Clinical Evaluation of Scout Accelerated Motion Estimation and Reduction Technique for 3D MR Imaging in the Inpatient and Emergency Department Settings

DISCUSSION

In this work, we have shown that motion correction using SAMER significantly reduced motion artifacts in volumetric T1-weighted MPRAGE images obtained in a clinical setting within clinically feasible reconstruction times. In most cases with moderate and severe motion artifacts, SAMER reconstruction was able to mitigate motion artifacts sufficiently to improve the quality of the examination from nondiagnostic to diagnostic. In cases with absent, minimal, or mild motion, application of SAMER provided comparable image quality without significantly impacting the overall motion grades. CNR and SNR were overall significantly higher in SAMER motion-corrected studies than baseline studies. Our findings suggest that motion-prone patients such as those scanned in the emergency and/or inpatient settings may greatly benefit from SAMER to reduce motion artifacts and prevent time-consuming repeat acquisitions and/or callback examinations, without sacrificing overall image quality.

Techniques for motion mitigation include alternative k-space sampling trajectories that are more robust to motion such as radial or spiral acquisitions, ultra-fast imaging techniques (eg, controlled aliasing in parallel imaging [CAIPI], wave-CAIPI, compressed sensing),^7,11-21 and retrospective or prospective motion-correction techniques.^3,22 Accelerated MR imaging techniques can reduce motion artifacts by reducing acquisition times. Very rapid brain MR imaging protocols have been developed with 1–2 minutes of the total scan time.^4,23 The highly accelerated MR imaging techniques rely on high-end multichannel receiver arrays that may not always be clinically available or feasible to implement, eg, in large patients. Accelerated imaging techniques and SAMER motion correction are not mutually exclusive, and a combination of these methods may provide additional benefits. Future work is needed to evaluate the potential synergistic benefits of applying SAMER motion correction to accelerated MR imaging techniques.

Prospective motion-correction methods include MR imaging–based motion navigator and optical-based motion tracking systems.^18,24 MR imaging–based navigator systems are motion-robust because they provide extra motion information from the oversampled central k-space.¹⁸ These techniques provide real-time positional information, allowing mitigation of motion through real-time updates of the imaging FOV. These techniques, however, require MR imaging systems that are capable of dynamic updates during image acquisition, installation of additional hardware, and extending the acquisition time to collect additional data used for motion estimation. Furthermore, prospective techniques can have measurement and estimation errors that can degrade image quality. In these quality-degraded cases in which the original uncorrected images are not available, re-acquisition may be required. These factors limit the implementation of prospective motion-correction techniques into existing radiology workflow without dramatic changes in operation, cost, software, and hardware.^25,26

In contrast, retrospective motion-correction techniques use motion information that is encoded from multichannel receiver arrays and is extracted for postacquisition correction through nonlinear inversion of a physics model.^8,27 Retrospective motion-correction techniques require an iterative approach and are often limited by high computational requirements and reconstruction times.¹⁸ The SAMER framework overcomes these limitations by exploiting a single 3- to 5-second scout image used to jump-start and stabilize the motion-trajectory estimations and an optimized linear + checkered sequence ordering in addition to the scout image to facilitate the separation of the image and motion parameter unknowns.⁸ The low-spatial-resolution and highly accelerated (R = 6) initial scout sequence used for motion correction has an echo-train length of ∼1 second, which essentially eliminates patient motion on the scout image and makes it a good motion-free baseline. SAMER further reduces the computational footprint by restricting readout voxels, using coil compression, and having only a single iteration step, reaching a median reconstruction time of 107 seconds in this study compared with several minutes with other retrospective motion correction techniques.^8,28 A further added advantage of all the retrospective motion-correction techniques is that they can be applied to existing clinical protocols, equipment, and workflow through modifications to the sequence and reconstruction software, without incurring additional burden to the patients or operators by obviating the need for external markers or cameras.^29-31

SAMER motion correction disproportionately benefited studies that had moderate and severe motion artifacts, with most nondiagnostic motion cases exhibiting an improvement in the motion grade. In fact, 69% of nondiagnostic cases were considered diagnostic with regard to motion artifacts following SAMER motion correction. In contrast, most examinations with mild-to-no motion, in which motion correction would offer little benefit in a clinical setting, were not significantly impacted by SAMER motion correction. These findings suggest that SAMER motion correction may provide the most benefit to examinations with moderate-to-severe motion, in which the image quality would otherwise be considered nondiagnostic and the underlying pathology might be obscured. In cases of extreme motion, it can be very difficult to perform retrospective correction accurately, seen in 2 cases in this study. The main reason is that extreme head rotation during image-acquisition causes large gaps in k-space that parallel imaging is unable to fill,³² ie, the gaps in k-space data were too large to allow retrospective motion correction with the existing data. This feature is a limitation of SAMER and other retrospective methods, and in these instances, either a full or partial repeat image acquisition was required.

Application of motion-correction techniques could potentially introduce unwanted reconstruction artifacts through imperfections in the motion estimation. SAMER uses a data-driven approach for motion correction, in which we optimize over a SENSE + motion model to estimate the patient’s motion trajectory.⁸ Small instabilities in this nonconvex optimization can lead to small inaccuracies in the motion parameters. This outcome was seen in 8 cases with uncorrected mild motion (grade 2), in which SAMER introduced a small number of unintended artifacts and worsened the motion grade by 1 point (Online Supplemental Data). However, even if SAMER causes slightly increased artifacts in a small number of cases, this result should not negatively impact the diagnostic quality of the examination because the original (non-motion-corrected) images are still available to the radiologist for reference and interpretation, safeguarding against potential worsening of motion due to the alternative k-space sampling strategy adopted to mitigate motion artifacts. As the technique is further refined and the negative impact on artifacts is diminished, only SAMER motion-corrected images may be needed for interpretation to minimize image bloat for the radiologist.

The SAMER framework also offers the ability to track the motion trajectory through use of motion-guidance lines at fixed k-space locations.⁸ In addition, SAMER allows motion states to be determined during acquisition from each single shot. This feature allows motion data to be made immediately available at the end of the acquisition rather than having to wait for data reconstruction. Although it was not explored here, the motion trajectories provided by SAMER and other approaches can be used to design and train classification systems for automated prediction of motion severity as well as the level of artifact reduction that SAMER and other motion correction techniques might provide. These types of systems could enhance the technologist and radiologist workflow by automatically identifying motion-degradation and correction viability, possibly even before scan completion. This enhancement may potentially allow the technologist to terminate an incomplete-but-nondiagnostic acquisition and troubleshoot to obtain a better repeat acquisition. While SAMER motion correction was only applied to the 3D T1-weighted MPRAGE sequence in this study, it can, in principle, also be extended to other sequences including 2D spin-echo and gradient-echo sequences.⁸ Motion correction in 2D sequences is limited by the greater section thickness and missing information from section gaps compared with 3D sequences. Deep learning models can improve image interpolation errors for 2D sequence motion-correction techniques^10,32 and could play a role in the extension of SAMER to 2D motion correction in the future.

Limitations of the current study include its retrospective nature and inclusion of only one 3D volumetric MR sequence. This was intended as a feasibility study to demonstrate the efficacy of SAMER on a single pulse sequence (3D T1-weighted MPRAGE) in a clinical setting. Testing of SAMER on additional MR images, 2D and volumetric sequences, is needed. In addition, the included MR imaging examinations were heterogeneous in that some examinations were performed without contrast and others were performed with contrast. While the results demonstrated that SAMER was able to improve motion artifacts in both noncontrast and contrast-enhanced MPRAGE images, the small sample size limited further evaluation of whether SAMER could mitigate motion to the same degree if contrast was present or absent. The focus of this study was to evaluate the ability of SAMER to mitigate motion and not on the conspicuity of findings. Nonetheless, future studies with larger numbers of patients are needed to explore how the presence of contrast may potentially affect the degree of motion correction achieved by SAMER.