Optimization of On-Resonant Magnetization Transfer Contrast in Coronary Vein MRI

Abstract

Magnetization transfer (MT) contrast has been used commonly for endogenous tissue contrast improvements in angiography, brain, body and cardiac imaging. Both off-resonant and on-resonant RF pulses can be used to generate MT based contrast. In this study, on-resonant MT preparation using binomial pulses were optimized and compared to off-resonant MT for imaging of coronary veins. Three parameters were studied with simulations and in vivo measurements: flip angle, pulse repetitions and binomial pulse order. Subsequently, first or second order binomial on-resonant MT pulses with eight repetitions of 720° and 240° flip angle were used for coronary vein MRI. Flip angles of 720° yielded contrast enhancement of 115% (P<0.0006) for first order on-resonant and 95% (P<0.0006) for off-resonant MT. There was no statistically significance difference between off-resonant and on-resonant first order binomial MT at 720°. However, for off-resonance pulses, much more preparation time is needed as compared to the binomials, but with considerably reduced specific absorption rate.

Introduction

Magnetization transfer (MT) has been applied widely to improve contrast in magnetic resonance imaging (MRI) (1–5). Several approaches for MT contrast enhancement have been proposed using either off- or on- resonant RF irradiation (2,6–8). Off-resonant MT is obtained by continuous wave (CW) (9) as well as pulsed irradiation (7,10–12) using RF excitation several kHz away from the water resonance frequency. CW MT yields complete saturation of macromolecular magnetization, i.e. better contrast improvement, but it has higher energy depositions in tissue and requires an additional excitation coil which limits its clinical use (9). The pulsed off-resonant RF irradiation has been applied for contrast enhancement in phantoms (13) and ex vivo (14,15) without a need for additional hardware. The elevated energy deposition from higher flip angle (i.e. higher specific absorption rate (SAR)) limits its in vivo use, especially at high magnetic field strengths and with imaging sequences with intrinsic high SAR. Therefore, several approaches have been proposed to reduce SAR by decreasing the MT flip angle (16,17). Lin et al. (16) proposed power optimized MT by decreasing the flip angle of the MT preparation pulses for outer k-space regions in a segmented acquisition. SAR could also be reduced by using longer pulse duration at the expense of a prolonged TR and imaging time (17). Low flip angle off-resonant MT results in reduced efficacy of MT for contrast enhancement (17,18). On-resonant MT, which consists of a series of block (12) or binomial composite pulses with a net flip angle of 0° (19,20), is an alternative approach to off-resonant RF irradiation (10–12,21).

MT preparation sequences have been shown to enhance visualization of coronary arteries and veins (18,22) by suppressing the myocardial signal while maintaining the blood signal. This is a significant advantage compared to alternative tissue-contrast enhancement methods such as T₂ preparation (18), which results in decreased blood signal as well.

In this study, we investigate an on-resonant MT preparation sequence using optimized binomial pulses as an alternative to off-resonant MT preparation for coronary vein MRI.

Materials and Methods

Figure 1 shows the on-resonant binomial MT preparation pulse, used in this study, which consists of composite RF pulses with 2 to 6 sub-pulse elements depending on the binomial order. The impact of on-resonant MT parameters, i.e. binomial order, flip angle and repetition, was investigated using numerical simulation and an in vivo optimization study. Subsequently, coronary vein MRI with on-resonance MT preparation was performed in a cohort of healthy adult subjects and the results were compared to images acquired with either an off-resonance MT or no preparation.

Figure 1.

Binomial pulses used in this study with the order one to five and a representative nominal flip angle of 720°. The total duration is the same for all pulses however sub-pulse duration is varied.

All imaging studies were performed using 1.5T Philips Achieva System (Philips Healthcare, Best, NL) using 5-element cardiac coil as receiver. All numerical simulations were performed in Matlab (The MathWorks, Natick, MA)

Numerical Simulation

We used a two pool model in which the free proton (Ma) and bound macromolecular spins (Mb) are considered as two separate magnetization groups with an inter-group magnetization transfer rate to study the MT effect (2,7,13). The time evolution of each pool's magnetization was calculated for a single or a train of on-resonance binomial RF pulses. Magnetizations of both bound and free spin pools experiencing MT were investigated considering two endpoints: a) maximal direct effect on the bound pool and b) minimal effect on the free pool. This allows improved contrast while maintaining the imaging signal. Previously reported parameters for myocardium at 37°C (23) were used for simulation (i.e. longitudinal relaxation rate (A free, B bound spin pool) R_A=1.02 1/s, R_B=1 1/s, transfer rate between spin pools R=57 1/s, transverse signal decay T_2A=38.3 ms, T_2B=8.5 µs, magnetization of spin pool M_0A=1 [a.u.] normalized, M_0B=0.072 % (relative to M_0A).

We used the coupled Block equations to simulate the longitudinal and transversal magnetizations of both free and bound spins experiencing an MT pre-pulse. The nominal flip angle, i.e. sum of absolute sub-pulse flip angles ([0°–720°]), the binomial pulse order ([0–5^th order]), and the number of repetitions ([1–8]) were varied to study the impact of these parameters. In our simulation, sub-pulses with constant amplitude (25µT) with no delay between sub-pulse components were used. The sub-pulse durations were varied according to the binomial order. For multiple repetitions of the MT preparation a gap of 1.5 ms between each repetition was used to reflect in vivo imaging sequence.

In Vivo Imaging

Two sets of in vivo studies were performed to investigate the on-resonant MT pulse. An initial study was performed to optimize on-resonant MT parameters in six healthy adult subjects. Subsequently, to demonstrate the efficacy of the on-resonant MT in comparison to off-resonant pulses, coronary vein MRI were acquired in seven healthy adult subjects using the optimized parameters. Written informed consent was obtained from each subject and the protocols were approved by the committee on clinical investigation.

In vivo Optimization

The impact of various MT pulse parameters were further studied in vivo by acquiring ECG triggered, 2D single-slice, single phase two chamber view images with a spoiled gradient echo imaging sequence (FOV=250×250 mm², spatial resolution 1.5×1.5 mm², slice thickness 8 mm, TR/TE/α=5.8ms/2.8ms/30° and 20 phase-encode lines per cardiac cycle, respiratory navigator gating, and spectrally-selective fat saturation) preceded by MT preparation sequence. On each subject, 80 images were acquired with different combination of MT preparations in a single imaging session which consisted of: a) off-resonant MT (20 ms Gaussian weighted sinc pulse with an offset of 500 Hz, a time-bandwidth-product of 6) with 20 combinations of flip angle (60°, 120°, 240°, 480° and 720°) and number of repetitions (2–8 in steps of 2), b) on-resonant MT with 60 combinations of binomial order (1^st to 3^rd order), number of repetitions (2–8 in steps of 2), and nominal flip angle (60°, 120°, 240°, 480°, and 720°). Each binomial pulse had duration of 0.3 ms for 60° and 2.0 ms for 720° and consisted of composite block pulses with maximum amplitude (25µT) with no gaps between sub-pulses. The net flip angle of all on-resonant binomial pulses was 0°. A baseline image without any tissue contrast enhancement was also acquired on each subject for comparison. SAR for each preparation sequence, calculated by the scanner software, was also recorded.

Coronary Vein MRI

Coronary vein MRI was acquired with the selected on-resonant and off-resonant MT preparation sequence. All images were obtained during the systolic rest period with free-breathing using respiratory navigator (18). A 3D GRE sequence with the following imaging parameters was used: TR/TE/α= 6ms/1.87ms/30°, FOV 270×270×30 mm³ with a spatial resolution of 1×1×3 mm³ reconstructed to 0.5×0.5×1.5 mm³, 20 slice, with 15 phase-encode lines per heart-beat.

Four coronary vein MRI scans were performed: 1) a baseline without any preparation pulse, 2) off-resonant MT, and 3–4) two on-resonant MT preparations. The off-resonant pulses had a frequency offset of 500 Hz, a flip angle of 720°, duration of 20 ms for each pulse and 8 pulse repetitions. Based on the results from in vivo study (discussed in result), the two on-resonant MT sequences were a) first order pulse with a nominal flip angle of 720° and 8 repetitions and b) second order binomial pulse with nominal flip angle of 240° and 8 repetitions. For the first order pulse, duration/flip angle/amplitude of 1.02ms/360°/25µT and 1.02ms/−360°/25µT was used for sub-pulse 1 and 2 respectively. For second order binomial RF pulse, duration/flip angle/amplitude was 0.17ms/60°/25µT, 0.35ms/−120°/25µT, 0.17ms/60°/25µT for sub-pulses 1, 2 and 3 respectively.

Data Analysis

SNR was defined as mean signal in the region of interest divided by standard deviation of the noise in the air across the chest wall, and CNR was defined as signal difference divided by noise. Signal measurements were performed in the myocardium and the left ventricular (LV) blood pool for the in vivo parameter optimization and in the lumen of the coronary sinus for the coronary vein MRI. Relative contrast (rCNR) and signal to noise ratios (rSNR) as a function of each parameter were calculated by linear regression. rSNR and rCNR is defined as the slope of the regression analysis.

Since multiple rSNR and rCNR was calculated for each of the subjects, the within-subject correlation of rSNR and rCNR was taken into account in a linear mixed-effects model (24,25). rSNR and rCNR was considered as the dependent variables in two different models and the MT preparation method (e.g. pulse types) as the independent 4-level categorical variable, representing the 4 binary dummy variables. The within-subject correlation was modeled by using the compound-symmetry structure for the variance-covariance matrix of error term. Once the fixed effect estimates for the dummy variables were obtained. Six pair-wise comparisons of the four methods were performed. The statistical analysis was corrected for multiple testing. A P-value < 0.05 was considered statistically significant. The SAS/STAT software (SAS Institute Inc. Cary NC) was used.

Results

Numerical Simulation

Figure 2 shows the longitudinal magnetization of a) free and b) bound spin pool experiencing a single binomial on-resonance MT pulse with different order and nominal flip angles. Saturation of the bound spin pool of more than 50% occurred at flip angles of 613°, 657°, 708° for first to third order respectively. Free magnetization was saturated less than 3% over the entire range of flip angles and binomial orders.

Figure 2.

Numerical simulation of the longitudinal magnetization (M_z) of (A) the free and (B) the bound spins experiencing a single binomial pulse of different orders (1^st to 5^th) as a function of nominal flip angle applied.

Figure 3 shows the longitudinal magnetization for different pulse repetitions using a nominal flip angle of 120°, 240°, 480° and 720° for binomial pulse orders of 1 to 5. The cumulative effect of multiple pulse application on the saturation can be seen in both spin pools, as previously shown (26). Direct saturation was decreased with higher order pulses. Saturation of the bound spin pool by 80% using on-resonant MT preparation can be reached by applying a nominal flip angle of 720° and 3 repetitions of first and second order pulses, 4 repetitions of third order, 5 repetitions of fourth order and 6 repetitions of fifth order pulses. 90% saturation could only be reached with 720° and 10 repetitions of first and second order pulses. Between 240° and 480° second and third order pulses yielded a good trade-off between saturating the free and the bound spin pool.

Figure 3.

Numerical simulation of the free (A) and bound (B) longitudinal magnetization experiencing a train of 1^st to 5^th order binomial pulse each with a nominal flip angle of 120°, 240°, 480° and 720°.

In Vivo Imaging

In vivo optimization

Figure 4 (A) shows the rCNR for different nominal flip angles using eight pulse repetitions. First order on-resonant and off-resonant MT only showed contrast enhancement for flip angles greater than 480°. In the range of flip angles between 240° to 480° second and third order pulses outperformed first order and off-resonant pulses with contrast enhancement by 39% (2^nd order, α=240°) and 35% (3^rd order, α=240°). Pair wise analysis of pulses at 240° resulted in a P-value of 0.017 for the difference between off-resonant and second order on-resonant preparation and 0.053 for the difference between off-resonant and third order on-resonant preparation. At the maximal flip angle of 720° the difference between first order on-resonant and off-resonant preparation was significant (P = 0.04). Figure 4 (B) shows the rCNR of data obtained using different pulse repetitions and a nominal flip angle of 720°. Increasing the pulse repetitions leads to an increased contrast. Going from two pulse repetitions to eight, rCNR was increased by additional 35% for first order, 21% for second order, 22% for third order and 13% for off-resonant MT.

Figure 4.

The in vivo rCNR for different flip angles (A) and pulse repetitions (B) of the used MT-preparation pulses. At low flip angles (≤480°) 2^nd and 3^rd order pulses outperform first order and off-resonant MT preparation.

The rSNR measurements in LV blood showed that off-resonant MT creates the most pronounced signal drop (20%, eight repetitions and 720°). This difference was found to be significant compared to all on-resonant MT preparation (P=0.001 for 1^st order, P=0.023 for 2^nd order, P=0.001 for 3^rd order). First order on-resonant MT showed the rSNR drop of 7% for flip angles of 720° and eight repetitions. The second and third order binomial pulses did not show any considerable rSNR drop.

The calculated energy deposition of single on-resonant MT preparation pulses (1^st to 3^rd order) was 16.7, 34.5, 69.1, 139.5 and 209.9 mJ/Kg and 0.8, 3.1, 12.6, 50.2 and 113.1 mJ/Kg for off-resonant at flip angles of 60°, 120°, 240°, 480° and 720° respectively.

Coronary vein MRI

Figure 5 shows example slice from the 3D coronary vein MRI acquired with (A) no preparation pulse (B) a train of eight 2^nd order binomial MT with nominal flip angle of 240° (C) a train of eight 1^st order binomial MT with nominal flip angle of 720° and (D) eight off-resonant MT preparation pulses with a nominal flip of 720° each. The mean contrast improvement for three MT preparations in coronary vein MRI for all subjects is summarized in Table 1. First order on-resonant MT preparation yielded the highest contrast between venous blood and myocardium (rCNR: 2.15±0.15, P<0.0006) followed by off-resonant (rCNR: 1.95±0.22, P<0.0006). The 2^nd order pulse with 240° resulted in a minimal but statistically non significant contrast enhancement by a factor of 1.36±0.16 (P= N.S.). Differences between first order on-resonant and off-resonant pulses were not significant

Figure 5.

A single slice from a 3D dataset of coronary vein MRI showing coronary sinus (CS) acquired with (A) no preparation pulse (B) a train of eight 2^nd order 240° binomial (C) eight 1^st order 720° binomial pulses and (D) eight off-resonant MT preparation pulses with a nominal flip of 720° each. The highest contrast enhancement can be obtained using large flip angle pulses for both on-resonance and off-resonance pulses. (LV= Left ventricle, RV= right ventricle, CS=coronary sinus, MYO=myocardium)

Table 1.

The blood-myocardium rCNR and blood rSNR for coronary vein MRI with the selected MT preparation parameters.

MT Pulse type	MT Pulse Parameters a	rCNR [a.u.]	rSNR [a.u.]	SAR (mJ/Kg)	P-value b CNR/rSNR
500 Hz Off-resonant	720° / 8 / 160	1.95±0.22	0.96±0.06	904.4	<0.0006 / N.S
1st order binomial	720° / 8 / 27	2.15±0.15	1.11±0.05	1679.1	<0.0006 / N.S
2nd order binomial	240° / 8 / 16	1.36±0.16	0.95±0.05	522.8	N.S / N.S

^aMT pulse parameters are nominal flip angle/pulse repetitions/ total duration of preparation sequence (ms),

^bP-value is for the pair-wise comparison with non-MT acquisition)

Discussion

We have performed numerical and in vivo studies to investigate utility of on-resonant MT for contrast enhancement in coronary vein MRI. The first order on-resonant MT with flip angle of 720° with eight repetitions provides similar contrast to off-resonance MT preparation and allows visualization of the coronary veins. For off-resonance pulses, much more preparation time is needed (20 ms) as compared to the binomials (2 ms), but with considerably reduced SAR.

Numerical simulations showed that for all orders the applied flip angle is a more important sequence tuning parameter considering the saturation of the bound magnetization, compared to pulse repetitions. It can be seen that multiple repetitions of low flip angle pulses do not saturate the bound magnetization as complete as a single application of a pulse with the same total flip angle.

The binomial pulses were not optimized for T₂ selectivity as proposed by Mathilde Pachot-Clouard and Luc Darrasse (27). The transversal magnetization of the restricted spin pool decays during pulse application, since the pulse duration is one order of magnitude longer than T₂ of bound spins, for the shortest MT preparation (α=60°).

Besides the three parameters (nominal flip angle, number of repetitions and binomial order) investigated in our study, the performance of on-resonant pulses is also dependent on off-resonance frequency. Our in vivo MT optimization study showed a significant rSNR drop for first order pulses with flip angles of 240°– 480°. This can be attributed to the high sensitivity of first order MT to off-resonance. The second and third order MT is more robust to field inhomogeneity. These observations were confirmed with a numerical simulation (data not shown).

In this study, the optimization of on-resonant MT pulses was performed to improve contrast between coronary veins (or arteries) and myocardium. The optimization result and approach can be generalized to various other MT applications for contrast improvements in other organs, but needs to be further investigated for each application.

Our study has limitations. We have only compared on-resonant to off-resonant MT for coronary vein MRI. However, exogenous contrasts such as Vasovist (Bayer Schering Pharma, Berlin, Germany) or Gadobenate dimeglumine (MultiHance, Bracco Imaging SpA, Milan, Italy) could also be used for coronary vein imaging with additional cost and risk of nephrogenic systemic fibrosis for patients with impaired renal function. Despite the fact that arterial and venous blood have different T₂ but similar MT (4), CNR/SNR optimization was based on the LV arterial blood during the transient state, therefore signal behavior for venous and arterial blood could be different. We did not present any clinical validation data to compare the MR to X-ray coronary venogram, which is the clinical gold-standard imaging modality for coronary veins in cardiac electrophysiology.

Conclusions

We present an optimized on-resonant MT preparation sequence as an alternative to off-resonant MT for coronary vein MRI. Both MT sequences yield similar contrast between blood and myocardium, however for off-resonance pulses, much more preparation time is needed but with considerably reduced specific absorption rate.