Magnetic resonance imaging of freely moving objects: prospective real-time motion correction using an external optical motion tracking system
Introduction
Subject motion is a limiting factor in most MR imaging applications. Living subjects have a restricted ability to maintain the position of their body parts for prolonged periods of time. This stringently confines the imaging time and thus image resolution, SNR and/or contrast for such applications as abdomen, extremity or head imaging. For abdominal imaging new techniques have emerged (Ehman and Felmlee, 1989, Wang et al., 1996), which allow for a free-breathing image acquisition by means of monitoring the diaphragm position during the breathing cycle. These navigated techniques assume the breathing motion to be essentially one-dimensional and reasonably periodic. Typically, only the data acquired in a narrow window around the exhalation phase, where practically no motion occurs, are accepted either prospectively or retrospectively for the final image reconstruction. The situation appears to be much easier for head imaging. Here it is generally accepted that healthy motivated volunteers are able to preserve their head position for about 10 min. This time, however, may be shortened substantially for less cooperative subjects. Additionally, some of the imaging techniques used in the head are extremely motion sensitive, either because of the intrinsic pulse sequence properties, as for multi shot diffusion-weighted imaging, because of artefacts of the statistical image analysis, as for stimulus-correlated motion in fMRI, or simply based on the high spatial resolution of the images for 3D techniques. There are two important issues which make head motion different from the effects observed in the abdomen. First of all, head (or scull to be more precise) motion can generally be described by six degrees of freedom (DOF) and may not be treated as one dimensional. This increases the requirements on the level of sophistication for motion detection approaches, because simple 1D navigators developed for the abdomen are not suitable for head motion characterization. Secondly, unlike breathing or cardiac motion, movements of the head are typically not periodic, which invalidates the use of gating techniques suitable for imaging of periodically moving organs.
Various approaches have been proposed to circumvent head motion in MRI. Traditionally in fMRI motion correction is performed in post-processing by applying affine transformations to the successive volumes in a time series (Friston et al., 1995). Both the detection and correction of motion is performed in image space. More recently a prospective approach has been proposed, where the motion data extracted from the most recent volume are used to correct for motion in the subsequent scan (Thesen et al., 2000). A common problem of all the image-based motion detection methods is the temporal resolution. If substantial motion occurred during the acquisition of a single multi slice package, the volume thus rendered will be intrinsically distorted and no affine transformation can be found to characterize this motion correctly. A slice-to-volume mapping approach (Kim et al., 1999) offers some limited remedy. However, it fails to perform accurate volume restoration in presence of out-of-slice motion.
A number of navigator-based approaches have been developed to characterize the motion of a sample with up to 6 degrees of freedom (Fu et al., 1995, Welch et al., 2002, Costa et al., 2005, Petrie et al., 2005). These methods reach an accuracy of up to 0.1 mm in translation determination and 0.2° in rotation. However, several RF excitations are required for these navigators with a scan time of about 30 ms. The prolonged scan time and the additional RF pulses applied make these navigator methods not entirely compatible with traditional imaging techniques such as FLASH (Haase et al., 1986), RARE (Hennig et al., 1986) or True-FISP (Oppelt et al., 1986).
Alternative k-space encoding techniques, such as motion compensated projection-reconstruction (Glover and Pauly, 1992), PROPELLER (Pipe, 1999) or self navigated spiral imaging (Khadem and Glover, 1994) have emerged recently, which allow to minimize motion induced image artefacts. Reduced motion sensitivity is commonly achieved by repeatedly oversampling the central region of k-space. This additional information is used for motion compensation. However, oversampling of certain regions of k-space typically implies a trade off in the scanning efficiency, even in the absence of motion. Additionally, the use of non-Cartesian k-space trajectories, especially in combination with parallel imaging, often dramatically increases the image reconstruction overhead.
Recently, the possibility of operating an external optical motion tracking system inside the scanner room has been demonstrated (Eviatar et al., 1999, Zhang et al., 1999, Tremblay and Graham, 2002, Dold et al., 2004a, Dold et al., 2004b, Zaitsev et al., 2004a, Zaitsev et al., 2004b) and first results have been presented using the tracking data for retrospective motion correction in fMRI (Tremblay et al., 2005).
In this paper, a general framework of using an external motion tracking system for prospective motion compensation in MRI is presented. The optical motion tracking system used is able to determine the sample position and rotation in 6 degrees of freedom in real time with high spatial accuracy. The tracking data are received directly by the hardware control unit of the MR imager, which updates the logical gradient orientations, RF frequencies and phases accordingly on every sequence repetition cycle. The usefulness of the approach is demonstrated both in phantoms and volunteers for several sequences such as 2D and 3D gradient echo, 2D spin echo with line-by-line position update and echo planar imaging with slice-by-slice correction.
Section snippets
Theory
There are numerous coordinate systems and transformations involved in performing prospective motion correction with an external tracking system. The coordinate systems considered in the following text are: tracked object (will be denoted with index “o”), optical tracking system (index “t”), MR scanner (index “s”) and imaging volume (index “i”). Tracking systems typically report positions and orientations of the tracked objects in the coordinate frame of the tracking system. Each target itself
Experimental setup and equipment
The experiments were performed on a Siemens Magnetom Trio 3T whole-body system (Siemens Medical Solutions, Erlangen, Germany). The optical motion tracking system (ARTrack1, Advanced Realtime Tracking GmbH, Weilheim, Germany) was based on a stereoscopic reconstruction of rigid bodies from grey scale images. The two progressive scan infrared video cameras of the tracking system were positioned in the magnet room as sketched in Fig. 1a, slightly elevated above the axis of the magnet in order to be
Results
In Fig. 4 the progression and the results of the proposed iterative calibration procedure are displayed. In this example the initial estimate of the coordinate transformation contained an error of several millimeters. With the iterative procedure the cross-calibration errors were reduced below 0.1 mm and 0.1 degree after 9 iterations. The whole cross-calibration procedure required about 30 min.
In Fig. 5 the results of a static correction test are presented. The gel-filled resolution phantom was
Discussion
This manuscript reports on the first successful implementation of a continuous real time 6DOF motion correction in MR imaging. A variety of imaging methods is covered including multislice EPI with slice-by-slice position update and traditional 2D and 3D Fourier imaging with line-by-line position update and sub-millimeter resolution. The results clearly demonstrate the advantages of prospective motion correction and imply benefits for practically every MR examination and pulse sequence.
A
Acknowledgments
Authors thank Advanced Realtime Tracking GmbH for providing the optical tracking system and Dr. Mark Schneberger for technical assistance. Dr. Valerij Kiselev is thanked for fruitful discussions during the manuscript preparation.
This research work was in part funded by the EU “Information, Society, Technologies” program (grant IST-2000-28168, project code name MRI-MARCB).
References (26)
- et al.
FLASH imaging. Rapid NMR imaging using low flip-angle pulses
J. Magn. Reson.
(1986) - et al.
NMR Fourier Zeugmatography
J. Magn. Reson.
(1975) - et al.
Using the axis of rotation of polar navigator echoes to rapidly measure 3D rigid-body motion
Magn. Reson. Med.
(2005) - et al.
The compensation of head motion artifacts using an infrared tracking system and a new algorithm for fMRI
Med. Meets Virtual Real.
(2004) - et al.
Updating of MRI gradients using a infrared tracking system to compensate motion artifacts
Int. Soc. Magn. Reson. Med.
(2004) - et al.
Adaptive technique for high-definition MR imaging of moving structures
Radiology
(1989) - et al.
Real time head motion correction for functional MRI
Int. Soc. Magn. Reson. Med.
(1999) - et al.
Spatial registration and normalization of images
Hum. Brain Mapp.
(1995) - et al.
Orbital navigator echoes for motion measurements in magnetic resonance imaging
Magn. Reson. Med.
(1995) - et al.
Projection reconstruction techniques for reduction of motion effects in MRI
Magn. Reson. Med.
(1992)
RARE imaging: a fast imaging method for clinical MR
Magn. Reson. Med.
High-order MR shimming: a simulation study of the effectiveness of competing methods, using an established susceptibility model of the human head
Appl. Magn. Reson.
Self correction for motion in spiral imaging
Int. Soc. Magn. Reson. Med.
Cited by (317)
Unsupervised dual-domain disentangled network for removal of rigid motion artifacts in MRI
2023, Computers in Biology and MedicineMotion robust magnetic resonance imaging via efficient Fourier aggregation
2023, Medical Image AnalysisServo navigators: Linear regression and feedback control for rigid-body motion correction
2024, Magnetic Resonance in MedicinePredicting dynamic, motion-related changes in B<inf>0</inf> field in the brain at a 7T MRI using a subject-specific fine-trained U-net
2024, Magnetic Resonance in MedicineDeep Learning for Retrospective Motion Correction in MRI: A Comprehensive Review
2024, IEEE Transactions on Medical Imaging