-Washburn1
1Laboratory of Imaging Technology, National Heart, Lung & Blood Institute, NIH, Bethesda, MD, United States, 2Cardiovascular Branch, National Heart, Lung & Blood Institute, NIH, Bethesda, MD, United States, 3Siemens Healthcare Ltd., Calgary, AB, Canada
Synopsis
Keywords: Data Acquisition, Low-Field MRI, MR value
Motivation: Conventional fixed duration acquisitions can result in patient-dependent image quality, leading to either unnecessarily long scan times or insufficient quality across patients.
Goal(s): We propose an inline automatic quality control based on signal-to-noise ratio (SNR) to achieve consistent diagnostic image quality and apply it to 2D phase-contrast flow MRI.
Approach: We designed a closed-loop feedback framework between image reconstruction and data acquisition to automatically stop the acquisition when a target SNR is achieved. Ten healthy volunteers and one patient were imaged at 0.55T.
Results: Deployed inline, the SNR stop threshold saved 53% of the scan duration, with a variation of ±1min across subjects.
Impact: The
inline automatic quality control enables a subject-specific optimized scan time
while ensuring consistent diagnostic image quality. The distribution of
automated stopping times across the population revealed the value of a
subject-specific scan time.
Introduction
Conventional MRI acquisitions are
optimized with a fixed duration to provide acceptable image quality for most
patients. However, image quality is patient-dependent which may lead to
unnecessary scan time for some patients or insufficient quality for others due
to body habitus or heart rate. Therefore, we proposed an inline automatic
quality control based on closed-loop feedback framework between image
reconstruction and data acquisition to efficiently achieve consistent
diagnostic image quality. Since the signal-to-noise ratio (SNR) is directly
related to confidence of flow measurements1, we applied this framework for quantitative flow measurements
with cardiac 2D phase-contrast flow MRI using a SNR threshold as a stop
criterion for a subject-specific imaging duration.Methods
The closed-loop feedback framework
is designed as follows (Figure 1): during the acquisition, SNR maps are reconstructed
in Gadgetron2 every 20s using 100 pseudo replicas3, then the SNR of the target tissue is extracted
automatically using a trained nnUNet4, and sent from Gadgetron to the sequence controller via the
“FIRE” research framework5 (Siemens Healthineers AG, Germany). The acquisition
automatically stops when a target SNR is achieved.
Ten
healthy volunteers (HV) (BMI=25.4±2.5, age=30±8 years old, male/female=4/6) and
one patient (BMI=29.3, age=73 years old, female) with a valvular implant were
imaged on a 0.55T MRI scanner (MAGNETOM Free.Max, Siemens Healthineers AG,
Germany) using a free-breathing GRE pseudo-golden-angle spiral flow research sequence
(TE/TR=2.0/10.5ms, FA=25°, 1.7mm2 resolution, venc=200cm/s,
FOV=384mm2) modified to listen for a ‘stop’ message. The acquisition
was run with (6 HVs) or without (10 HVs) closed-loop feedback with a maximum
scan time of 4min35s (ascending aorta, AAo) or 6min (main pulmonary artery,
MPA). At the end of the scan, data were retrospectively self-gated to 25
cardiac frames and reconstructed using T-CG-SENSE with a spatial and temporal
constraints (λs=0.1, λt=1).
The
stopping criterion (SNR threshold) was chosen to produce an accurate
measurement of cardiac output, defined as <5% absolute relative error (CO%error)
compared measurements from the full acquisition. This optimal targeted tissue
specific SNR was then applied inline.
For
the automatic segmentation, 2D nnUNet were trained on 128 patients and
evaluated on a test dataset including 10 patients (Dice Similarity Coefficient=0.95±0.02
for AAo and MPA), all of whom were imaged at 1.5/0.55T. Quality control and
image reconstructions were performed on a computer equipped with 4 GPUs (NVIDIA
A100-SXM, 80Gb). Subject imaging was performed with IRB approval.Results
Retrospective analysis demonstrated
that by choosing an SNR threshold of 175 (total across all cardiac frames) for
AAo and 140 for MPA (Figure 2), we ensured sufficient image quality to keep quantitative
flow measurements with an error ≤ 5% relative to the full duration measurement.
By applying these optimal SNR thresholds retrospectively (Figure 3),
acquisition would have automatically stopped at 2min 41s±62s/2min 39s±63s,
saving 41%/57 % of scan time for AAo and MPA.
Deploying the closed-loop feedback
inline, target SNR was reached at 2min27s±53s/2min50s±69s with SNR=181±5/145±3
and the acquisition stopped at 2min39s±67s/3min±80s saving 43%/51% of the scan
duration for AAo and MPA (Figure 4). Compared to the full acquisition, the CO%error
was 2.13±2.04%/6.34+3.73% with a maximum of 5.43%/11.5% for AAo/MPA. For one
subject with BMI=28.8, the full acquisition for AAo was too short to reach the
target SNR (max SNR=164). The SNR-driven acquisition was also successful and
generated diagnostic flow measurements in the one patient, despite a metallic
valvular implant (Figure 4).
Two example subjects are shown in
Figure 5 for illustration. Using a 2 min fixed acquisition time (top row), CO%error
was 5.4% compared to the full acquisition due to insufficient SNR. Whereas, in a
different subject (bottom row), the SNR-driven acquisition stopped the scan at
1min40 but with a CO%error≤1%, avoiding unnecessary
acquisition time.
When deployed inline, SNR map
computation required 12.98±5.49s and automatic segmentation required
1.08±0.09s, meaning that SNR calculation was always feasible within the 20s
assessment interval.Discussion
We demonstrated an inline SNR-driven
automatic quality control for adaptive subject-specific MRI acquisition time
applied to 2D phase contrast MR flow measurements of the great arteries. The
distribution of automated stopping times across the population (standard
deviation of 62/63s for AAo/MPA) revealed the value of subject-specific
acquisition time for consistent image quality instead of fixed duration (Figure
5). It resulted in a 41%/57% acquisition time savings while ensuring a
diagnostic measurement with an average error in quantitative flow parameters
lower than 2.2% and 6.4% for AAo and MPA respectively.Conclusion
Our inline SNR-driven automatic
quality control enables time-efficient subject-specific acquisitions while
ensuring diagnostic image quality and consistent quantitative flow measurement
compared to full time acquisitions. This framework could be adapted for other
applications using different quality metric.Acknowledgements
This work was supported by the National Heart, Lung,
and Blood Institute (NHLBI) Division of Intramural Research (Z01-HL006257,
Z01-HL006213). The authors are investigators on a US Government Cooperative
Research and Development Agreement (CRADA) with Siemens Healthcare to develop
0.55T MRI.References
1. Hansen
MS, Olivieri LJ, O’Brien K, Cross RR, Inati SJ, Kellman P. Method for
calculating confidence intervals for phase contrast flow measurements. J
Cardiovasc Magn Reson. 2014;16(1):46. doi:10.1186/1532-429X-16-46
2. Hansen MS, Sørensen TS. Gadgetron: An
open source framework for medical image reconstruction: Gadgetron. Magn
Reson Med. 2013;69(6):1768-1776. doi:10.1002/mrm.24389
3. Robson PM, Grant AK, Madhuranthakam
AJ, Lattanzi R, Sodickson DK, McKenzie CA. Comprehensive quantification of
signal-to-noise ratio and g -factor for image-based and k
-space-based parallel imaging reconstructions. Magn Reson Med.
2008;60(4):895-907. doi:10.1002/mrm.21728
4. Isensee F, Jaeger PF, Kohl SAA,
Petersen J, Maier-Hein KH. nnU-Net: a self-configuring method for deep
learning-based biomedical image segmentation. Nat Methods. December
2020. doi:10.1038/s41592-020-01008-z
5. Chow K, Kellman P, Xue H. Prototyping
Image Reconstruction and Analysis with FIRE. In: SCMR. Virtual Scientific
Sessions. ; 2021:838972.