Recently, a new imaging technique, alternate ascending/descending directional navigation (ALADDIN), was developed and used for perfusion-weighted (PW) imaging (1) and also magnetization transfer (MT) asymmetry imaging (2), which detects asymmetry of MT effects around the water resonance attributed to the chemical shift center mismatch between bulk water and (free mobile or solid-like) macromolecules. The ALADDIN technique sensitizes interslice blood flow (perfusion) and MT effects induced in 2D sequential multislice acquisitions and requires no separate ratio frequency (RF) pulse for spin preparation. In ALADDIN, scans for labeling and control in PW imaging and those for positive and negative frequency offsets in MT asymmetry (MTA) imaging are performed by changing acquisition orders (ascending/descending) and slice-select (SS) gradient polarities (positive/negative) of a 2D sequential imaging sequence. The ALADDIN PW and MTA imaging techniques share a common data acquisition process but require two separate data reconstruction methods that result in two distinctive ALADDIN image sets. However, these two ALADDIN image sets may not be uniquely determined when there are MR gradient imperfections such as eddy currents (3–9).
A balanced steady state free precession (bSSFP) sequence has been used for ALADDIN, in which the gradient is balanced (i.e., null zeroth moment) along each of the three directions. The phase-encode (PE) gradient, which is balanced between the positive and negative intensities within each period of time to repeat (TR), induces little eddy currents with the linear PE order (8). On the other hand, intensities of the positive and negative portions for the SS and readout (RO) gradients are not equal, thereby inducing eddy currents that are constant across multiple TR periods. Although contributions of these SS and RO eddy currents to conventional bSSFP imaging are negligible (8), they may generate mismatched MT frequencies leading to residual signals in the subtraction images of ALADDIN. In the previous studies (1, 2), ALADDIN signals were investigated with uniform phantoms that intrinsically have no MT effects.
In this article, we investigated the effects of gradient imperfections on ALADDIN signals at isocenter and off-center positions and proposed a new acquisition scheme that can suppress the potential errors from the gradient imperfections. We performed experiments on a distilled water phantom (no MT), agarose phantom (symmetric MT), and in vivo brain and skeletal muscle (asymmetric MT). We empirically demonstrated that gradient imperfections along RO direction associated with MT effects can generate subtraction artifacts on MTA images and that these artifactual signals can be suppressed by averaging signals over RO gradient polarities. To see the changes in residual signals induced by gradient imperfections along RO direction, we performed a TR-dependent study, where acquisition bandwidth (BW) hence RO gradient strength was modulated with TR. We discussed a potential explanation of relationship between eddy current-induced MT frequency shifts and artifactual signals and potential contributions of gradient imperfections to ALADDIN PW imaging.
Data Acquisition and Processing
All experiments were performed on a 3T whole body scanner (Siemens Medical Solutions, Erlangen, Germany). The study was approved by the institutional review board. Voxel-localized shimming was performed within a region covering whole brain. MTA images were acquired at the isocenter from a distilled water phantom, 4% agarose phantom, and in vivo brains of three normal male volunteers with a body coil for transmission and 12-element head matrix coil for reception. Additionally, images were acquired off-center along the X-axis from in vivo skeletal muscle of the three volunteers as well as the agarose phantom, with a circularly polarized RF coil for both transmission and reception. For ALADDIN imaging, we acquired four bSSFP datasets from alternating ascending/descending acquisitions and positive/negative SS gradients using positive RO gradient followed by the same four data acquisitions using negative RO gradient, as shown in Fig. 1. The same acquisition of these eight datasets was repeated once (total 16 repetitions). Imaging parameters were: matrix size = 128 × 128; field of view = 230 × 230 mm2; slice thickness = 5 mm; flip angle = 50°; PE order = linear; phase oversampling = 50% (used to increase MT saturation/labeling duration for subsequent slices in ALADDIN (1, 2), while typically used to avoid aliasing along the PE direction in conventional sequences); delay time between repetitions = 15 s for the distilled water phantom and 4 s for all the others; number of slices = 15; scan time per dataset = 4−4.5 min. The interslice gap was set to 140% of the slice thickness to avoid cross talk effects (1, 2). The RF pulse was Hanning-windowed sinc pulse with 1-ms duration and 1.6-kHz BW, inducing MT effects from the first prior slice at a 30-ppm frequency offset with the given slice thickness and gap values. The initial 10 PE steps with linearly increasing flip angles were used as dummy scans.
A TR-dependent study was performed at the isocenter with TR and acquisition BW varied from 3.76 to 4.18 ms and from 698 to 528 Hz, respectively, with average increment and decrement values of ∼0.05 ms and ∼19 Hz, respectively. The echo time was set to the half of the TR value in each case. For all the imaging at the isocenter, the scan direction and the PE direction were axial and anterior–posterior, respectively.
The off-center ALADDIN imaging was performed along the axial and sagittal directions at about 100 mm off the isocenter along the X-axis (left–right), with a fixed TR and acquisition BW values of 3.76 ms and 698 Hz. Both of the two possible PE directions were tested at each scan direction. The resonance frequency was recalibrated prior to each ALADDIN imaging to avoid B0 drifting effects.
Motion correction (10) was applied to datasets of varying SS gradient polarities and varying RO gradient polarities separately. This approach resulted in improved motion corrections, compared with those applied to the entire eight datasets or each of the eight datasets individually. After averaging the signals from the positive and negative RO gradients that were corrected for motion and averaged, percent signal changes (PSCs) between acquisitions with positive and negative MT frequencies were calculated (2).
To determine the PSC of the phantoms, a region of interest (ROI) covering >90% of the signal region was chosen from the center slice. The PSC values of MTA signals were calculated and averaged within the ROI. To determine the PSC of in vivo brain, whole white matter (WM) regions were manually segmented from the center slice of the baseline images. Only for in vivo brain, signal to noise ratio (SNR) was also calculated from the subtraction images, as the average signal within the WM ROI divided by standard deviation in a noise region manually defined outside the brain. MTA signals for in vivo muscle were computed by choosing a big ROI covering >90% of the muscular region within the center slice.
The agarose phantom, with the presence of symmetric MT effects, is suitable for testing contributions of gradient imperfections to ALADDIN MTA signals. Figure 2 shows the TR-dependent ALADDIN signals acquired for the agarose phantom at the isocenter. When only the positive RO gradient was used (+GRO), residual signals were detectable in MTA images (Fig. 2a,c). Within the TR range we tested, the residual signals became negative or positive depending on TR values and sometimes the polarities were mixed within a slice varying along the RO direction (e.g., second column from the right in Fig. 2a). Overall, the residual signals were stronger at lower TR values. With the new acquisition scheme (Avg.), the residual signals were suppressed independent of TR values and spatial locations (Fig. 2b,c). Note that susceptibility artifacts from air bubbles were observable in the MTA images (white spots in Fig. 2a) and could be suppressed by using the new scheme (Fig. 2b).
When the TR-dependent study was applied to human brain, the signal modulation as a function of TR was similar to that of the agarose phantom (Fig. 3a,c). The MTA signals varied less with TR by using the new acquisition scheme (Fig. 3b,c). In human brain, MTA signals in WM were ∼1.8% (dotted line in Fig. 3c), which was higher than those from the agarose phantom (∼0%, dotted line in Fig. 2c). SNR from the new acquisition scheme (∼5.0) was about times as high as that from the positive RO gradient only (∼3.6; Fig. 3d), indicating that the new method efficiently increased SNR in the same amount as the conventional averaging method. Unlike the agarose phantom and human brain, the distilled water phantom showed no signal or signal changes as a function of TR (data now shown), confirming that all the signal modulations in Figs. 2 and 3 are associated with combination of MT effects and gradient system imperfections.
The agarose phantom was also tested along axial and sagittal directions at an off-center position to investigate the effects of RO offsets and SS offsets, respectively, on the ALADDIN MTA signals. For these off-center imaging studies, we chose TR of 3.76 ms and acquisition BW of 698 Hz. The top row in Fig. 4a (+GRO) shows that the RO offset (∼100 mm) increased the MTA signals compared with those at the isocenter (leftmost column in Fig. 2a). When the RO direction was changed from right−left to anterior−posterior (Fig. 4b; RO offset was reduced down to ∼10 mm), the MTA signals became weaker than those in Fig. 4a. The residual signal levels for the sagittal scans were almost the same as Fig. 4b when the RO direction was the same (Fig. 4c vs. Fig. 4b) and significantly decreased when the RO direction was along feet to head (Fig. 4d). In either case, the polarity of MTA signals changed with the RO gradient polarity (middle rows in Fig. 4a−d), and the signals were eventually suppressed by averaging the positive and negative RO gradient data (bottom row in Fig. 4a−d, average 0.0% for all of them). In human muscle, we observed signal modulations similar to the agarose phantom (Fig. 4e−h). Note that the MTA signals in skeletal muscle became relatively consistent regardless of scan direction by using the new scheme (bottom row in Fig. 4e−h). MTA signals in the muscle were 0.51 ± 0.11% (N = 3), about 30% of those in brain WM.
In this study, we evaluated the effects of gradient imperfections on ALADDIN MTA signals and proposed a new acquisition scheme that could suppress them. The new averaging scheme will allow for more accurate MTA imaging with ALADDIN, especially at off-center positions. It requires a minimum number of repetitions to be ≥8 rather than ≥4 (i.e., doubling the scan time). However, ALADDIN requires number of repetitions to be ≥8 in most cases and the new scheme efficiently increased SNR in the same way as the conventional averaging method (Fig. 3d). Note that as long as the eight different kinds of datasets (Fig. 1) are acquired at least once (i.e., number of repetitions ≥8), number of repetitions does not have to be multiple of eight. Therefore, the penalty in scan time with the new method is minimal in ALADDIN. However, changing RO gradient polarities could be subjected to pixel shifts depending on the scan resolution and RO gradient strength, especially in the fatty tissue regions (chemical shift), although most brain regions could be intact under our experimental conditions.
Phase-encoding gradients in bSSFP are balanced in terms of both areas and intensities of the positive and negative portions, suppressing long-term eddy currents (main time constants of residual signals ≤1 ms) as demonstrated in the previous literature (8). However, intensities of the positive and negative portions are not the same for the RO and SS gradients, therefore they may be subjected to the long-term eddy current contributions. Our observation of stronger residual signals in lower acquisition BW (lower RO gradient, i.e., higher discrepancy between the positive and negative intensities; Fig. 2a) supports this notion. In this sense, higher acquisition BW is advantageous for ALADDIN due to reduction of eddy current contributions and also potential pixel shifts.
Eddy currents can shift not only interslice MT frequencies but also the location of slice excitation, which is subjected to the polarity of the SS gradient. Because the signals from the positive and negative SS gradients are averaged in ALADDIN, contributions of these slice shifts to ALADDIN signals would be minimal, as already demonstrated in the two previous ALADDIN studies for slice shifts associated with magnetic field inhomogeneity (1, 2). This indicates that the artifactual signals are likely associated with deviations in interslice MT frequencies.
A potential explanation for the relationship between the eddy current-induced MT frequency shifts and the artifactual signals are as follows. As the first order SS eddy current consistently increases or decreases the amplitude of the positive and negative SS gradients, its effect on the ALADDIN signals would be insignificant. This could be supported by an observation that an offset along the SS direction had no effect on artifacts in MTA images (Fig. 4) and PW images (data not shown). On the other hand, ALADDIN signals can be affected by frequency offsets in association with the RO eddy current (ΔFRO) and the SS B0 (zeroth order) eddy current ( ). Frequency deviations induced by the RO eddy current (ΔFRO) are independent of the SS gradient polarity, and thus will always be higher or lower than without the RO eddy current (Fig. 5b). As the polarity of is directly affected by the polarity of the SS gradient (GSS), the actual SS gradients will shift along the spatial distance domain as shown in Fig. 5c. The deviation along the frequency domain caused by the RO eddy current (vertical axis in Fig. 5b) can affect the MTA imaging, which is calculated with subtraction between datasets from the positive and negative MT frequencies. The deviation along the distance domain caused by the slice B0 eddy current (Fig. 5c) may affect the PW imaging calculated with subtraction between ascending and descending acquisitions. Again these frequency shifts affect both the interslice MT frequencies and the excitation locations of the slice of interest, but eventually the latter will not contribute to ALADDIN signals, as mentioned above.
We can roughly estimate how much MT frequency deviations and signal changes can be caused by RO eddy currents. The summation of the RO gradient (0.78 G/cm) and the RO dephasing gradient (−2.73 G/cm) will mainly account for the eddy currents that may survive until the subsequent RF excitation. If we assume that 0.1% of RO eddy current (0.00195 G/cm) persists in the subsequent RF excitation (11), it will induce a frequency offset of 83 Hz at a position of 10 cm offset from the isocenter, which corresponds to 0.65 ppm at 3 T. It should be noted that the eddy currents may be much more reduced in echo-planar imaging or spiral imaging sequences, because of relatively well balanced positive and negative portions of the RO gradients. According to the previous study (Fig. 1b in Ref.12), the slope of the tangential line for the MT-induced longitudinal magnetization (M0 = 1) versus saturation offset frequency was roughly 0.1 per 20 ppm, i.e., 0.5% of longitudinal signal change per ppm, at a 30-ppm frequency offset and 2-μT saturation power, which generated MT ratio close to our experimental conditions of ∼0.7 (data not shown). With linear PE order, the signal at the K-space center decreases down to ∼60% of that at the start of acquisition according to our simulation based on the two-pool MT model (data not shown), implying that the signal change would be reduced down to 0.3% per ppm at the K-space center. As the signal change associated with the frequency shift is applicable to both the positive and negative MT frequencies, the total frequency deviations (i.e., signal changes) in MTA imaging would be doubled, i.e., 0.6% of signal change per ppm at the K-space center. This indicates that the abovementioned frequency deviation of 0.65 ppm (under the assumption of 0.1% eddy current in the subsequent RF excitation) may induce ∼0.4% signal changes, comparable to those of ALADDIN. The signal changes can further increase or decrease depending on the zeroth order eddy current term (often comparable to the first order term (11)). Note that the frequency deviation of 0.65 ppm is relatively small, and thus we can assume that the MT effects change linearly within the range (12). Even if the linearity assumption is not valid, the error from the nonlinearity will simply increase or decrease the positive and negative frequency offsets in almost the same amount and thus will not significantly contribute to MTA imaging, because the response curve for the longitudinal magnetization versus saturation offset frequency is almost symmetric around the water resonance frequency (Fig. 1b in Ref.12).
PW imaging is calculated as subtraction between ascending/descending acquisitions and thus is expected to be less dependent on the RO eddy current because of no contribution of the RO eddy current to shift along the distance domain (Fig. 5b). In support of this expectation, neither the offset along the RO direction nor the RO gradient polarity changed residual signals on the PW images (data not shown). However, residual signals in PW imaging from the agarose phantom fluctuated as a function of TR with ±0.2% range (data not shown). They are potentially from the SS B0 eddy current that may induce frequency deviations different between the ascending/descending acquisitions (horizontal axis in Fig. 5c). The residual signals in PW imaging could be suppressed at certain TR values, which might be subjected to scan parameters and scanner conditions. Further studies in theory and experiments are necessary to understand sources of residual signals in ALADDIN PW imaging and develop a generic method to suppress them.
Although the above heuristic theory agreed with our experimental results and roughly explained the qualitative eddy current effects, contributions of eddy currents to ALADDIN may be much more complex in reality, as they are affected by various sources including cross terms among gradient axes and various decay constants. These factors may be coupled with the bSSFP acquisition in a complicated way. Therefore, the numbers and the potential theory described above should be treated qualitatively rather than quantitatively. Nonetheless, the experimental results clearly demonstrated the effectiveness of the new averaging scheme for suppression of the artifacts in MTA imaging.
It is not clear why polarity of the artifactual signals varied with small changes in TR (Figs. 2a and 3a). It may be related to the off-resonance response of bSSFP, which is periodic with the periodicity inversely proportional to TR. To our knowledge, the relationship between MT effects in bSSFP and the off-resonance response of bSSFP is underexplored. Considering the periodic off-resonance responses of bSSFP, more extensive studies are necessary to understand the MT effects of bSSFP.
The MT ratio of skeletal muscle has been studied previously (13–18). However, to our knowledge, no study on the MTA of skeletal muscle has been reported. The PSC values of MTA in skeletal muscle were only about 30% of those in brain WM. The ALADDIN PW signals in muscle were even smaller (PSC of 0.07 ± 0.02%) but spatially heterogeneous (data not shown). Further studies are necessary to investigate the clinical usefulness of ALADDIN in skeletal muscle.
The authors thank Dr. Chan-Hong Moon for sharing the agarose phantom and Jeffrey W. Barker for proofreading the manuscript.