Continuously Moving Table MRI with SENSE: Application in Peripheral Contrast Enhanced MR Angiography

Abstract

An integration of SENSitivity Encoding (SENSE) with continuously moving table (CMT) MRI for extended field-of-view (FOV) acquisitions is described. In this work, the approach in which receiver coils are attached to the object and move in synchrony with the scanner table is considered. Technical issues dealing with the implementation of SENSE-CMT are addressed, including coil calibration, correction for non-uniform magnetic gradients, and specific reconstruction steps. An explanation of combining SENSE with gradient non-linearity correction is given, as the latter becomes necessary in CMT acquisitions where a large sampling FOV is used. It is hypothesized that SENSE can provide at least a 2-fold improvement in lateral spatial resolution compared to non-accelerated CMT acquisitions. The hypothesis is tested in phantoms, where the effectiveness of both SENSE and gradient non-linearity correction to improve spatial resolution is shown. The SENSE-CMT technique is further demonstrated in vivo with contrast-enhanced MR angiography of the peripheral vasculature.

Recently, techniques for continuously moving table (CMT) MRI have been introduced (1–6), and a number of them have been applied to contrast-enhanced MR angiography (CE-MRA) of the peripheral vasculature (2,5,6). CMT methods dynamically acquire data as the table is moved through the scanner and can provide a seamless image over an extended longitudinal field-of-view (FOV). In peripheral CE-MRA, where the longitudinal FOV is typically 100 cm or more, a CMT technique is an alternative to a multi-station approach (7–10), eliminating the time overhead required in the latter technique to move the table from station to station.

In peripheral 3D CE-MRA, spatial resolution is critical. Specifically, in CMT acquisition, lateral spatial resolution depends on how long a specific level lies within the imaging volume and is proportional to the length of the actively sampled volume and inversely proportional to the repetition time TR and the table velocity used (2). Furthermore, aligning the sampling FOV over the arterial phase of the contrast bolus is required for a successful exam. This constrains the table velocity to a certain minimum and limits the obtainable spatial resolution. With increasing use of parallel imaging in fixed-FOV MRI (11), the purpose of this work was to investigate the feasibility of implementing SENSitivity Encoding (SENSE) (12) in CMT applications, specifically for the case of frequency readout along the longitudinal axis. We hypothesize that with otherwise fixed acquisition parameters, SENSE can be used to improve by at least 2-fold the lateral spatial resolution in CMT-MRA.

The integration of SENSE with CMT presents several technical challenges, such as coil coverage and sensitivity calibration, as described in preliminary works (13,14). These questions have recently been described for SENSE-CMT with a receiver pair fixed to the magnet gantry (15). In contrast, the current work describes SENSE-CMT using a coil array that moves along with the patient table. Additionally, we consider in detail the role of gradient non-linearity effects in SENSE-CMT, a factor that becomes important when a large sampling field-of-view is used, as is generally the case in CE-MRA. We test the hypothesis with a resolution phantom and apply the resultant SENSE-CMT approach to peripheral CE-MRA exams on 6 volunteers.

Methods

Surface Coils

Dedicated coil arrays have been applied to multi-station peripheral CE-MRA (8–10). Fig. 1 shows the 2 custom-made coil assemblies used in this work. Each is strapped to the subject and moves with the table during the CMT acquisition such that a receiver element is sensitive to the same portion of the imaging object during the scan. Each assembly contains 4 overlapping elements, providing 90 cm of longitudinal coverage from the iliac to the tibial arteries. Due to the current limitation of 4 receiver channels on our scanner, the 8 elements are sequentially activated and deactivated during a CMT scan. At any time during the scan, 4 elements (2 per side) are receptive to approximately 40 cm along the extended longitudinal axis of the interrogated volume.

FIG. 1.

SENSE-CMT coil assembly. Eight elements provide a longitudinal coverage of 90 cm. Pairs 1 and 2 are sensitive to the iliac and femoral arteries, while pairs 3 and 4 provide coverage distal to the popliteal trifurcation vessels.

Coil Calibration and Image Reconstruction

Coil sensitivity profiles are required for SENSE reconstruction. In CMT, profiles are needed along the entire longitudinal extent of the imaging volume. There are many options for collecting this information, and a suitable choice may depend on the severity of field non-linearity over the acquisition volume as limited by the encoding gradients. We refer to this as “gradient non-linearity” or “gradient warping.” In fixed-FOV MRI, gradient non-linearity effects are seen as warping of objects located at the edges of a large FOV (16). In CMT-MRI, they are manifested as additional blurring (2). We first describe calibration and reconstruction steps potentially applicable if gradient warping effects are not an issue, and then present the more generalized case that allows for them. In this discussion, a SENSE factor of 2 along the lateral Y-axis and a 2D CMT acquisition using sequential view order and readout in the longitudinal direction are assumed for simplicity. In practice, 3D acquisitions were performed. Also, we define FOV_S as the longitudinal extent of the actively sampled volume.

Negligible Gradient Warping

Consider first the case in which gradient warping is not a limitation, such as for FOV_S less than about 20 cm. For such cases, coil profiles can be acquired using either a fixed-table or CMT approach. Using the fixed-table approach, 1 or more calibration scans can be acquired for each coil pair with a static FOV_S centered within the imaging volume. These individual profiles are then juxtaposed to produce the extended FOV calibration, analogous to the approach described in Ref. 15. To spatially register the coil calibration and SENSE scans, the acquisition of central k-space views during the SENSE-CMT scan should occur at table positions matching the stationary FOV_S of the profile scan. This is facilitated by using the same CMT FOV_S for calibration and SENSE, with FOV_S not larger than the dimensions of the coil elements (15).

An alternative method is the CMT calibration approach, where profiles of the extended FOV are obtained in a single scan. A similar k-space synchronization requirement can be imposed where the table must travel the same longitudinal distance for one complete cycle of views in both the profile and the SENSE-CMT scans. Using Eq. [6] of Ref. (2), this can expressed as:

$$\begin{array}{c}{\text{v}}_{\text{profile}}\cdot {\text{N}}_{\text{TOT},\text{profile}}\cdot {\text{TR}}_{\text{profile}}={\text{FOV}}_{\text{S}}\\ ={\text{v}}_{\text{SENSE}}\cdot {\text{N}}_{\text{TOT},\text{SENSE}}\cdot {\text{TR}}_{\text{SENSE}}\end{array}$$ [1]

where v and N_TOT are the table velocity and total number of phase encodes, respectively. In the steps leading to the final reconstructed result for both fixed-FOV and CMT calibration schemes, raw SENSE images are first reconstructed using the CMT algorithm described in Ref. (2). They are then unaliased using standard SENSE methods (12).

Gradient Warping

It is often desirable in CMT-MRA to make FOV_S as large as practical in order to maximize the volume of coverage with each repetition. In our own experience, the use of FOV_S larger than 20 cm introduces blurring as a consequence of gradient non-linearity. A correction algorithm for such artifacts has been previously described (17). Fig. 2 shows the “barrel-like” gradient-warping-induced contours for the lateral right/left (R/liter) axis. In principle, each contour represents a distinct Y location in image space, but the gradient-related non-linearity in the R/L field causes the lateral FOV to in effect become progressively larger as one moves above or below the magnet isocenter. A pixel P experiences a progression of gradient non-linearity effects in CMT and is consequently blurred across a spectrum of lateral positions. In the final CMT image, P is represented as the sum of data acquired while it remained in the sampling FOV_S.

FIG. 2.

A particular pixel P (black circle) experiences a progression of gradient non-linearity effects in CMT. As the table moves along X, the location of P within FOV_S changes and the corresponding lateral position to which it is mapped varies, as dictated by the underlying warp contour (solid lines). Also shown is an R_Y = 2 SENSE-CMT acquisition, where aliased gradient non-linearity contours (dashed lines) are depicted. Points located ½ FOV_Y from P will alias (gray arrows and circles), but are subjected to gradient warping effects as well. As a result, a multitude of lateral positions fold onto P.

Also illustrated in Fig. 2 is an R_Y = 2 SENSE acquisition. Points that alias onto P are also subjected to the warped gradient field. A group of 3 aliasing points is illustrated along the right outermost contour. At time t₁, the rightmost of the group folds onto P, whereas at t₂ and t₃, the leftmost and middle of the group alias, respectively. Thus, a spectrum of lateral positions wraps onto P as it transits through FOV_S. Raw CMT images reconstructed prior to SENSE unfolding and uncorrected for gradient non-linearity are thus subject to a multitude of aliasing factors, potentially confounding the SENSE reconstruction process. Thus, SENSE unfolding must precede gradient warp correction.

Gradient non-linearity can be accounted for in SENSE-CMT by performing SENSE unfolding on a view-by-view basis, dictated by the notion that a separate 3D un-warping process is needed for each phase encode measurement (17). Thus, each measurement is first Fourier transformed into a sub-image corresponding to the specific k_Y-k_Z phase encode sampled. Gradient-warped calibration maps, calculated using coil images normalized by their sum-of-squares combination, are then used to unfold the aliased sub-image. The result is then corrected for gradient non-linearity. Finally, the SENSE-unfolded, gradient-warp-corrected sub-image is added to a running complex sum of such components, accruing until all acquired phase encodes are processed. To reduce computation load, a small group (e.g., 16–128) of consecutively sampled views can be reconstructed at a time if the corresponding distance of table travel is short (e.g., 1–2 cm), within which the differential rate of gradient warping is small (17). A flow chart of the proposed reconstruction algorithm is shown in Fig. 3.

FIG. 3.

Reconstruction algorithm for SENSE-CMT with gradient warp correction. Groups of echoes are treated progressively. (I, II) Data are position-corrected, Fourier transformed along X into hybrid space, and allocated to the X-position corresponding to the table location at the time of measurement. (III) Fourier transformation along the remaining axes yields a partial image, where SENSE-unfolding is performed. (IV) Gradient warp correction. (V) The result is progressively summed with corrected data from previous view groups.

In theory, calibration information obtained from either fixed-table or CMT acquisitions can unfold gradient-warp-corrupted SENSE-CMT data. Both assume approximations in the reconstruction and are compared in Fig. 4. In Fig. 4a, the positions of 2 adjacent stationary calibration scans (1 and 2) are shown schematically in the form of gradient warp contours. The gray circle represents a hypothetical phantom. Alternatively, an additional overlapping calibration scan (set 3) can be obtained at the expense of a longer calibration time. In Fig. 4b, several sub-images of length FOV_S are depicted as taken from a presumed SENSE-CMT acquisition. For illustration simplicity, only half of the gradient-warping contours are shown and aliasing contours are omitted. Similarly illustrated in Fig. 4c are sub-image components from a CMT calibration scan, the sum of which yields the final extended-FOV profile. For nearly all X locations, it is evident that neither calibration approach produces an exact SENSE reconstruction as a consequence of gradient non-linearity. Instead, both yield profiles that approximately reflect the aliasing and gradient warping effects present in the SENSE data. Consider the first SENSE sub-image (black lines in b). It can be unfolded by the fixed calibration approach since the coil maps closely reflect the warping and aliasing of the data. In contrast, unfolding of the third SENSE sub-image (red) with fixed-table calibration profiles will yield differential degrees of error since none of the 3 stationary sensitivity maps exactly coincides with the gradient warping effects of the SENSE sub-image. The summed CMT calibration profile, however, contains contribution from at least 1 sub-image component (dashed black, dashed red) that was acquired at the exact corresponding X locations as the first (black) and third (red) SENSE sub-images, and hence may produce a more accurate approximation of the SENSE reconstruction.

FIG. 4.

Gradient-warp corrupted SENSE-CMT data (b) can be more accurately unfolded using a CMT-acquired calibration (c) rather than fixed-FOV profiles (a). A SENSE-CMT acquired resolution phantom reconstructed using adjacent stationary calibrations 1 and 2 is shown in (d), with significant errors. (e) Using overlapping stationary calibrations (1–3) gives a better approximate reconstruction with reduced errors. (f) However, the same data reconstructed with a CMT calibration depicts a more accurate result. Lateral resolution improvement over a non-accelerated reference scan (g) is evident (dashed box).

Phantom Experiments

Two phantom experiments were performed on a 1.5T GE Signa LX scanner (GE Medical Systems, Milwaukee, WI, USA). In the first setup, a resolution phantom was used to compare the quality of the SENSE reconstructed image by using both fixed-table and CMT calibration scans in the presence of gradient non-linearity effects. We show that CMT calibration profiles yield superior SENSE reconstructions to those from fixed-FOV scans. In the second experiment, we demonstrate with a CMT acquisition that a combined SENSE and gradient-warping-corrected method can provide superior lateral spatial resolution in comparison to each of the techniques individually. A resolution phantom was used to assess R/L spatial resolution and was intentionally placed near the edge of the imaging FOV such that gradient-warp-induced blurring would be significant. For both experiments, a reference non-accelerated CMT acquisition was included with coil calibration and SENSE-CMT accelerated scans (R_Y = 2). A sum-of-squares reconstruction was performed for the reference scan. Additional parameters are summarized in Table 1.

Table 1.

Scan Parameters for Experiments

	Phantom experiment	Phantom experiment	Peripheral CE-MRA
Figure no.	4	5	6
Pulse sequence	3D Spoiled GRE	3D Spoiled GRE	3D Spoiled GRE
FOVs (cm)	36	36	25
FOV phase (cm)*	30	30	30 to 44
TR/TE (ms)	20.0/3.0	25.0/2.0	6.0/1.8
Flip angle (degrees)	30	30	30
Slab thickness (mm)	64	32	112 to 144
Partition thickness (mm)	4	4	7 to 9
Sampling matrix (read/phase/slice)	192/128/16	192/128/8	128/128/16
Voxel size, read × phase × slice (mm3)*	1.8 × 2.3 × 4	1.8 × 2.3 × 4	1.9 × (2.3–3.4) × (7–9)

^*Values reported for the reference (non-SENSE) scans. SENSE FOV along the Y-axis is reduced 2-fold, and corresponding pixel resolution is improved 2-fold.

Peripheral Vasculature CE-MRA Studies

Six volunteers underwent peripheral CE-MRA exams with SENSE-CMT using the coil array and reconstruction algorithm described. Informed consent was obtained from all participants, and the studies were approved by the institutional review board. Each exam was performed on a 1.5T GE Signa LX scanner and included 3 CMT scans: a 90-s coil calibration, a 50-s pre-contrast mask scan, and a 50-s contrast-enhanced study. Parameters for the latter 2 scans are shown in Table 1. In our experience, use of the CE-MRA gradient echo pulse sequence with a short TR and a large flip angle for coil calibration provides inadequate SNR for determining sensitivity profiles. Increasing the TR and reducing the flip angle yields calibration images that are more robust for SENSE reconstruction. With a minimum table velocity of 1.0 cm/sec on the scanner, a maximum TR of 11 msec was used for calibration with FOV_S = 24.6 cm, N_Y = 128, and N_Z = 16. Sequential view order was used with slice (Z-A/P) and phase (Y-R/L) encodings in the outer and inner loop, respectively. Subjects were injected with 20 mL of Gadolinium contrast (Omniscan, Amersham Health Inc., Princeton, NJ, USA) at 1.5 mL/sec followed by a 20 mL saline flush via an electronic injector (Medrad Spectris Solaris, Medrad, Inc., Indianola, PA, USA). Table motion was fluoroscopically triggered at the level of the abdominal aorta (18).

Results

Figs. 4d–g show results from the first phantom experiment. The relative location of the resolution phantom within the extended FOV is represented by the gray circle in Fig. 4a. Fig. 4d shows erroneous SENSE reconstruction using only adjacent fixed-FOV calibration profiles 1 and 2, with the gradient-warp-distorted image displayed at a window/level to accentuate the wrapping artifacts. Using all 3 overlapping calibration scans (sets 1–3), the approximate reconstruction is improved with reduced errors, as shown in Fig. 4e. Nevertheless, reconstruction of the same SENSE data with CMT calibration profiles yielded a more acceptable result, as shown in Fig. 4f. Lateral resolution improvement over a reference non-accelerated CMT scan (Fig. 4g) is evident (dashed box).

Fig. 5 illustrates results from the second phantom experiment. Fig. 5a shows the off-center placement of the resolution phantom with respect to the imaging FOV. Figs. 5b-e illustrate a magnified portion of the phantom. Figs. 5b and c were both reconstructed from the same non-SENSE data set, uncorrected and corrected for gradient non-linearity, respectively. Similarly, Figs. 5d and 5e show the same SENSE reconstructed data set, without and with gradient warp correction, respectively. SENSE reconstructed images show negligible unfolding artifacts, with a g factor range between 1.2 and 1.5. In Fig. 5e, values denote the number of line pairs per centimeter (lp/cm) for each resolution bar set. Comparing Fig. 5c (reference) with Fig. 5e (SENSE), the 2-fold improvement in lateral spatial resolution due to SENSE is evident, as the 1.52 lp/cm pattern is clearly seen in the former while the 3.13 lp/cm set is clearly resolved in the latter. The benefit of gradient non-linearity correction can also be observed by comparing either Fig. 5b with c or Fig. 5d with e. Note in both cases that all resolution bar sets, especially those located at the edge of the FOV (sets 1.52, 3.13, and 3.85), become markedly sharper with gradient non-linearity correction.

FIG. 5.

Results from second phantom experiment. (a) Off-center placement of phantom with respect to imaging FOV prior to table motion. (b) Reference image without gradient warp correction. (c) Reference image with gradient warp correction. (d) SENSE R_Y = 2 reconstruction without gradient warp correction. (e) SENSE R_Y = 2 with gradient warp correction. Note spatial resolution improvement due to gradient warp correction (row-wise comparison) and SENSE (column-wise comparison). Values in (e) denote line pairs per centimeter for each bar set.

Figs. 6a and d show mask-subtracted maximum intensity projections from 2 of the 6 in vivo CE-MRA studies, reconstructed following the algorithm in Fig. 3. Both illustrate approximately 90 cm of the peripheral vasculature from the femoral arteries to the tibial vessels with minimal unfolding artifacts. Enlarging the dashed region in Fig. 6a, Fig. 6b versus c illustrates the improvement in image quality provided by gradient warp correction. In the gradient-warp-corrupted image (b), secondary and tertiary vessels originating from the right femoral artery are blurred, due to their more lateral location within the FOV. Note the improved details of these lateral vessels with gradient warp correction (arrowheads in c).

FIG. 6.

(a) Mask-subtracted result from R_Y = 2 SENSE-CMT runoff CE-MRA study. (b, c) Enlargements of dashed region in (a), illustrated (b) without and (c) with gradient warp correction, respectively, from the same volunteer data set. Note improved conspicuity of secondary and tertiary branch vessels (arrowheads). In (d), result from another volunteer, SENSE reconstruction artifacts originating from aliased calibration scans are present, with lateral pixels beyond the prescribed FOV_Y experiencing three-fold wrap-around (center dashed oval).

Discussion

We have demonstrated a method to integrate SENSE with CMT imaging for the specific approach in which the receiver coils move with the subject. We have also described the effects and addressed gradient-warp-induced artifacts using a view-segmented reconstruction algorithm. In our own experience, these artifacts if uncorrected are problematic in CMT imaging if a sampling volume larger than about 20 cm is used. In this work, a SENSE-CMT approximate reconstruction algorithm was proposed that uses CMT calibration profiles. Since the calibration images are also subject to gradient warping and blurring, potential inaccuracies in local coil sensitivities can occur. However, this has not been a major issue because the degree of blurring is much smaller than the slowly varying nature of the profiles. In theory, an exact SENSE reconstruction can be obtained by first acquiring a continuum of separate stationary calibration images, but this would be impractical because of excessive acquisition time.

While the algorithm performed successfully in all 6 volunteers, 3 exams did not yield acceptable angiograms. One case was our first attempt, and the constructed coil array did not yield adequate signal for visualizing the vasculature. The remaining 2 cases were hindered by subtraction errors and misregistration between mask and contrast-enhanced images. This is likely due to patient motion or loosely secured coils that may have moved during the exam. Further work in improving CMT coil reception and secure attachment to the subject should remedy these issues. Using the presented algorithm, reconstruction of each volunteer data set took 30–45 min on a desktop computer (Pentium-4 2.4 GHz processor, 512 MB memory). Parallel processing is expected to reduce this time markedly.

With only 4 receiver channels on our current scanner, a total of 8 elements was integrated into the coil assembly. Upgrading the receiver platform to 8–16 channels is expected to broaden the capabilities of the SENSE-CMT technique. In all of the in vivo studies, a frequency readout resolution of 128 points was used. However, higher values are obtainable, subject to bandwidth limit and SNR loss. A further limitation was the use of thick coronal slices, principally because the desired table velocity and prescribed FOV_S constrained the number of partitions in accordance with Eq. [1]. Increasing the FOV_S, now possible with gradient warp correction, along with shorter TR capabilities, will allow for thinner slices.

The volunteer study shown in Fig. 6d used a lateral (R/L) FOV of 30 cm. This introduced 3-fold aliasing for lateral structures that fell outside of the prescribed FOV_Y in the SENSE acquisition. An R_Y = 2 SENSE reconstruction successfully unfolded voxels containing vasculature, while erroneous results were obtained for those voxels that experienced 3-fold aliasing (dashed ellipse). These erroneous voxels should not in principle obscure visualization of the main vasculature primarily because they are comprised of non-vascular anatomy. This observation has been reported elsewhere (19,20).

Irrespective of parallel-imaging, a high quality angiogram fundamentally depends on the acquisition of k-space when contrast-enhanced blood is present in the actively sampled FOV_S. For the CMT case, this requires that the table velocity be closely matched to that of the contrast bolus as it moves through the peripheral vasculature such that peak vessel enhancement is maintained within the FOV_S. As previously demonstrated (2,6), the tradeoff between table velocity, FOV_S, and sampling matrix size can often be contradictory in CMT. By assuring compatibility between SENSE-CMT and gradient non-linearity correction, we have increased the flexibility of the CMT method such that these and other acquisition parameters can be better adjusted on a patient-specific basis in CMT applications.

Conclusions

The integration of parallel imaging with CMT-MRI is feasible. In this work, the implementation of SENSE-CMT with surface coils attached to the imaging object was demonstrated. This has required addressing technical issues, including coil sensitivity calibration, SENSE unfolding, and most importantly, compatibility with gradient non-linearity correction.