Reviews and CommentaryFree Access

Performing Diffusion Tensor and Functional MRI in Patients with Metallic Braces

Published Online:https://doi.org/10.1148/radiol.2019192297

See also the article by Miao and Wu et al in this issue.

Dr Olaf Dietrich is a professor of experimental radiology and MR physicist in the Department of Radiology at the University Hospital of the Ludwig Maximilian University (LMU) of Munich, Germany. He has long-term research experience in the field of diffusion-weighted MRI, focusing on pulse-sequence development, assessment of image quality, and data postprocessing.

Echo-planar imaging (EPI) is an important MRI technique; its initial development dates back to the 1970s (1). Its outstanding feature is an extremely short acquisition time of typically about 50 msec per image. Because of its enormous speed, many particularly image-intensive applications such as functional brain MRI, dynamic susceptibility contrast perfusion MRI, or diffusion tensor imaging (DTI) rely on this well-established method. For neuroimaging, EPI may well be the pulse sequence that produces the largest total number of images from MRI every day.

At the same time, echo-planar images are hardly ever beautiful; they tend to feature low spatial resolution, ghosting artifacts, severe distortion, geometric deformation, and signal loss (2). Better gradient systems, which have become available in the last 10 years, have improved the image quality of EPI and enabled new applications in body regions that were previously considered unsuitable for EPI acquisitions. Nevertheless, the image quality of EPI generally remains inferior to that of most other pulse sequences.

The cause of these limitations is that EPI pulse sequences consist of a single fairly long (eg, 50 msec) data readout period after signal excitation by a radiofrequency pulse. The long readout gathers enough data to reconstruct a complete two-dimensional image. This is in contrast to other pulse sequences, which scan k-space line by line with numerous and much shorter readouts that are separated by radiofrequency pulses and take between about 1 msec and 10 msec. It is during this long EPI readout that various types of errors can accumulate and adversely affect the image quality due to magnetic field inhomogeneities, T2* signal decay, chemical shift effects, and gradient imperfections.

EPI artifacts are especially prominent in anatomic areas where nearby tissue-air boundaries or tissue types with different magnetic susceptibilities influence the static magnetic field. This results in signal loss and geometric distortion. Example regions include the inferior frontal lobes of the brain close to the air-filled spaces of the paranasal and frontal sinuses. Even worse artifacts occur if metallic material is present in or at the skull of the imaged participants such as aneurysm clips (3), deep brain stimulation devices (4), or metallic orthodontic braces. Many EPI-based functional MRI techniques are not applicable to patients with such image artifacts.

In this issue of Radiology, Miao et al (5) propose pulse sequences based on fast three-dimensional (3D) spoiled gradient-echo (GRE) acquisitions as a more robust alternative to EPI for patients with metallic braces. They assessed this approach for DTI and blood oxygen level–dependent (BOLD) functional MRI. The proposed sequences consist of an initial contrast preparation block (for either DTI or BOLD) followed by a 3D GRE readout, which is essentially the pulse sequence that is also used for abdominal volumetric breath-hold MRI or fast angiographic dynamic 3D acquisitions. The authors evaluated the 3D GRE approach in six healthy participants wearing removable metallic orthodontic braces. Image quality and functional metrics (of DTI and functional MRI) were compared between EPI and 3D GRE acquisitions, as well as between examinations with braces and without braces. The authors looked at image data in two brain regions of interest: one with strong susceptibility-induced artifacts in the frontal inferior brain and another one with minimal artifacts sufficiently distant from the dental braces.

In the regions affected by image artifacts due to braces, the authors found significantly better signal-to-noise ratios for both functional MRI and DTI acquisitions when the 3D GRE readout was used. No significant signal-to-noise ratio differences between both sequences were found in regions without relevant artifacts or in acquisitions with the braces removed. Although not statistically significant, 3D GRE appears to deliver systematically slightly lower signal-to-noise ratios than does EPI under optimal conditions (ie, in areas without metal artifacts).

The signal-to-noise ratio improvements in participants with dental braces also translated into improved functional metrics for both BOLD and DTI measurements in the artifact regions. The BOLD effect caused by a breath-hold paradigm and quantified as contrast-to-noise ratio (that is, the relative signal change multiplied by the signal-to-noise ratio) improved significantly from 0.29 ± 0.10 (standard deviation) in EPI to 0.83 ± 0.16 in 3D GRE acquisitions. The apparent diffusion coefficients and fractional anisotropy values normalized from unphysiologically low values of 0.16 ± 0.16 × 10-3 mm2/sec and 0.10 ± 0.12 in EPI acquisitions to 0.75 ± 0.08 × 10-3 mm2/sec and 0.45 ± 0.07, respectively, in 3D GRE acquisitions. These results of 3D GRE were comparable to echo-planar DTI measurements in brain regions without metal-induced artifacts. Moreover, not only data quality improved, but also geometric distortions of the brain were shown to be significantly lower in 3D GRE DTI than in EPI DTI.

These are promising results for future applications of functional MRI and DTI in participants with susceptibility artifacts in the brain. It is also worth noting that the scan duration for the acquisition of a complete 3D volume was similar for conventional multisection EPI and volumetric 3D GRE MRI in this study. For instance, both functional MRI protocols had a temporal resolution of 2 seconds for the acquisition of 40 sections.

From a physical perspective, there is one important principal difference between the two methods that requires further detailed analyses: EPI protocols include contrast preparation (either T2* weighting for BOLD functional MRI or diffusion weighting for DTI) separately for each individual two-dimensional slice. In contrast to EPI, this preparation is performed only once prior to the acquisition of the complete 3D volume in the proposed 3D GRE approach. After the contrast preparation (and before the readout), the resulting transverse magnetization is flipped back into the longitudinal direction. The data readout after this preparation block then takes between 1 second and several seconds. During the readout module, relatively small fractions of the prepared longitudinal magnetization are used for each of several hundred subsequent low-flip-angle GRE acquisitions. But at the same time, longitudinal relaxation is taking place and—over the period of a few T1 time constants—is gradually diminishing the effect of the contrast preparation.

Thus, the 3D GRE readout cannot be prolonged arbitrarily, because after a few seconds the BOLD (or DTI) contrast will have relaxed away. Consequently, the spatial resolution or matrix size of the 3D GRE readout is restricted by these timing constraints and cannot be chosen as high as might be desirable for a given application. This limitation can be mitigated (and indeed it is in the present study) by a centrically reordered readout, in which the k-space center is read first when the prepared contrast is still at its maximum. More peripheral k-space areas may obtain less functional contrast information, but this is typically not obvious in reconstructed image data because only geometrically small structures are affected. Whether this can be considered acceptable in actual study protocols is still to be evaluated.

There were a few other limitations of this first proof-of-concept study that demand more detailed follow-up. One is that the authors compare a conventional T2*-weighted GRE BOLD EPI sequence with a spin echo–prepared T2-weighted BOLD 3D GRE sequence, which is known to exhibit different sensitivity properties (6). It will be interesting to see if a T2*-weighted BOLD 3D GRE sequence, which is currently being developed by the same group, will provide even higher sensitivities to neural activations. A second issue is the relatively simple functional approaches used for this evaluation. The assessment of functional MRI was based on a rather robust breath-holding paradigm, which can be expected to exhibit a high BOLD contrast and high levels of activation even at reduced signal-to-noise ratios. Similarly, the DTI evaluation involved only relatively robust apparent diffusion coefficient and fractional anisotropy calculations.

Ultimately, studies in larger patient groups and with more focused protocols are required to evaluate the feasibility and sensitivity of the proposed pulse sequences, for example, for resting-state functional MRI applications or for (multitensor) tractography analyses with high spatial resolution. However, even if 3D GRE sequences turn out to be generally less sensitive than optimized EPI protocols in such studies, the 3D GRE approach is certainly expected to excel in all situations, in which otherwise unavoidable susceptibility artifacts affect the data quality of EPI acquisitions.

Disclosures of Conflicts of Interest: O.D. disclosed no relevant relationships.

References

  • 1. Mansfield P. Multi-planar image formation using NMR spin echoes. J Phys Chem 1977;10(3):L55–L58. Google Scholar
  • 2. Zhou XJ. Echo planar imaging. In: Bernstein MA, King KF, Zhou XJ, eds. Handbook of MRI pulse sequences. Burlington, Mass: Elsevier Academic Press, 2004; 702–739. Google Scholar
  • 3. Khursheed F, Rohlffs F, Suzuki S, Kim DH, Ellmore TM. Artifact quantification and tractography from 3T MRI after placement of aneurysm clips in subarachnoid hemorrhage patients. BMC Med Imaging 2011;11(1):19. Crossref, MedlineGoogle Scholar
  • 4. Sammartino F, Krishna V, Sankar T, et al. 3-Tesla MRI in patients with fully implanted deep brain stimulation devices: a preliminary study in 10 patients. J Neurosurg 2017;127(4):892–898. Crossref, MedlineGoogle Scholar
  • 5. Miao X, Wu Y, Liu D, et al. Whole-brain functional and diffusion tensor MRI in human participants with metallic orthodontic braces. Radiology 2020;294:149–157. LinkGoogle Scholar
  • 6. Sanchez Panchuelo RM, Schluppeck D, Harmer J, Bowtell R, Francis S. Assessing the spatial precision of SE and GE-BOLD contrast at 7 Tesla. Brain Topogr 2015;28(1):62–65 https://doi.org/10.1007/s10548-014-0420-4. Crossref, MedlineGoogle Scholar

Article History

Received: Oct 11 2019
Revision requested: Oct 14 2019
Revision received: Oct 15 2019
Accepted: Oct 16 2019
Published online: Nov 12 2019
Published in print: Jan 2020