Enhanced Image Quality in Black-Blood MRI by Using the Improved Motion-Sensitized Driven-Equilibrium (iMSDE) Sequence

Abstract

Purpose

To propose an improved motion-sensitized driven-equilibrium (iMSDE) pulse sequence to enhance the tissue signal-to-noise ratio (SNR) while maintaining the same flow suppression capability in black-blood carotid artery MRI.

Materials and Methods

Compared to the traditional MSDE sequence, the iMSDE sequence uses an extra refocusing pulse and two extra gradients to achieve SNR improvement. Computer simulation and phantom studies were used to evaluate both eddy currents and local B₁ inhomogeneity effects on SNR behaviors on both MSDE and iMSDE images. To further assess the SNR improvements brought by iMSDE in vivo, 5 healthy volunteers were also scanned with both sequences. Paired t-test was used for statistical comparison.

Results

Both simulations and phantom studies demonstrated that eddy currents and local B₁ inhomogeneity will cause image SNR reduction in the MSDE sequence, and that these factors can be partially compensated for with the iMSDE sequence. In vivo comparison showed that the iMSDE sequence significantly improved the tissue-lumen contrast-to-noise ratio (CNR) and static tissue SNR (p<0.001 for both), while maintaining low lumen SNR in carotid MRI.

Conclusion

Compared to the traditional MSDE sequence, the iMSDE sequence can achieve improved soft tissue SNR and CNR in carotid artery MRI without sacrificing flow suppression capability and time efficiency.

INTRODUCTION

Efficient flowing blood signal suppression is critical for accurate morphology measurements and diagnosis in MR vessel wall imaging (1–3). Due to the complicated flow patterns in the carotid artery bifurcation, however, current black-blood (BB) imaging techniques are frequently compromised by plaque-mimicking artifacts (4). Widely used BB imaging techniques such as in-flow saturation (IS) (5) and double inversion recovery (DIR) based (6–9) techniques are limited by the blood replenishing rate and therefore prone to flow artifacts in the presence of recirculating or stagnant flow.

To improve blood suppression, flow-dephasing BB imaging techniques, such as the motion-sensitization driven equilibrium (MSDE) sequence, have recently been utilized for carotid artery vessel wall imaging (10–12). As discussed in previous works (12, 13), the flow suppression capability of the MSDE sequence is not limited by the flow replenishing rate, but rather determined by the first-order moment (m₁) of the flow sensitizing gradient pair as well as the intravoxel spin velocity distribution. Therefore, unlike IS and DIR techniques, the MSDE technique can potentially eliminate any slow flowing blood artifacts, as long as the m₁ is sufficiently high.

A practical limitation of the MSDE technique is the signal loss (12), and the suspected sources for signal drop include the inherent T₂ decay, the diffusion effect, and instrumental factors. Both the T₂ decay and diffusion attenuation are determined by the intrinsic tissue properties and are therefore difficult to avoid. Among the instrumental factors that potentially may affect MSDE performance are eddy currents and B₁ inhomogeneities.

In this study, we present an improved MSDE (iMSDE) sequence that can partially compensate for the above instrumental factors and therefore considerably improve image quality in black-blood MRI. The novel design was tested through numerical simulations, phantom studies, and in vivo experiments.

MATERIALS AND METHODS

iMSDE Sequence

Compared to the traditional MSDE sequence (Fig. 1(a)), the major difference in the iMSDE sequence is the extra 180° refocusing pulse (Fig. 1(b)). The group of 4 radiofrequency (RF) pulses is constructed in an MLEV-4 scheme (14), and both 180° pulses are composite 90x-180y-90x pulses. If the total duration between the two 90° pulses is defined as TE_prep, the delays between RF pulses are TE_prep/4, TE_prep/2, TE_prep/4, respectively. In the reported variant of the sequence, all RF pulses are rectangular, and the duration of a 90 degree pulse is 0.4 ms.

Fig. 1.

Diagram of both the traditional (a) and improved (b) MSDE (iMSDE) preparative pulse sequences. All open trapezoids represent motion sensitizing gradients. The spoiling gradients are marked by ‘S’. Both sequences are executed before a standard Turbo Spin-Echo (TSE) acquisition sequence. Diagrams showing how m₁ changes as a function of scanning parameters are also shown for MSDE (c) and iMSDE (d) sequences. In both diagrams, the gradient strengths are assumed to be 20 mT/m.

In the iMSDE sequence, the motion-sensitized gradients were constructed to: 1) keep the zeroth order gradient moment (m₀) as zero, such that the phase coherence among stationary spins can be maintained; and 2) maximize the first order gradient moment (m₁) for a given TE_prep, such that the phase coherence among flowing spins can be eliminated effectively (12). For both MSDE and iMSDE sequences, m₁ values are also plotted as functions of sequence parameters (Fig. 1(c, d)).

Simulations for the Eddy Currents Effect

Eddy currents are known to be a critical source of artifacts in diffusion-weighted imaging (DWI) (15). A double-refocusing scheme conceptually similar to iMSDE (16) was shown to be more immune to eddy currents than the standard spin-echo DWI sequence. Despite this similarity, however, the mechanism of image quality degradation in MSDE and DWI is different. In DWI, eddy currents mostly affect phase properties of the signal during a readout sequence (typically echo-planar) and, therefore, cause image distortions like shear, scaling and translation (17). In the MSDE/iMSDE sequences, the m₀ of the motion sensitizing gradients needs to be fully balanced to avoid signal loss caused by spin dephasing. If a gradient waveform, however, is distorted by the eddy currents effect, the m₀ will not be perfectly balanced, thus resulting in the overall reduction of the observed signal on top of the inherent signal drop caused by T₂ relaxation or diffusion effects.

To estimate the effect of eddy currents on the magnetization, the residual uncompensated gradient area (m₀) was calculated over the duration of a preparative sequence by integrating a gradient waveform with shape distortions caused by eddy currents. Integration limits were from the first to the last 90° pulse for both sequences. The resulting signal attenuation was calculated by integrating a phase distribution produced by the uncompensated gradient across the voxel size, which was fixed to 2 mm based on the typical slice thickness in carotid imaging. Simulations were performed for a series of m₁ values assuming typical sequence parameters relevant to clinical MSDE applications (12). The same m₁ were used for MSDE and iMSDE sequences, which presumably correspond to the similar flow suppression efficiency and enables fair comparison. Sequence parameters for simulations are summarized in Table 1. The distortions of a gradient waveform caused by eddy current were modeled as a superposition of exponential decay terms with corresponding time constants τ_i and weighting factors w_i(18). For example, the eddy current for a single gradient slope can be described as (18),

Table 1.

Parameters for MSDE and iMSDE at different m₁ levels

m1 (mTms2/m)	Sequence	Gradient Strength (mT/m)	Motion sensitizing gradient duration a (ms)	Total duration of prepulse b (ms)
185	MSDE	20	1.9	7.1
	iMSDE	20	0.7	10
500	MSDE	20	3.8	10.9
	iMSDE	20	1.6	13.6
950	MSDE	20	5.6	14.6
	iMSDE	20	2.5	17.2
1545	MSDE	20	7.5	18.4
	iMSDE	20	3.4	20.9
2280	MSDE	20	9.4	22.1
	iMSDE	20	4.4	24.7
3160	MSDE	20	11.3	25.9
	iMSDE	20	5.3	28.3

^aThe duration of one gradient lobe

^bThe time duration from the start of the first 90° RF pulse to the end of the last 90° pulse.

$$G_{\mathit{eddy}}=-G_0\sum _{i=1}^N{w}_{i}exp((t_0-t)/{\tau }_i),$$ [Eq.1]

where the negative sign indicates the polarity of G_eddy (relative to the original gradient), G₀ is the initial eddy current amplitude, N is the number of time constants, t is the time from the beginning of the sequence, t₀ is the start time of the gradient slope (attack or decay). The number of time constants was set to 4, and the values were chosen based on existing literature (18): τ_ι=1.36, 16.8, 93 and 492 ms. Similarly, weightings used in simulations were 0.43, 0.13, 0.19, 0.25, according to values reported previously (18).

The signal decay was simulated assuming that spins across each dimension evenly dephased by the unbalanced gradient area (19). The dephased spin vectors were then summed, and the ratio of the resulting vector to the equilibrium magnetization was used to estimate an expected signal loss. The simulations were conducted for each time constant separately. To calculate the final signal level after combining all time constants, the phase dispersion from each time constants were first summed and then the total phase dispersion is used to calculate the overall signal loss.

Simulations for the B₁ Inhomogeneity Effect

To evaluate the magnetization decay caused by inhomogenous B₁ field in MSDE and iMSDE sequences, Bloch equation based simulations were performed to compute the residual magnetization after each prepulse. All RF pulses were approximated as instantaneous due to their originally short duration. The simulations were carried out in a custom-coded MATLAB (Mathworks, Natick, MA) program.

The effect of B₁ field inhomogeneity was evaluated as a function of a relative B₁ scaling factor (rB₁). Other parameters used in this simulation were T₁, T₂ and TE_prep times. A T₁ value of 600ms and a T₂ value of 400ms were used to reflect a realistic prediction for the subsequent phantom experiment. Representative TE_prep time used in this simulation is the same as used in eddy currents simulation, as listed in Table 1 for m₁=950 mTms²/m.

Phantom Study

MR images of a CuSO₄ water solution (770mg CuSO₄/L) were acquired on a clinical 3T scanner (Philips Achieva R2.5.3, Best, the Netherlands). The maximum available gradient strength and slew rate are 33mT/m and 200mT/m/ms, respectively. A proton density weighted MR image without any prepulse was first acquired as a baseline. To evaluate the signal reduction at different m₁ conditions for both techniques, identical imaging parameters to the previous computer simulations were used, as summarized in Table 1.

Other than the imaging parameters listed in Table 1, the rest of the parameters were exactly the same for both MSDE and iMSDE images, as well as the baseline reference image. The parameters were: Turbo Spin Echo (TSE) sequence, TR/TE 3000/8.5 ms, FOV 160×120mm, in-plane resolution 1×1mm, slice thickness 10mm, echo trail length (ETL) 12, number of signal averages (NSA) 1. The coil used for this phantom experiment was an 8-channel head coil for its relatively larger coverage and improved SNR compare to the body coil.

Relative SNR is calculated by dividing the phantom SNR at MSDE/iMSDE images by the phantom SNR acquired without prepulse. SNR here is defined as ratio of average phantom signal divided by noise level measured in background air, after incorporating a coil correction factor (20). All image analyses were conducted using a custom-designed image analysis software CASCADE (21).

In Vivo Comparison

The study was performed in compliance with the guidelines of the local Institutional Review Board. Five healthy volunteers with no known cardiovascular disease (2 Male 3 Female, mean age: 56) were recruited for this study after obtaining their informed consents.

MR images of the carotid arteries were obtained from the same 3T clinical scanner with a custom-designed four-channel phased-array bilateral carotid coil. The local Institutional Review Board has approved the coil for human research.

To compare the blood suppression efficiency of both techniques, transverse images centered on the carotid bifurcation were obtained from all volunteers at identical anatomic locations with both MSDE and iMSDE sequences. To ensure a fair comparison, both techniques in the same group utilized the same gradient strength and same m₁. Two groups of m₁ values, representing regular (m₁=950 mTms²/m) and high m₁ (m₁=1545 mTms²/m) conditions were used. The regular m₁ value represents m₁ levels comparable to those used in clinical scans (12) and the high m₁ value represents the highest m₁ values suitable for current clinical carotid scan settings without causing dramatic image quality degradation for both preparative sequences. Detailed parameters for two m₁ levels used here were again the same as used in previous computer simulations and phantom studies, as summarized in Table 1. All sequences above used the same proton density (PD) weighted TSE acquisition sequence: TR/TE 4000/8.5ms, FOV 160×120mm, matrix 256×192, slice thickness 2mm, ETL12, NSA 1, 16 slices, total scan time 2min 16sec.

For the in vivo comparison study, the SNR in the carotid artery lumen was used as a measure of flow suppression efficiency, and the SNR of the sternocleidomastoid (SM) muscle was used as a measure of the overall signal intensity because it is relatively homogenous and close to the carotid artery (12). Images acquired at the same location with different sequences were manually registered using the image processing software, CASCADE (21). SNR measurements were performed as described previously (12).

Analysis was performed on the 6 images centered at the carotid bifurcation of each artery because more distal artery segments are typically not prone to plaque-mimicking artifacts. For each subject, bilateral arteries were included in the analysis.

Statistical analysis was performed using Microsoft Excel (Microsoft Corporation, Redmond, WA, USA). Two tailed paired Student’s t-tests were conducted for both lumen and SM muscle SNR comparison in all groups. In all tests, statistical significance was defined at the p<0.05 level.

RESULTS

Simulations for the Eddy Current Effect

The simulated signal decrease caused by unbalanced m₀ moment at different time constants is plotted in figure 2. It can be seen that the iMSDE sequence retains much higher signal level at all τ values, except for the shortest time constant (τ=1.36ms). Nevertheless, as can be seen in the plot calculated for the weighted sum of all decay components (Fig. 2e), the iMSDE sequence provides much better immunity to the total eddy current effect than the MSDE sequence.

Fig. 2.

Simulated signal loss in MSDE and iMSDE sequences caused by eddy currents. Panels (a–d) compare the uncompensated m₀ between the two techniques for different components of eddy currents. Panel (e) compares the uncompensated m₀ after combining all exponential components with corresponding weightings. The m₁ range corresponding to typical clinical carotid imaging conditions (12) is marked by the double line box on panel (e). It can be seen that iMSDE sequence enables a much higher signal at all m₁ levels.

Simulations for the B₁ Inhomogeneity Effect

Representative plot comparing signal behavior for two sequences at m₁=950 mTms²/m is shown in Fig. 3. The simulations revealed that the iMSDE sequence can consistently improve signal levels when the B₁ field is inhomogenous. As seen in Fig. 3, the iMSDE sequence presents almost invariant signal levels at different rB₁ conditions, while the MSDE sequences is characterized by a progressive signal loss with a deviation of rB₁ from unit.

Fig. 3.

Comparison of simulated signals as functions of relative B₁ inhomogeneity between MSDE and iMSDE sequences. The iMSDE sequence provides consistently higher signal intensity compared to the MSDE sequence, and the signal improvement becomes more pronounced when rB₁ value drifts further away from unit. Data are calculated for m₁=950 mTms²/m.

Phantom Study

In the phantom study, trends similar to what were predicted by the computer simulations were observed. The iMSDE sequence provided consistently better SNR when compared to the MSDE sequence. As shown in Fig. 4, although an apparent non-uniform signal loss can be identified on MSDE image, especially when m₁ is high, no visible image quality degradation is seen on matched iMSDE image.

Fig. 4.

Example images obtained in phantom experiments. Top row: images acquired with MSDE sequences at m₁ levels of 185, 950, 1545 and 3160 mTms²/m. Bottom row: images acquired with iMSDE sequences at matched m₁ levels. The iMSDE sequence provides visible improvement, of image quality, especially for high m₁.

Quantitative SNR comparison (Fig. 5) demonstrates that iMSDE provides significant signal improvement over MSDE. As indicated by the double lined box (Fig. 5), the iMSDE sequence enables a 20–60% higher signal for typical m₁ values used in clinical carotid artery imaging.

Fig. 5.

Measured relative SNR values in the phantom for MSDE and iMSDE sequences at different m₁ values. iMSDE provided significantly higher signal levels for all conditions. The m₁ values that are proper for clinical carotid scans are highlighted in the double-lined box. In this range, iMSDE has 20–60% higher SNR.

In Vivo Comparison

For in vivo imaging, iMSDE images presented consistently higher signal intensities compared to MSDE for all m₁ groups acquired at the same location. The signal improvement of iMSDE sequence becomes more visually appreciable at higher m₁ levels, as seen in Fig. 6. The vessel wall on MSDE image is totally dimmed when m₁ is high (panel b), but well visible on the corresponding iMSDE image at the same m₁ (panel d). Another noteworthy observation is that the flow artifacts (arrows) on the images from one subject were visible on both regular m₁ images (Fig. 6(a), (c)) but removed on both high m₁ images (Fig. 6(b), (d)). Furthermore, the size and visibility of flow artifacts were comparable for MSDE and iMSDE images acquired at the same m₁ conditions.

Fig. 6.

Example image comparison between MSDE and iMSDE techniques. Panels (a, b) MSDE images acquired at regular and high m₁ conditions; Panels (c, d) iMSDE images acquired at regular and high m₁ conditions. iMSDE images generally present a higher signal level than the corresponding MSDE images, especially when the m₁ is high. The flow artifact (arrows) that is visible on the regular m₁ images (a, c) was completely removed on high m₁ images (b, d).

Eight out of the 60 total locations were excluded because of poor image quality, which was primarily caused by the combination of both the severe signal loss on high m₁ MSDE images and low coil sensitivity for peripheral locations. Quantitative results of image analysis are summarized in Table 2. iMSDE provided significantly higher CNR and sternocleidomastoid muscle SNR at both regular and high m₁ values when compared with the standard MSDE sequence (p<0.001, Table 2), Similar lumen SNR values were found between MSDE and iMSDE images at regular m₁ (p=0.58 and 0.001, respectively), and significantly lower lumen SNR were identified for iMSDE at high m₁.

Table 2.

In vivo SNR and CNR measurement comparison

			SNR (Mean ± SD)	p-value
SM – Lumen CNR (N=52)	Regular m1	MSDE	14.6 ± 7.3	<0.001
		iMSDE	17.3 ± 8.3
	High m1	MSDE	8.1 ± 4.2	<0.001
		iMSDE	13.7 ± 6.9
SM muscle SNR (N=52)	Regular m1	MSDE	17.9 ± 8.2	<0.001
		iMSDE	20.6 ± 9.3
	High m1	MSDE	11.2 ± 4.5	<0.001
		iMSDE	16.4 ± 7.4
Lumen SNR (N=52)	Regular m1	MSDE	3.3 ± 1.3	0.58
		iMSDE	3.4 ± 1.4
	High m1	MSDE	3.1 ± 1.0	0.001
		iMSDE	2.7 ± 0.9

DISCUSSION

The signal loss in the MSDE sequence is associated with multiple factors. Although the actual contribution from eddy currents and B₁ inhomogeneity is still unknown due to the difficulty of controlling each individual factor in an actual scanning environment, iMSDE sequence is still shown to present consistently higher SNR (over 15%) in both phantom and in vivo studies. The SNR increase can greatly improve the robustness of cardiovascular vessel wall scans which have been limited by insufficient resolution and/or inadequate SNR. It is noteworthy that the signal improvement was achieved at almost no cost: the iMSDE sequence has the same flow suppression efficiency and similar time efficiency as the MSDE sequence. A potential disadvantage of iMSDE sequence is the employment of an additional refocusing 180° pulse, which may make the sequence reach specific absorption rate (SAR) limits faster. The SAR limitation can be reached even faster for higher magnetic field applications. In our current in vivo carotid imaging protocol at 3T, however, this extra pulse did not limit the total number of slices available in each TR (8 slices for every TR=4000ms).

Other than eddy currents and B₁ inhomogeneity, the off-resonance RF excitation caused by B₀ field inhomogeneity could potentially lead to signal drop in MSDE images. However, any frequency shifts of less than several hundred hertz will not cause a significant off-resonance effect for non-selective hard pulses used in the sequence design (0.4 ms duration, 2500 Hz bandwidth). For the particular carotid imaging setup investigated in this study, B₀ nonuniformity does not exceed ±60Hz across the FOV, as confirmed by B₀ mapping (data not shown). Therefore, B₀ inhomogeneity plays a negligible role in carotid imaging using MSDE or iMSDE compared to other factors (eddy currents and B₁ inhomogeneity). However, for potential future MSDE/iMSDE applications, such as coronary or peripheral artery imaging, this factor may be considerable due to larger B₀ non-uniformity in corresponding anatomical areas.

The signal drop measured in the phantom study is caused by both the eddy currents and B₁ inhomogeneity issues. The actual signal loss measured on the phantom is not as significant as the combined results from simulations. This discrepancy could be attributed to the eddy current compensation provided by the scanner. Although the simulation is made by assuming no eddy current compensation is present, most modern commercial scanners compensate for eddy currents automatically. As a result, the signal loss observed from the experiment is considerably lower than that predicted in simulations.

In the simulations of the eddy current effect, the numbers and values of the time constants were obtained from a different type of imaging system since this is the only data we obtained from the previous literature. This fact, however, does not undermine the validity of this simulation. The robustness of iMSDE sequences to eddy currents is still well demonstrated since consistently better signal level were found except for the shortest time constant simulation when compared to the MSDE. We would expect to see similar results when other time constants were used for the simulation.

Although the flow suppression capability of the iMSDE sequence improved along with the m₁ increase, the overall carotid artery CNR between SM muscle and CA lumen dropped (Table 2). Generally, a CNR decrease indicates a less preferable situation for image analysis. In vessel wall imaging, however, a solely decreased CNR may not always indicate a less favorable situation, but should be evaluated in combination with the effectiveness of flow artifact elimination. As revealed in Fig. 6, the high m₁ images generally had a lower CNR than the regular m₁ images, but suppression of flow artifacts was consistently better at high m₁. A good application for the high m₁ iMSDE imaging, therefore, is to be used as a lumen/wall boundary identification sequence in carotid artery imaging protocols. It can also be combined with an isotropic 3D sequence (such as TFE) for accurate plaque burden assessment. In this way, the lumen/wall boundary can be reliably delineated through the high m₁ iMSDE sequence and the plaque tissue segmentation (1, 3) can be performed using other multi-contrast imaging sequences.

The lumen SNRs observed at high m₁ in vivo experiments were statistically different between MSDE and iMSDE groups. This was not considered as being caused by the inherent different blood suppression capabilities between the two sequences. Rather, the reason was that some image distortions caused by eddy current could be identified on the MSDE images, such that the lumen signal was sometimes contaminated by the artifacts originated from the wall, which resulted in a slightly higher signal.

Some other potential applications of the iMSDE pulse sequence include peripheral artery imaging and high-field imaging applications. In peripheral artery imaging, the flow velocity is much lower than in major arteries such as the carotid. Therefore, a sequence with both high blood flow suppression capability and good signal level, such as the iMSDE sequence, is desired. For high-field imaging applications, increased field strength may be a more challenging environment to achieve a homogeneous B₁ field. In this situation, the iMSDE sequence, which is less sensitive to B₁ variations, may be a more practical solution.

In conclusion, an iMSDE sequence was proposed for black-blood vessel wall imaging. Simulations and phantom studies validated the hypothesis that eddy current and local B₁ inhomogeneities cause considerable signal loss in the standard MSDE sequence, while both adverse factors can be effectively compensated using the iMSDE sequence. The in vivo iMSDE images confirmed the simulation and phantom findings and demonstrated significant CNR improvements compared to the traditional MSDE technique.