Modified Pulsed-continuous arterial spin labeling for labeling of a single artery

Abstract

Imaging the contribution of different arterial vessels to the blood supply of the brain can potentially guide the treatment of vascular disease and other disorders. Previously available only with catheter angiography, vessel selective labeling of arteries has now been demonstrated with pulsed and continuous arterial spin labeling (ASL) methods. Pulsed continuous labeling, which permits continuous labeling on standard scanner RF hardware, has been used to encode the contribution of different vessels to the blood supply of the brain. Vessel encoding requires a longer scan, a more complex reconstruction algorithm, and may be more sensitive to fluctuations in flow, however. Here a method is presented for single artery selective labeling in which a disc around the targeted vessel is labeled. Based on pulsed-continuous labeling, this method is achieved by rotating the directions of added in-plane gradients. Numerical simulations of the simplest strategy show good efficiency but poor suppression of labeling at large distances from the target vessel. Amplitude modulation of the rotating in-plane gradients results in better suppression of distant vessels. In-vivo results demonstrate highly selective labeling of individual vessels and a rapid falloff of the labeling with distance from the center of the labeling disc, in agreement with the simulations.

INTRODUCTION

Arterial spin labeling (ASL) (2,3) employs spatially selective RF pulses to create noninvasive perfusion or angiographic images. Conventional ASL uses RF pulses to label the endogenous blood water protons from all feeding blood vessels. However, control over which arteries are labeled can be used to measure the tissue regions that are perfused by particular vessels (4–8) and to characterize the dynamics of flow through vessels, occlusions (9), arteriovenous malformations (10), and aneurysms (11). Related information obtained with catheter angiography is already widely used for clinical management (12–14), but catherization poses much greater risk than MRI (15).

Arterial selective labeling with ASL requires creating an image where only blood flowing through the artery of interest is labeled while the other arteries are unlabeled. Multiple approaches to achieving such selective labeling have been proposed, as described in a recent review (16). The use of selective RF pulses for pulsed labeling can achieve hemispheric selective or even individual vessel selective labeling (6,8,16). Continuous ASL methods can provide higher perfusion signal and signal-to-noise ratio and may provide simpler and more precise labeling of individual vessels. A separate RF coil to label blood can be used for vessel selective continuous labeling (17–19), but this requires special hardware and a substantial distance from labeling plane to the imaging region. As an alternative, magnetic field gradients can be used for localization of labeling with continuous ASL. A continuous ASL method using an oblique labeling plane combined with a head transmit receive RF coil has been reported (7,20). This method is not compatible with large volume transmission and array coil reception, however. Werner et al. (21) introduced a modification of the amplitude modulated control method (22) for multislice continuous ASL in which the labeling plane is tilted and rotated such that only blood flow in the neighborhood of a single point could be efficiently labeled. Though the tilting of the labeling plane requires a greater distance between the labeling location and the imaging region than for non-selective ASL, this approach was highly successful for selective labeling.

The introduction of pulsed continuous ASL (pCASL) (23,24) has made continuous labeling more practical on clinical scanners and also increased labeling efficiency relative to the amplitude modulated control technique. Since pCASL introduced gaps between RF pulses, additional gradients can be added to localize labeling without tilting the labeling plane. Wong has reported a strategy to encode all vessels of interest using Hadamard type encoding (25) based on pCASL. To achieve this encoding, images are acquired with fixed gradients perpendicular to the labeling plane added. While this approach is efficient for characterizing flow through all vessels, it requires a relatively complex geometric prescription of the scan, and may require measurement of labeling efficiencies of each vessel to correct for imperfect Hadamard encoding. Measuring selective flow though one vessel may be all that is required, for example, when stenosis in a particular vessel, or collateral flow from a particular territory is being assessed. Selective labeling of an individual vessel, analogous to single vessel catheter injection, may be desirable.

Here we propose a single artery selective labeling method that requires only the specification of the target vessel position as input. The method employs pCASL to achieve high continuous labeling efficiencies. Rotating in-plane gradients are added to pCASL to achieve localized labeling while spoiling the undesired labeling of other vessels. This method effectively labels a disc around the targeted vessel (Fig. 1)

Figure 1.

The geometry of single artery selective labeling. Only a disc within the pulsed-continuous labeling plane surrounding the targeted vessel is labeled. A is the center of the targeted vessel, C is the isocenter, x is defined as the projection distance of AC along the gradient x direction, and y is defined as the projection distance of AC along the gradient y direction

MATERIALS AND METHODS

Pulse Sequence Design

The selective labeling method is based on PCASL (23), which can achieve high labeling efficiency for vessels across the labeling plane. The labeling of PCASL is achieved with a train of equally spaced, selective RF pulses. The labeling achieved is comparable to that of a constant RF and gradient with amplitudes equal to the time average of the RF and gradients for the PCASL. Our modification of the PCASL pulse sequence to achieve single artery selective labeling is shown in Fig. 2. In addition to the labeling gradient along the flow direction, we introduced in-plane gradients between the RF pulses. The in-plane gradients produce a phase shift between vessels. Single vessel selectivity is then achieved by rotating the direction of the in-plane gradients in analogy to the method of Werner et al. (21). Application of the gradients in the gaps between the RF pulses does not tilt the labeling plane, and hence the interference with nearby tissue above the labeling plane that can occur with the Werner et al. method need not occur with PCASL.

Figure 2.

The single artery selective pulsed-continuous sequence: label and control RF, and gradient waveforms. The same gradient waveforms were applied for label and control. For the rotating strategy with fixed amplitude, the amplitude of G_xy is fixed, but the direction of G_xy is changing from t to t.

The in-plane rotating gradients should selectively label a disc. The center of the selective disc is on the targeted single vessel. The center of the disc is controlled in the pulse sequence by the phases of the RF pulses. Each RF pulse must be incremented in phase relative to the prior pulse by an angle determined by the applied gradients and the desired disc center. This phase shift is given by

$$\mathrm{\Delta }\phi =\gamma G_{\mathit{zave}}\mathrm{\Delta }z\mathrm{\Delta }t+\gamma G_{\mathit{xave}}\mathrm{\Delta }x\mathrm{\Delta }t+\gamma G_{\mathit{yave}}\mathrm{\Delta }y\mathrm{\Delta }t$$ [1]

where γ is the gyromagnetic ratio, Δt is the time spacing between the center of two RF pulses, G_zave, G_xave, G_yave are the time average gradients between the center of two RF pulses along z, x and y directions respectively, Δz is the offset distance from the isocenter to the labeling plane, Δx is the projection distance of the segment from the isocenter to the target vessel along the gradient x direction, and Δy is the projection distance along the gradient y direction (see Fig. 1).

The fully refocused control approach (23) was found unsuitable for selective labeling because the control always produced less labeling than the label. Instead, the control of Wong (25), which employs the same gradients as the label but with alternating RF pulse sign was used.

Numerical Simulations

Numerical simulation of the Bloch equations was used to explore the labeling efficiency for vessels as a function of distance from the targeted vessel. Integration of the Bloch equations was performed for a blood isochromat starting 10s x velocity below the labeling plane, until ending 10s x velocity above the labeling plane, with an integration step size of 0.01 ms. The integration employed finite rotations around the effective field to increase accuracy with modest step size (26). T₁ and T₂ relaxation were included in the simulation but any magnetization transfer (MT) effects were not. The simulation was implemented in MATLAB (MathWorks Inc.). It required 1 s to calculate the efficiency for each velocity on a 1.8 GHz Pentium III Dell Inspiron 700m computer with Windows XP.

A Hanning window shaped RF pulse of 500 μs duration with an average B₁ amplitude of 1.7 μT and a spacing between two RF pulses of 1500 μs was used. A longitudinal gradient amplitude G_zmax of 9 mT/m during the RF pulse and a time average z gradient G_zave of 0.5 mT/m were assumed. Average in-plane gradient strengths G_xy (0.2–1.0 mT/m), and rotation rates between two consecutive RF pulses Δθ (10°–90°) were explored. Due to unsatisfactory results with fixed amplitude in-plane gradients (see results), amplitude modulation (AM) was explored. In the rotating strategy with AM, only a 15° rotation rate between two consecutive RF pulses was simulated. We multiplied the in-plane gradient amplitudes by a modulation function. This modulation function was constant within each Δt. Across Δt’s, the modulation function was a periodic triangle function. A periodic triangle function with minimum value 0.5 and maximum value 1.5, with period T_AM of N_AM (192 in this study) times the Δt was used (Fig. 3). Several kinds of different modulation waveforms were explored. The waveforms were based on triangle function y (i) = 2/N_AM (i%N_AM − N_AM/2) + 0.5, square function $y(i)=4.8/N_{AM}^2{(i\%N_{AM}-N_{AM}/2)}^2+0.6$, square root function $y(i)=6/7·(\sqrt{2/N_{AM}·\mid i\%N_{AM}-N_{AM}/2\mid }+0.5)$, and exponential function y(i) = (e^{2/T_AM (i%T_AM−T_AM/2)} +0.5)/(e − 0.5), where y(i) is the modulation value for the ith Δt, and % is modulus operator. The triangle function was chosen because the residual peaks of the triangle function were the smallest compared with other functions. The much larger period of the AM than the gradient rotation period was chosen to achieve slow modulation of the in-plane gradient amplitudes. The modulation was tended to modify the amplitude of the rotating gradients in a slow speed so that it can mimic the superstition of selective profiles with different rotating gradient amplitudes. For the rotating strategy with fixed amplitude, the residual peaks were both with positive and negative signals. Changing gradient amplitude corresponds to changing radius of labeling disc, which tends to cancel positive and negative gradients. Relaxation times T₁ and T₂ of 1.55 s (27) and 0.25 s (28) for blood were assumed.

Figure 3.

Schematic diagram of in-plane gradients used for the rotating strategy with amplitude modulation (AM). The in-plane gradient amplitude for each t is drawn as a single point. Top: inplane gradient with fixed amplitude is a constant (dashed line), while in-plane gradient amplitude with AM is a slowly-varying periodic triangle function relative to the rotating speed (solid line). Middle and bottom: the modulation pattern of x and y gradients from t to t.

Because simulations of the single vessel PCASL approach show sensitivity to the relative timing of the RF waveforms and the time the spin crosses the labeling plane, the labeling efficiency for a single velocity was calculated by averaging the efficiency with different delays between the RF waveform and the spin position timing. The delays varied from zero to the total period of the RF and gradient waveforms. A step size equal to one fifth of an RF spacing was used. In the case of the rotating strategy with the fixed amplitude for in-plane gradients, the period was Δt times 360/Δθ; while in the case of the rotating strategy with amplitude modulation, the period was T_AM. Laminar flow (v_max is the velocity at the center of the vessel) was assumed to estimate the labeling efficiency over the vessel’s cross-section. The total efficiency for a vessel with laminar flow was calculated as the weighted average of the efficiencies at many different individual velocities (26):

$$\mathit{eff}=\frac{2}{v_{max}^2}{\int }_0^{v_{max}}v\epsilon (v)dv$$ [2]

where ε(v) is the T₁ decay corrected efficiency for spins with velocity v. T₁ decay corrected efficiency compensates for blood T₁ relaxation after passing the labeling plane. The integration was evaluated within an error of 1% of the efficiency value using a recursive adaptive Simpson quadrature algorithm (29). The labeling efficiency was simulated for a range of vessel velocities.

Because static magnetic field perturbations may cause the resonance frequency of spins at the labeling plane to differ from the expected value for a uniform field and gradient, we also explored the effect of off-resonance on labeling efficiency.

In vivo Measurements

In vivo studies were performed to show the feasibility of the single artery selective ASL on human subjects and to verify the consistency between the in vivo efficiency and simulation results. Single artery selective PCASL was implemented to test for the artery selectivity of the pulse. A hanning window shaped pulse of 500 μs duration with average RF irradiation amplitude of 1.7 μT, a time between successive RF pulses of 1.5 ms, and a labeling duration of 1.5 s were chosen; Longitudinal gradient amplitude of 9 mT/m during the RF pulse, with average gradient amplitude of 0.5 mT/m were used; Average in-plane gradient amplitude of 0.28 mT/m, and a rotation rate of 15° increment between RF pulse repetitions were chosen. A post-labeling delay of 1.5 s (30) was selected to allow enough time for the labeled blood to arrive in the region of interest. Nonselective PCASL was performed within the same acquisition, immediately after selective pulsed-continuous ASL sequence, to serve as an efficiency reference.

All labeling strategies were implemented before a single-shot Fast spin echo (FSE) sequence with a partial k-space image acquisition. The sequence was performed with a TE of 25 ms in a 5-mm-thick axial slice through the superior part of the lateral ventricles. FSE images were obtained using 64 × 64 matrix on a 24-cm field of view. Raw data were saved from each channel of the coil array. The total time between image acquisitions, including labeling preparation and FSE echo train was 6 s.

Eight different subjects (five males and three females, 21–50 years old) were studied. The study protocols were approved by the local institutional review board and all subjects provided written informed consent. Imaging was performed on a GE 3.0 Tesla HD scanner using the receive-only 8-channel head array coil and the body transmit coil. Two separate studies were performed. Both studies began with a brain localizer and a 3-dimensional MR angiogram to choose the labeling location.

The first study was designed to demonstrate the feasibility of artery selective labeling with near optimal parameters from the simulations. Perfusion territory images were acquired with labeling of the left internal carotid, right internal carotid and basilar artery in four volunteers. The amplitude modulated version of the selective labeling strategy was employed. Each sequence automatically acquired 4 dummy scans to achieve steady state magnetization, 12 interleaved artery selective label and control pairs and 12 interleaved nonselective pairs for a total scan time of 5.2 minutes.. The superior inferior labeling location of the basilar artery was placed approximately 1.3 cm superior to the merging of the vertebral arteries. At this level, the internal carotid arteries are approximately 2.6 cm from the basilar artery. The labeling plane for the internal carotid arteries was selected 1.3 cm inferior to the merging of vertebral arteries so the two internal carotid arteries are well separated. The center of each target vessel was measured on the 3D MR angiogram and then numerical values were entered into the sequence as user control variables.

The second study was designed to measure the inversion efficiency profile as a function of distance offset between the center of labeling and the target artery. Four volunteers were imaged in this study. Perfusion signal was measured at distance offsets from the targeted internal carotid artery of 3 mm towards the posterior direction, 0 mm and 3 mm, 9 mm, 15 mm, 22 mm, 29 mm, 36 mm, 43 mm, 50 mm, 70 mm towards the anterior direction. The amplitude modulated version of the selective labeling strategy was employed. For each distance offset, the sequence was performed for 28 repetitions: 4 repetitions for signal stabilization, 6 label and control pairs of artery selective labeling, and 6 pairs for nonselective labeling. We chose the right internal carotid as the target artery. As an add-on to this study, we investigated whether the amplitude modulated in-plane gradients were required as predicted by simulation. We measured the efficiencies for the rotating strategy both with fixed amplitude and with amplitude modulation at certain fixed distance offsets. Due to time restrictions, we could only measure the efficiency at a single distance offset in each volunteer (115 mm with two volunteers, 235 mm with one volunteer, 330 mm with one volunteer, all towards the anterior direction).

To show the feasibility of volumetric acquisition with selective labeling, control minus label difference images were acquired in 2 volunteers with the 2D sequence replaced with an interleaved 3D sequence. The amplitude modulated version of the labeling was employed and applied to the right internal carotid artery. Background suppression with the selective region expanded to a 192-mm axial slab was performed and image acquisition was an interleaved stack-of-spirals FSE acquisition. One of eight interleaved spiral gradient waveforms was performed for each excitation at each of 40 centrically-ordered slice encodes. The eight interleaves were acquired in separate acquisitions with a TR of 6 s. Three averages of label and control pairs required a total of 288 s. The resulting images had nominal in-plane resolution of 3.4 mm and slice thickness of 4 mm.

Analysis

All image data were saved as raw echo intensities and reconstructed offline with custom software. Partial Fourier raw data was acquired at lines −m ≤ u ≤ N/2 (m = 6, N= 64). The low frequency phase map of each coil was estimated from the average signal of the control image in the nonselective PCASL. The low-resolution phase maps were used to correct the phase of the image from each coil. The final image was combined from each coil image, weighted by the conjugate of the low-resolution phase map (31).

For the first study, region of interests (ROIs) for each subject were manually drawn on the selective PCASL images: left internal carotid territory, right internal carotid territory, and basilar territory. Regions of interest for the two anterior and the posterior territories were defined using the known anatomy and the clear separation of the territories on the selective images. To avoid any contribution of multiple vessels near the boundaries of the territories, ROI’s were restricted to the center of the territories. For the second study, the target artery (right internal carotid) territory was drawn on the selective PCASL image. Contralateral ROI’s were drawn on the subtraction image between nonselective and selective.

The difference between control and label was averaged across all repetitions for selective PCASL and nonselective PCASL. The relative efficiency of selective PCASL was defined as the ratio of the average difference signal over the ROI between selective PCASL and nonselective PCASL. The relative efficiency was calculated for the target territory ROI and the contralateral ROI for each distance offset of the second study.

RESULTS

Simulations

Simulations of the rotating gradient strategy with fixed amplitude showed an initial rapid drop in efficiency with distance from the target, but with substantial residual peaks at large distances (Fig. 4a). The simulation results prefer a slow gradient rotation rate (a small angle increment between two consecutive in-plane gradient directions). Simulation with faster gradient rotation rates (not shown) tended to have worse suppression of efficiency at large distances characterized by larger peaks in the efficiency vs. distance profile. These residual peaks indicate partial labeling of potential vessels at large distances from the targeted vessel. Addition of amplitude modulation to the slower gradient rotation appeared to spoil the coherent “ringing” at large distances and resulted in good suppression of distant vessels (Fig. 4b). To test whether the noise like appearance of the labeling at a distance from the labeling location was caused by the numerical instability, simulations were also performed with 10x smaller step. No difference in efficiency great than 0.1% was observed. This indicated that the numerical instability was not present in the simulation.

Figure 4.

Simulated inversion efficiency as a function of offset from targeted vessels. (a) G_xy rotating at different rates: 150, 300 increments between two consecutive RF pulses but with fixed amplitude. (b) G_xy rotates in 150 increments. The gradient amplitude was either fixed (dashed line) amplitude or slowly modulated in amplitude (AM) between RF pulses (solid line). An average xy gradient of 0.028 mT/m was used.

The calculated sensitivities of the single vessel selective labeling efficiency profile to different velocities and frequency offsets are shown in Fig. 5a and Fig. 5b respectively. The radius of the selective disc decreases as the velocity drops. Frequency offsets caused marked inversion efficiency loss even within a few millimeters from the center of the targeted vessel.

Figure 5.

Simulated inversion efficiency as a function of offset from targeted vessels. The transverse gradient was rotated in 150 increments between two RF pulses with amplitude modulation. (a) calculated inversion efficiency at velocities of 40.3 and 10 cm/s. (b) calculated inversion efficiency at different frequency offsets of 0, 1/(8 t), 3/(16 t) with an assumed peak vessel velocity of 40.3 cm/s. All efficiencies are relative to the efficiency for 0 offset, no off-resonance, and 40.3 cm/s

In-vivo Results

In-vivo measurements of the spatial selectivity profile of vessel selective PCASL were qualitatively consistent with the simulation results. Profiles of distal perfusion labeling as a function of distance between the target region and the vessel are shown in Fig. 6. Though the efficiency relative to non-selective PCASL was slightly less than in the simulations, the shape of the curve is similar. Note the excellent suppression at large distances. In contrast, the rotating strategy with fixed amplitude produced substantial labeled signal at the three distance offsets that were measured (Fig. 7). The average labeling efficiency from the three distance offsets were 0.20 ± 0.09 for the rotating strategy with fixed amplitude, but with a much smaller efficiency of 0.05 ± 0.03 for the rotating strategy with amplitude modulation. These results support the necessity of amplitude modulation to avoid the distant labeling of the fixed amplitude strategy.

Figure 6.

Mean In-vivo inversion efficiency as a function of the distance offset from the right internal carotid artery (posterior directions shows as positive offsets, and anterior as negative offsets). (a) inversion efficiency drop-off profiles (solid curve) and simulation profile (dashed curve) with amplitude modulation for a velocity 40.3 cm/s. (b) contralateral efficiency relative to nonselective PCASL. Error bars are standard deviation averaged across four subjects.

Figure 7.

Comparison of the inversion efficiencies for the rotating strategy (a) with fixed amplitude; (b) with amplitude modulation, both positioned at a distance offset of 23 cm from the right internal carotid artery; and (c) the corresponding nonselective PCASL image. The signal on (a) and (b) were multiplied by 5 for better visibility. The undesired residual labeling of the distal artery that appears in fixed amplitude, as expected from the simulations, is spoiled with amplitude modulation.

Separation of the two anterior and the posterior perfusion territories was highly successful, Fig. 8. In-vivo inversion efficiencies in the different arterial territories (left internal carotid, right internal carotid and basilar artery) are shown in Fig. 9. Efficiencies relative to pulsed-continuous ASL were 84%, 87% and 80% for left internal carotid, right internal carotid and basilar artery respectively. Cross talk between the territories was small but an approximate 8% labeling is observed in the basilar during internal carotid labeling and vice versa. Given the close proximity of the basilar artery to the internal carotids, this degree of labeling is not surprising. Based on the acquired data, we cannot exclude the possibility that the cross-territory labeling we measured between the posterior and anterior circulation could represent actual mixing of flow across the territories, but the mixing is qualitatively consistent with labeling imperfections.

Figure 8.

Artery selective imaging of (a) right internal carotid artery, (b) left internal carotid artery, (c) basilar artery, compared with (d) nonselective labeling.

Figure 9.

In-vivo inversion efficiency in the different territories when different vessels are labeled with the single artery selective strategy. The efficiencies relative to pulsed-continuous ASL are 84%, 87% and 80% for left internal carotid, right internal carotid and basilar artery respectively. For one internal carotid selective labeling, the perfusion signal in the other internal carotid artery territory is close to the noise level, but the perfusion signal in the basilar artery may be visible in some subjects; for the basilar selective labeling, the perfusion signal in both internal carotids territories

The acquisition of multiple slices using the single artery selective technique was readily achieved. Control minus label difference images from a 3D acquisition using the rotating strategy with amplitude modulation are shown in Fig. 10. The images clearly show labeling of the right internal carotid perfusion territory without contamination from other territories.

Figure 10.

Multislice perfusion difference images from a 3D whole brain acquisition by selectively labeling the right internal carotid artery using the rotating strategy with amplitude

DISCUSSION

Our results demonstrate the feasibility of precise selective labeling even close to the target imaging region. Our computer simulation proved useful at predicting the undesirable labeling at large distances when rotation of a fixed amplitude gradient is applied. The improvement of selectivity with slow amplitude modulation of the gradient suggested by the simulation also was confirmed in-vivo.

The agreement between the simulations and the in-vivo results were not perfect in our study. In particular, the efficiency at the center was lower than the simulations. Other studies using selective forms of pCASL have reported variable efficiencies (25). Their efficiencies ranged from 54 to 91%. It may be that a combination of hardware imperfections and magnetic field non-uniformity contribute to variation from the ideal. Still, the basic properties of selectivity and improvement suppression of distant flow with amplitude modulation were in agreement between the simulations and the in-vivo results.

The single vessel selective ASL method described here provides an efficient selective labeling of individual arteries. The in-vivo labeling efficiency for the selected vessel is above 80% of that achieved in the nonselective, pulsed-continuous ASL experiment, although the in-vivo labeling efficiency is lower than the simulation predicted. B₀ field inhomogeneity, imperfect placement of the center of the selective disk, and potentially gradient eddy currents could contribute to this loss of efficiency.

Though our vessel selective strategy was proven highly selective, interaction between the basilar and the internal carotid artery territories was observed (Fig. 8b and 9). While we cannot exclude the possibility of true physiologic mixing of supply, our measurements (Fig. 6) suggest that the 8% residual labeling is consistent with a 2 cm distance from the target. The residual signals can be removed by increasing the amplitude of in-plane gradients. However, the labeling efficiency may be reduced by the increased gradient amplitude.

The single vessel selective strategy is a simple yet effective approach, which does not require the geometries of other vessels as an input to calculate the sequence parameters (25). For the CASSL approach based on the amplitude modulated control (21), the labeling plane was tilted and rotated so that only the neighborhood of a single vessel could be efficiently labeled. But the labeling plane may interfere with the image region, which is undesirable imaging artifact. The labeling plane of the selective PCASL approach was designed as rotated but not tilted plane. Therefore the selective PCASL approach removes any issues of artifact associated with overlap between the tilted labeling plane and the desired imaging region.