Fast Isotropic Volumetric Coronary MR Angiography Using Free-Breathing 3D Radial Balanced FFE Acquisition

Transcription

1 Fast Isotropic Volumetric Coronary MR Angiography Using Free-Breathing 3D Radial Balanced FFE Acquisition C. Stehning, 1 * P. Börnert, 2 K. Nehrke, 2 H. Eggers, 2 and O. Dössel 1 Magnetic Resonance in Medicine 52: (2004) A shortcoming of current coronary MRA methods with thin-slab 3D acquisitions is the time-consuming examination necessitated by extensive scout scanning and precise slice planning. To improve ease of use and cover larger parts of the anatomy, it appears desirable to image the entire heart with high spatial resolution instead. For this purpose, an isotropic 3D-radial acquisition was employed in this study. This method allows undersampling of k-space in all three spatial dimensions, and its insensitivity to motion enables extended acquisitions per cardiac cycle. We present initial phantom and in vivo results obtained in volunteers that demonstrate large volume coverage with high isotropic spatial resolution. We were able to visualize all major parts of the coronary arteries retrospectively from the volume data set without compromising the image quality. The scan time ranged from 10 to 14 min during free breathing at a heart rate of 60 bpm, which is comparable to that of a thin-slab protocol comprising multiple scans for each coronary artery. Magn Reson Med 52: , Wiley-Liss, Inc. Key words: 3D radial; volumetric imaging; isotropic resolution; coronary artery imaging; 3D-PR Clinical interest in magnetic resonance imaging (MRI) for cardiac diagnostics has grown over the recent years, as this noninvasive technique provides excellent soft-tissue contrast and enables the assessment of morphology and cardiac vasculature function. However, a shortcoming of current coronary MRA methods is the time-consuming examination necessitated by the need for extensive scout scanning prior to the actual image acquisition. These preparations are necessary for locating individual anatomical details, such as the coronary arteries, and planning particular measurements (1). A desirable improvement would be to acquire an extended volume (e.g., the entire heart) with high, isotropic resolution (2), which would make initial planning rather simple. Slices and arbitrary views could then be reconstructed retrospectively from the volume data set. However, to maintain an acceptable total scan time, a high acquisition speed is mandatory. Recently, fast volumetric MRA based on Cartesian scanning using parallel imaging techniques was introduced for this purpose (3). Alternatively, non-cartesian techniques, such as radial and spiral acquisitions, are promising candidates. In particular, projection reconstruction or radial techniques (2D PR) have been introduced as robust tools for cardiac studies (4,5), since they allow higher in-plane resolution per time unit than Fourier techniques. A particular asset of the radial technique is that the point-spread function (PSF) is robust with respect to undersampling. Thus, aliased signal energy will appear only as slight streaking and increased pseudo-noise, whereas undersampling in a Cartesian acquisition will result in severe ghost images. Projection reconstruction in three dimensions (3D-PR) offers an even greater potential for reducing scan time by undersampling the outer regions of k-space in all three spatial dimensions (6). In the present work, such a 3D radial acquisition was combined with a navigator-gated, balanced FFE sequence to acquire large imaging volumes covering the entire heart. Different views and slices (containing, e.g., the left anterior descending artery (LAD) and the right coronary artery (RCA)) were reformatted retrospectively after the scans were completed. MATERIALS AND METHODS 3D Radial Sampling of k-space A nearly isotropic coverage of k-space with radial sampling in 3D is achieved by distributing the end points of all projections along a spiral running on a sphere from one pole to the equator (7). For coronary MRA, the total data acquisition is subdivided into i MAX interleaves, where each interleaf is acquired within one cardiac rest period in late diastole. Each interleaf comprises p MAX projections, and successive interleaves are derived from the first one by a polar rotation about a polar angle,. The corresponding normalized readout gradient strengths G X, G Y, G Z are given by: with G X p,i cos( ) 1 G Z 2 p G Y p,i sin( ) 1 G Z 2 p G Z p,i p i MAX 0.5 p MAX, [1] 1 Institute of Biomedical Engineering (IBT), Karlsruhe, Germany. 2 Philips Research Laboratories, Hamburg, Germany. *Correspondence to: Christian Stehning, Institute of Biomedical Engineering (IBT), Kaiserstrasse 12, D Karlsruhe. stehning@directbox.com Received 3 November 2003; revised 14 January 2004; accepted 10 February DOI /mrm Published online in Wiley InterScience ( Wiley-Liss, Inc p MAX 2 i sin i 1 (G Z p ), [2] MAX i MAX where p and i denote the p-th projection of the i-th interleaf. This parameterization was chosen because it provides nearly isotropic k-space coverage for an arbitrary number of interleaves and projections per interleaf. The resulting

2 198 Stehning et al. trajectories are illustrated in Fig. 1. The overall anisotropy of the distribution, as measured by the standard deviation (SD) of the distance between adjacent points on the sphere, is 10% for a total number of projections n p MAX i MAX 100 (7). Neglecting the slight anisotropy, n ( /2) s 2 radial projections are necessary to fully satisfy the Nyquist criterion. In this context, s denotes the number of samples along one projection that is necessary to sample the desired FOV with the given resolution. To ease a comparison of scan times between a 3D-radial and a 3D-Cartesian acquisition, a sampling ratio defined as n/s 2 is introduced. Thus, a sampling ratio of 100% fulfills the Nyquist criterion for a Cartesian acquisition, but already implies undersampling for a radial acquisition. The corresponding point-spread functions (PSFs) for different sampling ratios are shown in Fig. 2. To ease visualization, only the central slice of the 3D PSFs is given. For sampling ratios lower than 100%, aliased energy from undersampled regions of k-space becomes visible as radial streaks within the FOV, leading to small non-zero background signal. More importantly, aliasing affects the overall intensity distribution, and may lead to intensity modulations over the image (8). The maximum amplitude of the streaking energy is approximately 1% of that of the central peak for the lowest sampling ratio of 12.5%. FIG. 1. 3D radial sampling with nearly isotropic k-space coverage. The acquisition is divided into interleaves (two of which are shown). 3D Image Reconstruction and SNR Considerations The images were reconstructed with the use of a fast gridding algorithm (9) extended to 3D. To compensate for the nonuniform k-space sampling density, an appropriate weighting of the individual samples is necessary. For this purpose, two different strategies were investigated in this work. The first one is based on the analytical functional determinant for 3D polar coordinates, which results in a quadratic weighting function. The second strategy is based on a numerical calculation of a weight for each sample using an iterative procedure (10). Given a set of weights W m, an improved set W m 1 is derived according to W m W m 1 W m C, [3] where C denotes the gridding kernel, and V is the convolution operation (10). An initial estimate W 0, for instance, FIG. 2. PSF over two FOVs for the central slice for different sampling ratios: (a) 100%, (b) 50%, (c) 25%, and (d) 12.5%. Level and window settings were adapted to brighten low-intensity aliasing components.

3 Volumetric MRA Using 3D-PR Balanced FFE 199 FIG. 3. Weighting functions for a single projection of a 3D radial acquisition. Dashed line: Analytical (quadratic) weighting function. Solid line: Iteratively calculated set of weights for one projection. is obtained from the analytical (quadratic) approach. For a typical 3D PR trajectory, 10 iterations usually provided a sufficient accuracy. The resulting weighting function for one projection of a 3D radial acquisition with 256 samples per projection and a sampling ratio of 25% is shown in Fig. 3. The numerical result resembles the analytical result for the inner part of k-space, but flattens toward the outer part of k-space, which is in good agreement with the results in Ref. 10 for a 2D radial acquisition. For both strategies, the strongly nonuniform data weighting entails a loss in the signal-to-noise ratio (SNR). In analogy to the derivation in Ref. 11, the remaining SNR for a quadratic weighting function (cf., Fig. 3) results in an SNR factor of SNR Radial SNR Cartesian [4] 9 compared to an uniform data weighting. The numerically calculated set of weights (cf., Fig. 3) yields a slightly improved result (Eq. [5]) because the variance of the data weights is slightly smaller compared to the quadratic weighting: SNR Radial, Iterative SNR Cartesian [5] However, the SNR losses in Eqs. [4] and [5] are an inherent property of the radial acquisition, and must be offset in practice by other advantages of radial sampling to justify its use. Moreover, it requires a high signal level to start with, which calls for the use of multiple surface coils for signal reception, as well as adequate sequences. Phase Correction and Eddy Current-Related Artifact Reduction Compared to Cartesian sampling, radial sampling is more prone to image artifacts caused by errors in the actual k-space trajectory, since all acquired projections pass through the center of k-space. Therefore, correction of the deviations is mandatory. Gradient channel anisotropies, approximated as short temporal delays of the corresponding waveform, are a major source of such deviations (12). In analogy to Ref. 13, we measured these delays by acquiring six projections along the main physical gradient channel axes immediately after the image acquisition. The shift k in k-space corresponds to a measurable linear phase twist exp(j2 k) between the forward and backward projections on each physical axis. The calculated shift was used to adjust the trajectory in the gridding reconstruction. To correct the absolute phase, we calculated a mean absolute phase for each projection and adjusted it to zero before image reconstruction, using a magnitude-weighted phase estimation. Another possible source of artifacts is rapid changes of the dephasing gradient orientation that may result in eddy current-related imperfections in the balanced gradient scheme. The resulting signal oscillations result in image artifacts, as recently reported for Cartesian sampling schemes (14). For a radial projection order, it is therefore important to keep the spatial angle increment between subsequent radial projections small. A typical implementation of a 3D-radial imaging sequence using the projection order discussed in the first section has by design a smooth gradient alternation with an almost constant angular increment of 8. This helps to minimize the above-mentioned eddy current-related artifacts. MR Sequence Phantom and in vivo experiments were performed on a whole-body 1.5T MR system (Gyroscan INTERA 1.5T, Philips Medical Systems) equipped with a five-element cardiac synergy receive coil. To test the achievable resolution at different sampling ratios, we used the resolution phantom shown in Fig. 4 (diameter 220 mm, thickness 50 mm). We measured the spatial resolution by positioning an intensity profile across the structures. For a signal drop between two structures of more than 50%, the resolution was considered sufficient to resolve this structure. For the SNR evaluation, we defined SNR as the ratio of the mean signal intensity (SI) of a bright region within the phantom and the SD of the noise SI in a region outside the phantom. The in vivo experiments were performed on 10 healthy male volunteers (age range years). Written informed consent was obtained from all participants. A fat-suppression prepulse was applied prior to signal sampling to suppress signal from epicardial fat. The image acquisition was ECG-triggered, with a delay of 600 ms, which was slightly altered according to each subject s heart rate. This helped us position the data acquisition window in late diastole, where motion is expected to be minimal (15). To eliminate artifacts caused by respiratory motion, the measurement was navigator-gated with a gating window of 5 mm. The navigator, based on a 2D pencil beam RF pulse, was placed at the right hemidiaphragm. It was applied immediately before the image acquisition to ensure precise motion information. To cope with residual motion within the defined respiratory gating window, a prospective motion correction technique was employed. The underlying rigid body motion model assumes a fixed correlation factor (cf 0.6) between the SI translation of the diaphragm and the SI displacement of the heart (16). For motion compen-

4 200 Stehning et al. FIG. 4. a: Reconstructed images of a resolution phantom. The phantom was imaged with different sampling ratios: (b) 100%, (c) 50%, (d) 25%, and (e) 12.5%. b e: A selected region of the phantom. sation, the excitation pulse and the acquired radial echoes were translated in frequency by appropriate shifts f tr and f rc, respectively, f tr G s cf d NAV [6] f rc G m cf d NAV cos( ) [7] where denotes the gyromagnetic ratio, G s and G m are the strengths of the slice selection and readout gradients, d NAV is the actual navigator displacement, and is the azimuthal angle of the current projection. A balanced FFE sequence ( 60 ) was used to sample with a high SNR and to increase the contrast between myocardium and blood. A /2-TR/2 start-up sequence (17) followed by 20 dummy cycles was employed to establish the steady state. The data acquisition window within the cardiac cycle was set to 136 ms, which is rather long but is feasible for most volunteers (18). The following sequence parameters were used: matrix 256 3, FOV (300 mm) 3, measured voxel size (1.17 mm) 3,TE 2.1 ms, TR 4.2 ms, radial sampling ratio 25%, and receiver bandwidth 781 Hz/pixel. The data from each receive coil were separately reconstructed, and the resulting image magnitudes were combined. The reconstruction time was approximately 30 s per coil on a standard PC (Pentium 1 GHz). To visualize the RCA and LAD, the isotropic image data set was reformatted (19). The total scan time was min at a respiratory gating efficiency of 50%. RESULTS Phantom Study Experiments using four different sampling ratios (100%, 50%, 25%, 12.5%) were performed. Furthermore, the two different sampling density compensation functions were used during image reconstruction. Selected phantom results are shown in Fig. 4. The measured SNR and achieved spatial resolution at the different sampling ratios are summarized in Table 1. The measured SNR achieved with the iterative weighting was higher than that obtained with the quadratic weighting. The differences were larger than the theoretical values calculated in Eqs. [4] and [5], since the iterative weighting strategy underweights high spatial frequencies. Consequently, a slight decrease of the resolution was observed at the lowest sampling ratio when the iterative weighting approach was used. However, a loss in resolution was not observed when a sampling ratio of at least 25% was chosen. The SNR further decreased with the sampling ratio, as the actual amount of measured data decreased. For a comparison of the image quality, separate phantom images reconstructed with the two different sampling density correction strategies are shown in Fig. 5. Both images were acquired with a sampling ratio of 25%. The additional effort of iteratively calculating a set of data results in a slightly improved image quality, as fewer artifacts are visible and the overall image homogeneity is improved. Table 1 Resolution and SNR at Different Sampling Ratios Using Different Weighting Functions Sampling ratio (%) Res. (mm) a Res. (mm) b SNR a SNR b a Denotes the analytical, quadratic weighting approach. b Denotes the iteratively determined weighting approach.

5 Volumetric MRA Using 3D-PR Balanced FFE 201 FIG. 5. Reconstructed images of a resolution phantom using a sampling ratio of 25% and two different weighting functions: (a) analytical (quadratic) and (b) iteratively calculated set of weights. On the bottom right of a and b, a section of the resolution phantom is shown at a larger scale. In Vivo Results Throughout the volunteer study, a large volume comprising the entire heart was scanned. To illustrate the volume coverage, different views obtained from a single 3D data set are shown in Fig. 6. No significant streaking artifacts are visible. Furthermore, no motion artifacts appear in the reconstructed images. In Fig. 7, selected results are shown for the RCA and LAD. We were able to visualize the RCA and proximal parts of the LAD retrospectively from the 3D data set. DISCUSSION With regard to the phantom study, it is noteworthy that with the use of the presented 3D radial acquisition, the sampling ratio may be decreased to 12.5% compared to a Cartesian acquisition. Acceptable artifact levels, and no significant decrease in resolution were observed. This result is in good agreement with the findings from the PSF calculation described in the Materials and Methods section. Therefore, artifacts are not a predominant limitation of undersampled 3D radial acquisitions for high-contrast applications such as MRA. However, the SNR strongly decreases with the sampling ratio. The minimum useful sampling ratio for in vivo measurements is therefore not limited by the undersampling artifacts; rather, it is predominantly limited by the available SNR. The observed decrease in SNR is in fair agreement with theory (square root of number of measured data). However, the way the SNR was measured did not distinguish between random, uncorrelated noise, and systematic errors introduced, e.g., by radial streaking artifacts. For this reason, some deviations between the measured and the expected SNR are to be expected. For coronary MRA, a sampling ratio of 25% was chosen, which offered a good trade-off between SNR and scan time. The iterative weighting increased both SNR and image quality without significant loss of resolution. Since individual weights are calculated for each sample, the numerical weighting strategy also takes the slight anisotropy of the sampling pattern as well as shifts of the projections introduced by eddy currents into account. However, the numerical weighting strategy requires considerable computation time. Currently, the weights have to be calculated in advance for each trajectory. Compared to a conventional coronary MRA protocol, the examination was considerably simplified. One breath-hold localizer scan was sufficient to plan the volumetric image acquisition. All of the scans were completed successfully, and we were able to visualize the major parts of the coronary arteries retrospectively from the volumetric data set. Only the reformatting of the coronary arteries along a FIG. 6. Reformatted coronal, sagittal, and transversal views of a 3D volume data set, illustrating the large volume coverage.

6 202 Stehning et al. FIG. 7. Reformatted views of the coronary arteries reconstructed from a volumetric data set acquired with 25% undersampling. curved surface remained a tedious procedure that required considerable user interaction. One possible way to improve this procedure would be to implement a simple planar reformatting approach (for instance, with a threepoint-plan-scan utility). The total scan time of a conventional protocol comprising a high-resolution localizer scan ( 3 min at 60 bpm during free breathing) and two individual high-resolution scans for the RCA and LAD ( 5 min each) is approximately 13 min, not including the time needed for planning or repetition of misplanned scans. The total scan time of our protocol was approximately min (at 60 bpm during free breathing), which is comparable to or even shorter than the scan time required for a conventional MRA protocol. Undersampling k-space in all three dimensions at a sampling ratio of 25%, and prolonging the acquisition window resulted in a decrease in scan time by a factor of 6, compared to an equivalent volumetric, isotropic Cartesian acquisition. No significant artifacts were visible. This is partly due to the still acceptable PSF of the 3D-radial acquisition at low sampling ratios, but is also a consequence of the robustness against motion of the radial acquisition combined with the prospective motion correction technique implemented in this sequence. Improved image quality or decreased scan time, or both, may be achieved by the use of a 3D-affine motion correction approach (20). In principle, this idea could also be extended to cope with cardiac motion in the image acquisition window for increased scan efficiency. The contrast between myocardium and blood could be improved by the use of magnetization preparation prepulses (T 2 prep) or higher flip angles; however, because of SAR limitations, these options have not yet been applied. The overall noise level was higher compared to that in images obtained with conventional thin-slab techniques (1,5). This drawback results partly from the SNR decrease due to undersampling and nonuniform data weighting, but it is also a consequence of the small voxel volume (1.17 mm) 3 in such an isotropic scan. Techniques that generally provide high SNR, such as balanced FFE sequences and phased-array coils, were not sufficient to reach an acceptable SNR level. For future applications, improved multiple-element receive coils and a transition to higher field strengths may further help to overcome the SNR problem. However, a more detailed comparison with conventional MRA protocols still needs to be performed. CONCLUSIONS A 3D radial sampling method combined with a balanced FFE sequence and prospective motion correction was implemented on a clinical scanner and employed for coronary MRA. The volumetric scanning technique offers a considerable improvement over current protocols, since the imaging session is greatly simplified and the risk of misplanned slices is minimized. Multiple anatomical details can be reconstructed from a single volumetric data scan without the need for additional imaging. The employed 3D radial acquisition allows a high degree of undersampling, which increases the acquisition speed significantly. However, there is a trade-off between the reduced scan time and a decreased SNR. A comparison with alternative fast volumetric imaging strategies, such as Cartesian sampling using parallel imaging techniques, could be the aim of a future study.

7 Volumetric MRA Using 3D-PR Balanced FFE 203 REFERENCES 1. Stuber M, Botnar RM, Danias PG, Sodickson DK, Kissinger KV, Van Cauteren M, De Becker J, Manning WJ. Double-oblique free-breathing high resolution three-dimensional coronary magnetic resonance angiography. J Am Coll Cardiol 1999;34: Börnert P, Jensen D. Coronary artery imaging at 0.5 T using segmented 3D echo planar imaging. Magn Reson Med 1995;34: Weber OM, Martin AJ, Higgins CB. Near-isotropic whole-heart coronary MRA. In: Proceedings of the 11th Annual Meeting of ISMRM, Toronto, Canada, p Glover GH, Pauly JM. Projection reconstruction techniques for reduction of motion effects in MRI. Magn Reson Med 1992;28: Larson AC, Simonetti OP, Li D. Coronary MRA with 3D undersampled projection reconstruction TrueFISP. Magn Reson Med 2002;48: Barger AV, Block WF, Toropov Y, Grist TM, Mistretta CA. Time-resolved contrast-enhanced imaging with isotropic resolution and broad coverage using an undersampled 3D projection trajectory. Magn Reson Med 2002;48: Wong ST, Roos MS. A strategy for sampling on a sphere applied to 3D selective RF pulse design. Magn Reson Med 1994;32: Lauzon ML, Rutt BK. Effects of polar sampling in k-space. Magn Reson Med 1996;36: O Sullivan JD. A fast sinc function gridding algorithm for Fourier inversion in computer tomography. IEEE Trans Med Imaging 1985;MI4: Pipe JG, Menon P. Sampling density compensation in MRI: rationale and an iterative numerical solution. Magn Reson Med 1999;41: Pipe JG, Duerk JL. Analytical resolution and noise characteristics of linearly reconstructed magnetic resonance data with arbitrary k-space sampling. Magn Reson Med 1995;34: Aldefeld B, Börnert P. Effects of gradient anisotropy in MRI. Magn Reson Med 1998;39: Wieben O, Brodsky E, Mistretta CA, Block WF. Correction of trajectory errors in radial acquisitions. In: Proceedings of the 11th Annual Meeting of ISMRM, Toronto, Canada, p Scheffler K, Hennig J. Eddy current optimized phase encoding schemes to reduce artifacts in balanced SSFP imaging. In: Proceedings of the 11th Annual Meeting of ISMRM, Toronto, Canada, p Kim WY, Stuber M, Kissinger KV, Andersen NT, Manning WJ, Botnar RM. Impact of bulk cardiac motion on right coronary MR angiography and vessel wall imaging. J Magn Reson Imaging 2001;14: Wang Y, Riederer SJ, Ehman RL. Respiratory motion of the heart: kinematics and the implications for the spatial resolution in coronary imaging. Magn Reson Med 1995;33: Deimling M, Heid O. Magnetization prepared true FISP imaging. In: Proceedings of the 2nd Annual Meeting of ISMRM, San Francisco, p Stehning C, Börnert P, Nehrke K, Schäffter T, Dössel O. Radial balanced FFE imaging with extended sampling windows for fast coronary MRA. In: Proceedings of the 11th Annual Meeting of ISMRM, Toronto, Canada, p Etienne A, Botnar RM, Van Muiswinkel AM, Boesiger P, Manning WJ, Stuber M. Soap-bubble visualization and quantitative analysis of 3D coronary magnetic resonance angiograms. Magn Reson Med 2002;48: Manke D, Nehrke K, Börnert P. Novel prospective respiratory motion correction approach for free-breathing coronary MR angiography using a patient-adapted affine motion model. Magn Reson Med 2003;50: