0578

Magnetic Resonance SIGnature MAtching (MRSIGMA) for Real-Time Volumetric Motion Tracking
Li Feng1 and Ricardo Otazo1,2

1Department of Medical Physics, Memorial Sloan Kettering Cancer Center, New York, NY, United States, 2Department of Radiology, Memorial Sloan Kettering Cancer Center, New York, NY, United States

Synopsis

New MR-Linac systems allow for simultaneous imaging and radiation-treatment, and hold great promise for adaptive radiation-treatment of moving organs. However, even with the latest MRI acquisition and reconstruction technologies, tracking volumetric motion in real-time is still challenging. This work describes a novel technique called MR SIGnature MAtching (MRSIGMA), which consists of (1) offline learning, where 3D motion states and corresponding unique rapid signatures are learned; and (2) online matching, where only rapid signature data are acquired to determine corresponding motion states. Initial implementation using golden-angle radial sampling is shown for liver motion tracking with a latency of < 200ms.

INTRODUCTION

MRI offers many advantages for planning, adaptation and evaluation of radiation therapy [1]. The new MR-Linac systems that combine an MRI scanner with a linear accelerator are now available for simultaneous imaging and radiation-treatment [2-3], and holds significant promise for adaptive treatment of moving organs, such as the lung and liver. However, even with the latest MRI acquisition and reconstruction technologies, tracking volumetric motion in real-time with a high spatial and temporal resolution is still challenging. In order to address this challenge, this work proposes a novel real-time 4D MRI technique called MR SIGnature MAtching (MRSIGMA), which first learns a database of 3D motion states and motion signatures (offline learning) and then matches signature data acquired in real-time at a high temporal resolution with the corresponding database entry (online matching). MRSIGMA is demonstrated for real-time motion tracking of the liver using a stack-of-stars golden-angle radial sampling scheme.

METHODS

MRSIGMA general idea: As shown in Figure 1, MRSIGMA consists of two steps: (a) offline learning of 3D motion states and motion signatures; and (b) online matching of high temporal resolution signature-only data with the motion state/signature database entry. Offline learning uses motion navigation and motion-resolved imaging over multiple motion cycles to create a database of high spatial resolution 3D motion states and corresponding unique motion signatures. A key point is that the acquisition of the signature data must be very fast (<200ms). During the online matching step, high temporal resolution signature-only data are acquired, and the 3D motion state whose signature best matches the real-time data is selected.

Implementation of MRSIGMA using golden-angle radial imaging: MSIGMA is implemented using stack-of-stars golden-angle radial sampling [4-5]. For the offline learning steps, motion signatures are generated from the centers of k-space for each radial angle (Figure 2), and high-resolution 3D motion states are reconstructed from continuously acquired data over multiple motion cycles using the XD-GRASP technique [6] (Figure 3). Online matching is performed by acquiring signature data only (kx-kz plane for each rotation angle), online signature extraction (generation of respiratory projection profiles followed by motion detection as shown in Figure 2) and correlation-based signature matching.

Evaluation: Two 3D liver datasets previously acquired using a prototype fat-saturated stack-of-stars golden-angle sequence on a 3T MRI scanner (TimTrio, Siemens Healthineers, Germany) were used to test the MRSIGMA framework in-vivo. Relevant imaging parameters included: acquired matrix size=256x256x35, FOV=320x320x216mm3, TR/TE=3.40/1.68ms. A total of 1000 spokes were acquired, resulting in a total scan time of 178 seconds. Offline learning was performed using the first 900 spokes to reconstruct a 4D respiratory-resolved image database using XD-GRASP [6], resulting in 10 motion states spanning from expiration to inspiration. Online matching was performed with the last 100 spokes to generate 100 3D images from the pre-learned database using the online matching algorithm, leading to a temporal resolution of 0.178 seconds per volume.

RESULTS

Figure 4 compares 4D respiratory-resolved liver images in different motion states (right two columns, estimated during offline learning using XD-GRASP) with corresponding real-time x-z projections (left column, generated from online matching data). The total latency, including the acquisition of online motion signature (~178ms) and the matching process (～30ms), was around 200ms. Figure 5 shows real-time liver motion tracking using MRSIGMA compared to projections in the x-z plane, which were obtained by performing a 2D FFT on the kx-kz plane for each of the last 100 spokes (online signature data). Given the challenges to obtain a real-time 4D image for comparison, high-temporal resolution x-z projections were used to validate the performance of MRSIGMA. The motion displacement of the liver dome, with respect to the top edge of the FOV, was 23.51±9.84 pixels and 24.35±10.89 pixels in the 4D image and x-z projections, respectively, showing great correlation (R2=0.925).

DISCUSSION

MRSIGMA enables volumetric motion tracking in real-time by performing all the time-consuming data acquisition and image reconstruction work during the offline learning step and leaving simple and rapid tasks for the online matching step. The latency during the online matching step can be further reduced by acquiring only one navigation spoke as the online motion signature data. The idea of MRSIGMA can easily be extended to golden-angle Cartesian or 3D radial sampling. MRSIGMA is originally intended for the MR-Linac system, where a number of calibration procedures are usually performed with the patient on the table before treatment, and thus a sufficient amount of time is available for offline learning. However, MRSIGMA is not limited to MR-Linac and can also be used for other applications where real-time motion-tracking is needed.

Acknowledgements

The authors thank Dr. Lihua Chen from the Southwest Hospital in Chongqing, China for sharing the stack-of-stars golden-angle liver datasets.

References

[1] Lagendijk J et al. MR guidance in radiotherapy. Phys Med Biol. 2014; 59:R349.

[2] Raaymakers BW et al. Integrating a 1.5 T MRI scanner with a 6 MV accelerator: proof of concept. Phys. Med. Biol. 2009; 54:N229–37.

[3] Dempsey J et al. A device for realtime 3D image-guided IMRT. Int J Radiat Oncol Biol Phys. 2005; 63(1):S202.

[4] Chandarana H et al. Free-breathing radial 3D fat-suppressed T1-weighted gradient echo sequence: a viable alternative for contrast-enhanced liver imaging in patients unable to suspend respiration. Invest Radiol. 2011 Oct;46(10):648-53.

[5] Winkelmann S et al. An optimal radial profile order based on the Golden Ratio for time-resolved MRI. IEEE Trans Med Imaging 2007;26:68–76.

[6] Feng L et al. XD-GRASP: Golden-angle radial MRI with reconstruction of extra motion-state dimensions using compressed sensing. Magn Reson Med. 2016 Feb;75(2):775-88.

Figures

Figure 1: Description of the MRSIGMA general idea.. (1) Offline learning acquires signature and 3D imaging data over multiple respiratory cycles to create a database of high-resolution motion states. (2) Online matching acquires signature data only at high temporal resolution. The 3D motion state whose signature best matches the newly-acquired signature data is selected. Real-time 3D motion tracking is accomplished by performing all the time-consuming acquisition and reconstruction work during offline learning, and leaving just signature data acquisition and correlation analysis in the online matching step, to minimize latency.

Figure 2: (a) Stack-of-stars golden-angle radial sampling for MRSIGMA. (b) High temporal resolution signature data are generated by taking a projection along the z-dimension for each angle. (c) Examples of respiratory signals detected during the offline learning step (red curve) and the online matching step (green curve).

Figure 3: Offline learning and online matching steps separated by a break interval. During offline learning, the database is generated, where each entry includes a pair of motion signature (red line) and motion state (3D image). During online matching, online signature data (green line) are matched to the corresponding offline signature (red line).

Figure 4: Comparison of x-z 2D projections with corresponding 3D images obtained with MRSIGMA at different motion states. The x-z 2D projections, which serve as the online motion signature data, are treated as references to validate the motion pattern in MRSIGMA. This example shows that MRSIGMA is able to generate high-resolution 3D images in real time, with its motion pattern is well-correlated with the reference 2D projection (yellow dashed lines).

Figure 5: Movies comparing x-z 2D projections with corresponding 3D images obtained with MRSIGMA at different motion states in two subjects. There is a lesion in subject 2 (bottom row) that is clearly delineated in the high-resolution 3D images. While it is hard to see image information in the 2D x-z projections, they only serve as online motion signature data and can be used to validate the motion pattern in the corresponding 3D high-resolution images.

Proc. Intl. Soc. Mag. Reson. Med. 27 (2019)
0578