Edited by: Steffen Erhard Petersen, Queen Mary University of London, United Kingdom
Reviewed by: Tim Leiner, University Medical Center Utrecht, Netherlands; Emmanuel Androulakis, Royal Brompton & Harefield NHS Foundation Trust, United Kingdom
This article was submitted to Cardiovascular Imaging, a section of the journal Frontiers in Cardiovascular Medicine
This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.
Contrast- enhanced magnetic resonance angiography (CE-MRA) is a well-established, highly trusted imaging technique for the non-invasive assessment of peripheral artery disease (PAD) (
This stepping table subtraction method is available for clinical use from all major vendors of MR systems but suffers from several intrinsic drawbacks. First, a subtraction for background suppression is prone to misregistration artifacts resulting from patient movement, including involuntary (such as peristaltic) motion (
Recently, a clinical study employing a subtractionless method was performed at 1.5T in patients with suspected PAD (
Due to the large chemical shift between water and lipids at 3T, a better selection of echo times that allows for both short scan times and efficient water- fat separation is possible. In theory, this may yield a further improvement in SNR, as the reconstruction of water-only images uses data from two independent echoes in an approximately optimal fashion.
The depiction of the vessel lumen against the background is key in all CE-MRA variants. However, a manual evaluation of VBC based on user-defined regions-of-interest (ROIs) in angiograms is very cumbersome and, to some extent, user-dependent. In order to simplify the workflow and reduce user dependence, a tool for semi-automated VBC analysis was developed and employed.
The purpose of this work was hence to investigate the feasibility of subtractionless first-pass peripheral MRA at 3T in patients with known or suspected PAD using a three- positions stepping table approach with a single dose of contrast agent and to evaluate the performance in terms of SNR and VBC, as well as subjective image quality, compared to the conventional subtraction method.
The resulting SNR in a subtraction image is given by
where S1 and S2 are the signal and σ is the standard deviation of the noise in the pre- and post-contrast images, respectively. Since the noise in the pre- and post-contrast images is uncorrelated, the standard deviation of the noise simply scales with the square root of two in the subtraction image. However, unlike in averaging, the signal does not double. At best, the signal in the vasculature in the pre-contrast image is negligible due to the long T1 of unenhanced blood and the short TR and high flip angle employed, leading to strong signal saturation. In an exemplary measurement in the abdominal aorta the pre-contrast signal intensity was below 10% of the post-contrast signal intensity and was therefore considered negligible. Consequently, the subtraction decreases the SNR by at least the square root of two, or more if the pre-contrast signal intensity is significant, or if a complex subtraction is employed and the signal phases of the pre- and post-contrast images interact in an unfavorable fashion.
In contrast, the gain in SNR in a water-only image reconstructed with a subtractionless method involving two post-contrast acquisitions at different echo times TE1 and TE2 is given by
where ΔTE = TE2-TE1 denotes the echo spacing and ΔfF denotes the offset of the resonance frequency of fat relative to water, which is ~ −421.5 Hz at 3T, and
Overall, the achievable gain in SNR using a subtractionless mDixon reconstruction instead of a conventional subtraction reconstruction amounts to a factor of 2. This factor is composed of a √2 gain by omitting the subtraction and a √2 gain by the water-fat separation and an appropriate choice of echo times. The theoretical SNR gains obtained for each table position with the employed sequence parameters are summarized in
Echo times per anatomical location and respective SNR gains by the water-fat separation (SNR) and by eliminating the subtraction (Total SNR).
Abdominal position | 1.48/2.84 | 1.38 | 1.95 |
Upper leg position | 1.51/2.83 | 1.39 | 1.97 |
Lower leg position | 1.58/2.88 | 1.40 | 1.98 |
Ten patients with known or suspected PAD and clinically indicated MRA were examined on a clinical 3T scanner (Ingenia, Philips Healthcare, Best, Netherlands). Exclusion criteria included common contraindications for magnetic resonance angiography as well as refusal or inability of the patients to get through the examination. During and after injection of 10 ml Gadovist (Bayer Healthcare, Berlin, Germany) at 0.5 ml/s, contrast-enhanced images were acquired successively at three table positions, each with a field of view (FOV) of 430 × 400–450 × 180–200 mm3, using a 3D T1-weighted spoiled dual-gradient-echo sequence with a TE1/TE2/TR of 1.5–1.6/2.8–2.9/4.4–4.7 ms. The measured spatial resolution increased from 1.3 × 1.3 × 17 mm3 at the first, abdominal position to 1.0 × 1.0 × 1.5 mm3 at the third, lower leg position. Scan times ranged from 17 os for the first position to 23 s for the third position, with an up to 8-fold acceleration by SENSE and a partial Fourier factor of 0.7. RF shimming was performed individually at each position. A direct comparison between the subtraction and subtractionless methods was enabled by additionally collecting corresponding non-contrast-enhanced images before injection using the same sequence.
Water images were reconstructed from the contrast-enhanced images using mDixon with a multi-peak spectral model of fat (
All image acquisition was part of routine clinical practice. Conduct and reporting of this study were carried out in accordance with the Helsinki Declaration as revised in 2013. The studies involving human participants were reviewed and approved by the ethics committee of the Charité—Universitätsmedizin Berlin. Both the institutional data protection office and the local ethics committee waived the need for informed consent and provided permission to analyse the anonymized images obtained with the imaging protocol as described below.
We prespecified 23 clinically relevant vessel segments that were assessed with regard to image quality by two cardiologists (3 and 6 years of experience, respectively), each blinded to the rating of the other. Segments were rated based on a four-point scale: 0 = not evaluable, no arteries visible (non-diagnostic); 1 = poor to moderate quality, not all arterial segments evaluable due to noise, heterogeneous vascular enhancement or poor fat suppression (partly non-diagnostic); 2 = acceptable quality but some noise or heterogeneous signal, all arterial segments evaluable for diagnostic purposes; 3 = good quality, all arterial segments evaluable for diagnostic purposes without artifacts.
The volumetric data was reformatted into the axial view, and cylindrical regions of interest (ROIs) centered on the main vessels were manually defined in the mDixon images and labeled according to the respective vessel segment. For each of the 23 vessel segments, 3 different ROIs were evaluated resulting in a total of 69 ROIs per patient and a total of 690 ROIs. The ROIs were copied across to the corresponding subtraction images without modification.
In order to simplify the workflow and reduce user dependence, an automated algorithm to derive the VBC from each ROI was employed, where the VBC was defined as:
The algorithm segmented each ROI in the mDixon images into vessel and background by first searching for the 10 most intense local maxima, where the maxima were sorted by distance to the ROI isocenter and the innermost maximum was identified as a seed point for the vessel. The vessel lumen was then determined and grouped using a flood fill algorithm. From the remaining voxels, vessel branches, and other hyperintense areas were removed in order to approximate the background. The resulting segmentation was identically applied to the mDixon and the subtraction images.
As described in previous studies (
Values were tested for normal distribution using Shapiro–Wilk test. Normally distributed values such as VBC, SNR, and image quality scores are reported as means ± standard deviation and Student's
All 10 patients (4 female, 6 male) were successfully scanned. Median patient age was 69 years (interquartile range 68–74 years), mean body mass index was 29.0 ± 3.6 kg/m2, resulting in a mean contrast agent dose of 0.123 ± 0.013 mmol/kg per patient. The overall mean image quality score ± standard deviation was 2.88 ± 0.32 for the mDixon method and 2.57 ± 0.48 for the subtraction method (
Subjective image quality for the mDixon and subtraction methods averaged for both readers.
Total | mDixon | 2.88 ± 0.32 | |
Subtraction | 2.57 ± 0.48 | <0.001 | |
Abdominal position | mDixon | 2.99 ± 0.08 | |
Subtraction | 2.75 ± 0.28 | <0.001 | |
Upper leg position | mDixon | 2.97 ± 0.15 | |
Subtraction | 2.68 ± 0.37 | <0.001 | |
Lower leg position | mDixon | 2.60 ± 0.50 | |
Subtraction | 2.13 ± 0.60 | <0.001 |
For the subtraction method, one reader rated 7 of 230 vessel segments (3.04%) to be partly (five segments) or completely (two segments) non-diagnostic, the other reader rated 10 vessel segments to be partly non-diagnostic (4.35%). Noticeably, of the 14 segments that were rated with a score of 1 or 0 by at least one reader, only 3 segments were deemed partly or completely non-diagnostic by both readers.
For the mDixon method, one reader considered all vessel segments evaluable for diagnostic purpose, whereas the other reader rated three vessel segments to be non-diagnostic (1.3%). Non-diagnostic vessel segments were predominantly found in the lower leg table position, mostly due to misregistration artifacts (see
Coronal maximum intensity projections obtained by the subtraction method
Coronal maximum intensity projections obtained by the subtraction method
Overall, interobserver agreement with regard to subjective image quality showed moderate agreement with a κ value of 0.534 and an acceptable reliability with a Cronbach's alpha of 0.704 (
Subjective image quality per anatomical location and observer.
mDixon |
2.98 |
2.99 |
2.96 |
2.98 |
2.53 |
2.67 |
|||
– |
0.66 | 0.72 | |||||||
Subtraction (mean ± SD) | 2.88 |
2.63 |
2.81 |
2.56 |
2.20 |
2.05 |
|||
0.09 | 0.30 | 0.55 | |||||||
0.002 | <0.001 | 0.014 | <0.001 | 0.006 | <0.001 |
Subjective image quality scores as given by both raters for mDixon images (3b) and subtraction images (3c) of the abdominal position.
ROI placement was feasible in all prespecified vessel segments. Segmentation into vessel and background was successfully carried out in all ROIs. An example of this segmentation is shown in
The values for VBC were heterogeneous between patients and table positions, with the best results in the upper leg position, followed by the abdominal position (
Mean vessel to background contrast total and per anatomical location.
mDixon (mean ± SD) | 23.16 ± 8.4 | 23.69 ± 6.8 | 31.33 ± 6.3 | 15.60 ± 4.2 |
Subtraction (mean ± SD) | 19.00 ± 8.1 | 18.98 ± 8.5 | 22.60 ± 4.3 | 15.56 ± 4.4 |
Factor | 1.22 | 1.25 | 1.39 | 1.00 |
<0.001 | <0.001 | <0.001 | 0.91 |
Mean gain in signal-to-noise ratio was 1.82 or 82% in the mDixon images as compared to the subtracted images. It was highest in the upper leg position (2.07) and gradually less pronounced in the lower leg position (1.95) and the abdominal position (1. 57). Detailed results are summarized in
Noise and mean SNR gain by the mDixon method as compared with the conventional subtraction method.
mDixon (mean ± SD) | 102.81 ± 30.33 | 81.56 ± 17.3 | 129.69 ± 22.81 | 130.95 ± 17.03 |
Subtraction (mean ± SD) | 187.11 ± 76.21 | 127.78 ± 27.41 | 253.72 ± 43.96 | 271.26 ± 39.98 |
Factor | 1.82 | 1.57 | 2.07 | 1.95 |
<0.001 | <0.001 | <0.001 | 0.091 |
Contrast- enhanced magnetic resonance angiography with a single dose of contrast agent using the two-point modified Dixon method is feasible at 3T and reliably provides good image quality in patients with known or suspected peripheral artery disease. We did not encounter disturbing field inhomogeneity or water-fat swapping artifacts despite the relative closeness of the TEs to in- and out-of-phase echo times at 3T (
Subjective image quality was high in our study, regardless of method, table positions, or reader. However, we found that uniformly, independent of reader or anatomical position, image quality ratings were higher with the mDixon method. A prominent advantage of the mDixon method in this regard is its elimination of misregistration artifacts. Depending on the reader, seven or ten vessel segments were considered non-diagnostic in the subtraction method in the lower leg position, but of sufficient image quality to allow stenosis assessment in the mDixon method (
Detail of angulated coronal maximum intensity projections obtained by the subtraction method
In accordance with the visually assessed image quality improvement with the mDixon method, our study showed an increase in VBC of 22% and in SNR of 82% compared to the conventional subtraction method. The measured SNR gain is in good agreement with the predicted values at the selected echo times. Semi-automated VBC evaluation was successfully carried out in all patients. We substantially reduced the expenditure and complexity of defining the true background signal surrounding a vessel by automatically eliminating adjacent vessels and other local signal intensive structures. The fact that we could not show a significant improvement for vessel-to-background contrast in the lower leg position is somewhat contradictory, given the respective improvements in subjective image quality and signal-to-noise ratio. This may in part be attributed to an increased contrast accumulation in the background tissue at the late acquisition stage in the lower leg, which generally renders the calculation of a true background signal difficult. While some, including the authors of this study, consider an elevated background signal as an advantage, because the visibility of anatomical landmarks is preserved, it may hamper the acceptance of the subtractionless approach by others. Advanced post-processing algorithms were presented (
In our work, the overall gain in SNR was 1.82, thus reasonably close to the theoretically possible values as described in the methods section, which supports the hypothesis that at 3T, the noise propagation in the water-fat separation in the present application is even more favorable than at 1.5T. For the desired resolution, the available gradient performance, and the predominant importance of speed, the resulting echo spacing of ~1.2–1.4 ms is much closer to the, from an SNR perspective, optimal 1/(2* ΔfF) at 3T than at 1.5T.
Our study is focussed on a direct comparison between subtraction and mDixon angiography, where acquisition times and spatial resolution were very similar to those obtained in prior studies at 1.5T (
The feasibility of single- dose contrast- enhanced magnetic resonance imaging for peripheral arteries has been shown before, as well as the feasibility of using the mDixon method at 1.5 T to achieve substantial improvements in image quality, VBC and SNR (
Semi-automated processes to define the extents of vessels and to derive VBC have been described before for magnetic resonance angiography (
Our study has several limitations. First of all, we did not compare the images to the gold standard digital subtraction angiography and no invasive measurement of stenosis severity was employed. Our reference method, CE-MRA, has however been proven to be a reliable imaging method with a high diagnostic accuracy for peripheral artery disease detection (
In conclusion, this study demonstrated the feasibility of subtractionless first-pass peripheral MRA at 3T in patients with known or suspected PAD using a three- positions stepping table approach with a single dose of contrast agent and conducted a comprehensive comparison with conventional subtraction angiography. Our results indicate that the predicted increase in SNR for the given echo time selection at 3T, as well as its robustness against motion artifacts translates well into improved image quality in clinical practice.
The original contributions presented in the study are included in the article/supplementary material, further inquiries can be directed to the corresponding author/s.
The studies involving human participants were reviewed and approved by ethics committee of the Charité—Universitätsmedizin Berlin. Written informed consent for participation was not required for this study in accordance with the national legislation and the institutional requirements.
HE, BS, MK, and SK conceived and planned the image acquisition. KW, CS, and MN performed the measurements. SK, PS, and BP were involved in planning and supervised the work. KW and SK processed the experimental data, performed the analysis, drafted the manuscript, and designed the figures. HE and MK implemented and optimized the modified Dixon MRA technique. CS implemented the algorithms for semiautomated calculations. All authors discussed the results and commented on the manuscript.
HE was employed by the company Philips Research, Hamburg, Germany. CS and BS were employed by the company Philips Healthcare, Hamburg, Germany. MK was employed by the company Philips Healthcare, Best, Netherlands. The remaining authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
We thank the technicians Corinna Else, Gudrun Grosser, and Janina Dentzer for their work on this study.
Contrast-enhanced magnetic resonance angiography
Peripheral artery disease
modified Dixon
Magnetic resonance: MRA, Magnetic resonance angiography
Signal-to-noise ratio
Vessel-to-background contrast
Tesla
Echo time
Repetition time
Region of interest
Offset of resonance frequency of fat relative to water
Field of view
Radiofrequency
Sensitivity encoding
Maximum intensity projection.