High frequency orientation estimates for fast real-time motion correction using vector observations of gravity and the static magnetic field (B0).
Adam M.J. van Niekerk1, Paul Wighton2,3, Ali Alhamud1, Matthew D. Tisdall2,3, Andre J.W. van der Kouwe2,3, and Ernesta M. Meintjes1

1Human Biology, MRC/UCT Medical Imaging Research Unit, University of Cape Town, Cape Town, South Africa, 2Athinoula A. Martinos Center, Massachusetts General Hospital, Boston, MA, United States, 3Radiology, Harvard Medical School, Boston, MA, United States

Synopsis

In this study we propose a novel approach to motion correction in MRI that separates the challenges of tracking orientation and translation. We developed an external hardware device capable of high frequency orientation estimates independent of the pulse sequence. The device takes vector observations of gravity and the MRI scanner’s static magnetic field (B0) and is therefore free from many constraints of some existing external motion tracking techniques. Most notably, no scanner specific calibration is required and the device can be miniaturised. Translation estimates are achieved through the use of 3 high-speed orthogonal navigators. Line by line rigid body motion correction is implemented in a spoiled gradient echo pulse sequence.

Taget Audience

This study is relevant to clinicians and researchers who scan subjects prone to move during image acquisition or use motion sensitive contrasts. Researchers using markers or optical motion tracking could benefit from the novel approach used for orientation estimates.

Purpose

We aim to make in-bore motion tracking more accessible through the use of miniature, low-cost sensors found in a large variety of consumer electronics. Most current external motion tracking techniques rely on a set of translation measurements to estimate orientation1,2, or a retrograte marker which spatially encodes motion parameters on the marker itself3. This fundamentally limits the minimum size of the apparatus the patient must wear as the reliability of orientation estimates falls off as the marker/s are brought closer together1,2 or reduced in size3. We identified the need for orientation estimates independent of the imaging pulse sequence that can be estimated at a temporal resolution unattainable by navigator-based methods. Many external motion correction techniques are also limited by complex calibration procedures or an increased level of subject interaction4, with line-of-site1,3 or limited measurement range restricting marker placement2,3. We address some of these issues using an approach with zero scanner-specific calibration.

Method

An MRI compatible sensor array was developed comprising a 3-axis magnetometer, 3-axis accelerometer and 3-axis angular rate gyro (Figure 1). The sensors allow for high temporal resolution (up to 1 kHz data rate) measurements of the scanner’s static magnetic field vector and the earth’s gravitational field vector. Orientation estimates are obtained by comparing vector observations from the accelerometer and magnetometer in the subject frame with measurements of the a priori reference vectors (gravity [-y] and B0 [+z]) in the scanner frame. To reduce the effects of linear accelerations on the gravity vector estimate, a modified version of the non-linear complementary filter proposed by Mahony5 et al. (originally designed for unmanned aircraft guidance/control) was introduced, which combines angular rate measurements with the vector-based orientation. All sensor signals were digitised onboard the device where a 32-bit micro-controller implemented the complementary filter with single precision floating point arithmetic. A two step, one-time, calibration of the device was performed to correct misalignment of the 3-axis sensors caused by soldering imperfections and variations in the gains of the sensor axes. Data were transmitted to a laptop over a serial optical link (300 Hz). The laptop acts as a TCIP server on the local measurement network for the scanner to query for the latest orientation estimate. By combining the orientation estimates with translation estimates obtained using three orthogonal high speed linear navigators6, full rigid body motion correction is achieved. Translation estimates were obtained in the corrected imaging frame by computing the frequency shift of the raw MR signal relative to a reference acquisition. A multi-channel fitting algorithm improved the stability of the translation estimates (Figure 2). Translation and orientation updates were applied for each line of k-space in a standard 3D spoiled gradient echo pulse sequence (TE 7 ms, TR 12 ms, 256x256x64) with a non-selective RF pulse (Figure 3). Four brain acquisitions were performed in three subjects on a 3 T Siemens Skyra: (1) no motion, correction active, (2) occasional pose changes, correction active, (3) occasional pose changes, no correction, (4) continuous motion, correction active. Motion correction was applied in real-time and corrected images were generated immediately on the scanner. All scans were conducted in accordance with protocols approved by the Faculty of Health Sciences Human Research Ethics Committee of the University of Cape Town.

Results

The magnetometer measurements of the fields within the MRI scanner during image acquisition allow for an exceptionally high signal to noise vector reference. This and that gravity and the static magnetic field are almost perfectly orthogonal result in more precise orientation estimates than what are conventionally possible with similar sensor arrangements. Figures 4 and 5 show the motion parameters measured during each of the corrected scans and representative images for all 4 acquisitions in one subject.

Discussion

High temporal resolution orientation estimates enable robust motion correction free from scanner specific calibration despite noise in the translation estimates. Tiny changes in orientation could result in substantial errors on the periphery of the image due to a ‘lever arm’ effect. These artefacts are difficult to correct restrospectively highlighting the importance of reliable orientation tracking. Changes in posture of 0.1 degrees over one TR (12ms) were often observed showing the importance of high temporal resolution correction. State parameters such as angular rate and acceleration could be combined with navigator measurements in future implementations to improve translation estimate accuracy in high frequency applications.

Acknowledgements

The South African Research Chairs Initiative of the Department of Science and Technology and National Research Foundation of South Africa, Medical Research Council of South Africa, NIH grants R21AA017410, R01HD071664.

References

1. Melvyn B Ooi, Murat Aksoy, Julian Maclaren, Ronald D Watkins, and Roland Bammer. Prospective motion correction using inductively coupled wireless RF coils. Magnetic resonance in medicine : official journal of the Society of Magnetic Resonance in Medicine / Society of Magnetic Resonance in Medicine, June 2013. ISSN 1522-2594. doi: 10.1002/mrm.24845. URL http://www.ncbi.nlm.nih.gov/pubmed/23813444.

2. M Zaitsev, C Dold, G Sakas, J Hennig, and O Speck. Magnetic resonance imaging of freely moving objects: prospective real-time motion correction using an external optical motion tracking system. NeuroImage, 31(3):1038–50, July 2006. ISSN 1053-8119. doi: 10.1016/j. neuroimage.2006.01.039. URL http://www.ncbi.nlm.nih.gov/pubmed/16600642.

3. B. Armstrong, T. Verron, L. Heppe, J. Reynolds, and K. Schmidt. RGR-3D: simple, cheap detection of 6-DOF pose for teleoperation, and robot programming and calibration. Proceedings 2002 IEEE International Conference on Robotics and Automation (Cat. No.02CH37292), 3(May), 2002. ISSN 10504729. doi: 10.1109/ROBOT.2002.1013678.

4. Benjamin Zahneisen, Brian Keating, and Thomas Ernst. Propagation of calibration errors in prospective motion correction using external tracking. Magnetic resonance in medicine : official journal of the Society of Magnetic Resonance in Medicine / Society of Magnetic Resonance in Medicine, 72(2):381–8, aug 2014. ISSN 1522-2594. doi: 10.1002/mrm.24943. URL http://www.ncbi.nlm.nih.gov/pubmed/24123287.

5. Robert Mahony, Tarek Hamel, and Jean-Michel Pflimlin. Nonlinear Complementary Filters on the Special Orthogonal Group. IEEE Transactions on Automatic Control, 53(5):1203–1218, June 2008. ISSN 0018-9286. doi: 10.1109/TAC.2008.923738. URL http://ieeexplore.ieee.org/lpdocs/epic03/wrapper.htm?arnumber=4608934.

6. van der Kouwe, André JW, Thomas Benner, and Anders M. Dale. "Real-time rigid body motion correction and shimming using cloverleaf navigators." Magnetic Resonance in Medicine 56.5 (2006): 1019-1032.

Figures

Figure 1: Printed Circuit Board (PCB) with microcontroller (Larger central IC), sensors (right of controller), supporting circuitry (above controller) and fibre optic connector (underneath left). Dimensions (22mm x 45mm x 10mm)

Figure 2: Multichannel fitting shows the resulting fitted plane (blue) for a translation estimate in the phase encode direction. Without outlier detection (red +) the points with higher index values would induce a bias in the translation estimate.

Figure 3: Pulse sequence diagram showing the 3 mutually orthogonal navigators (imaging and gradient frames are perfectly aligned). The small amplitude navigator pulses allow for a relatively low noise modification to the standard gradient echo pulse sequence.

Figure 4: The measured motion parameters for acquisitions with: (1) no motion, correction active; (2) occasional pose changes, correction active; and (4) continuous motion, correction active. The motion during scan (3) with occasional pose changes, no correction was similar to (2).

Figure 5: Resulting images from the 4 aquisitions. Without correction, even small pose changes caused severe blurring of the anatomy in (3) even though motion was similar to that in (2). Although there is a large amount of ringing in (4), anatomy is still visible despite considerable continuous motion. No filtering or retrospective correction was applied to these images.



Proc. Intl. Soc. Mag. Reson. Med. 24 (2016)
0338