Pseudo-Continuous Transfer Insensitive Labeling Technique

Abstract

Transfer insensitive labeling technique (TILT) was previously applied to acquire multi-slice cerebral blood flow (CBF) maps as a pulsed arterial spin labeling (PASL) method. The magnetization transfer (MT) effect with TILT is well controlled by using concatenated radiofrequency pulses. However, use of TILT has been limited by several challenges, including slice profile errors, sensitivity to arterial transit time and intrinsic low signal-to-noise ratio (SNR). In this work, we propose to address these challenges by making the original TILT method into a novel pseudo-continuous ASL approach, named pseudo-continuous transfer insensitive labeling technique (pTILT). pTILT improves perfusion acquisitions by 1) realizing pseudo-continuous tagging with non-adiabatic pulses, 2) being sensitive to slow flows in addition to fast flows, and 3) providing flexible labeling geometries. Perfusion maps during both resting state and functional tasks are successfully demonstrated in healthy volunteers with pTILT. A comparison with typical SNR values from other perfusion techniques shows that although pTILT provides less SNR than inversion-based pseudo-continuous ASL techniques, the modified sequence provides similar SNR to inversion-based PASL techniques.

INTRODUCTION

Arterial spin labeling (ASL) permits noninvasive cerebral blood flow (CBF) measurements by utilizing water in blood as an endogenous perfusion contrast agent (1,2). Currently, there is a large family of ASL methods. Continuous arterial spin labeling (CASL) uses long flow-driven adiabatic inversion pulses (2–4 s) to label fast flow spins in main feeding arteries of the brain. CASL should theoretically produce the highest signal-to-noise ratio (SNR) (3,4) among all the ASL methods, however, CASL is limited by the availability of near continuous wave radiofrequency (RF) transmission and other imperfections in multi-slice acquisition (5,6). Pulsed arterial spin labeling (PASL) applies a short RF inversion pulse to a thick labeling slab proximal to the imaging slices to generate a bolus of labeled magnetization (7–9). Although more easily implemented than CASL techniques, the PASL techniques suffer from significant reductions in SNR. Recently, an intermediate strategy, named pseudo-continuous arterial spin labeling (pCASL), was proposed, by using a train of discrete RF pulses and gradient fields to mimic the flow-driven adiabatic mechanism (10–12). By combining the advantages of PASL and CASL, pCASL acts as a desirable alternative approach for perfusion measurements with good SNR but without the need for specialized hardware.

Transfer insensitive labeling technique (TILT) and its derivatives have been used as a PASL method by employing concatenated RF pulses (13–16). The technique was designed to completely eliminate magnetization transfer (MT) effects from the ASL signal. The inversion in TILT is achieved by using two concatenated slice-selective 90° RF pulses in a particular way: the 2^nd 90° RF pulse is inverted in time and the 2^nd slice-selective gradient has opposite sign compared to the 1^st one. For control preparation, the 2^nd RF pulse has a 180° phase shift. With this RF concatenation scheme, the MT signals from the static tissue in imaging slices are cancelled out between control and label sessions, as the same RF pulses with the same timing are employed in both sessions. However, in addition to intrinsic low SNR and sensitivity to arterial transit time effect as a PASL method (13,15,16), TILT can suffer from potentially large slice profile artifacts from static tissue in the subtracted perfusion weighted images, leading to significant perfusion quantification errors. A straightforward solution is to make the gap bigger between labeling and imaging slabs, but this will reduce the available imaging coverage of the brain. An alternative approach to remove the effects of labeling pulses on the imaging volume was proposed in (15). Three saturation pulses followed by strong gradient spoilers were implemented in the imaging region subsequent to the TILT labeling. In the current work, we will use narrow labeling planes to limit the contribution of the labeling pulses to the imaging plane.

Retaining the magnetization transfer insensitive feature, here we propose to make TILT into a novel pCASL technique, named pseudo-continuous transfer insensitive labeling technique (pTILT). pTILT improves perfusion acquisitions by 1) realizing pseudo-continuous tagging with non-adiabatic pulses, 2) being sensitive to slow flows in addition to fast flows, and 3) providing flexible labeling geometries. A set of studies were carried out to explore the feasibility of pTILT as a novel pCASL approach. First, numerical simulations were performed to exam the labeling efficiency of pTILT as a function of flow velocity. Second, we investigated and compared the slice profile artifacts of TILT and pTILT. Third, perfusion acquisitions during resting state and functional activations were demonstrated in human brain with pTILT.

MATERIALS AND METHODS

MR pulse sequence

In TILT, a single pair of concatenated 90° RF pulses were implemented in a thick labeling slab (140 mm) to achieve an inversion slice profile (13). Figure 1 shows the labeling and imaging geometry for TILT and our proposed pTILT acquisitions. For the conversion of TILT to pTILT, we propose three major modifications: (1) rather than implementing the RF concatenation pair only once, we repeated it for 100 times to reach pseudo-continuous labeling status. (2) The labeling slab was reduced to 10 mm in pTILT, as shown by the yellow box below the imaging slices in Figure 1b. (3) We use 45° RF pulses in pTILT, instead of the 90° pulses in TILT, therefore, we use saturation for labeling. Although this results in some loss of SNR, we gain some of the lost SNR from the transition to pseudo-continuous labeling. The detailed MR pulse sequence parameters are listed in Table 1.

Figure 1.

Schematic of the labeling geometry for (a) TILT, (b) global pTILT, (c) localized pTILT for visual area and (d) localized pTILT for motor area. Orange box: imaging slices and dashed yellow box: labeling slab. pTILT pulse sequence for control (e) and label (f) sessions. T_ps: RF pair spacing, τ: RF spacing, T_s: total labeling duration, w: post-labeling delay.

Table 1.

MR pulse sequence parameters:

Imaging parameters: FOV = 22 cm, in-plane matrix size = 64×64, TR/TE = 5000/44 ms, SE-EPI readout, imaging slice thickness = 6 mm, slice gap = 3/1.2 mm. Averages = 30. TILT: labeling slice thickness = 50 mm, RF pair repetitions = 1, postlabeling delay (w) = 1 s, gap between imaging and labeling regions = 10 mm. Global pTILT: labeling slice thickness = 10 mm, RF pair repetitions = 100, total labeling duration = 3 s, postlabeling delay (w) = 1 s. Localized pTILT: labeling slice thickness = 10 mm, RF pair repetitions = 100, total labeling duration = 3 s, postlabeling delay (w) = 0.5 s, gap between imaging and labeling regions = 10 mm. RF concatenation pair: windowed-sinc 90°/45° RF pulse with duration = 2560 us, RF spacing (τ) = 2960 us, RF pair spacing (Tps) = 30 ms, spoiler duration and amplitude = 4000 ms/[±10, ±12, ±14, ±16 mT/m].

Human subjects

Four subjects (three male and one female, 20–30 years old) were studied on a 3 T Siemens (Erlangen, Germany) Trio scanner using a standard body coil transmission and a twelve-channel head array receive coil, following a protocol approved by the Institutional Review Board of the University of Illinois at Urbana-Champaign. To reduce movement, padding was used to stabilize the subject’s head.

Perfusion quantification

Pairwise subtraction between control and label images was used to generate flow-weighted images. Perfusion in the units of mL/100mL/min was calculated based on Eq. [1], in which a single compartment model and no blood exchange are assumed (17,18):

$$\mathit{CBF}=\frac{\mathrm{\Delta }M}{M_{o,\mathit{CSF}}}·\frac{6000}{2·{\lambda }_{\mathit{blood}}·\alpha ·T_{1,\mathit{blood}}}·exp\left(\frac{w+T_{\mathit{slc}}·(n-1)}{T_{1,\mathit{blood}}}\right)·exp\left(\frac{TE}{T_{2,\mathit{blood}}}\right),$$ [1]

where ΔM is the flow-weighted image, M_o,CSF is the measured intensity of cerebrospinal fluid (CSF) in a voxel in the ventricles. λ_blood, α and w are the water content of blood (0.76 as used in (17)), labeling efficiency and post-labeling delay respectively. T_1,blood (1680 ms at 3 T) and T_2,blood (275 ms at 3 T) are the longitudinal and transversal relaxation rates of blood (19). T_slc is the EPI readout duration of one single slice, and n is the index of acquired slice.

We perform three studies to show the performance of pTILT. For the first study, we examine the labeling efficiency of the proposed pTILT technique using simulations and show its sensitivity to a wide range of flows. In the second study, we compare pTILT to TILT in terms of static tissue contamination in the flow signal by implementing both techniques in volunteer subjects. Finally, in the third study, we examine the performance of pTILT in measuring baseline and task-related flows in the brain of healthy volunteers.

Study 1: Simulation of labeling efficiency

Labeling efficiency is one of the most critical parameters in perfusion measurement with pCASL. Numerical Bloch equation simulations developed in MATLAB (MathWorks Inc.) were performed to explore the labeling efficiency of pTILT as a function of blood velocity. The simulation parameters used in this analysis were the same as in Table 1, identical to those in the human studies. Plug flow was assumed with a range of 0–50 cm/s. The labeling efficiency α was then calculated by Eq. [2]:

$$\begin{array}{l}{\alpha }_{\mathit{label}}=\frac{M_{z0}-M_z^{\mathit{label}}}{2M_{zo}},{\alpha }_{\mathit{control}}=\frac{M_z^{\mathit{control}}}{2M_{zo}}\\ \alpha =\frac{M_z^{\mathit{control}}-M_z^{\mathit{label}}}{2M_{zo}}\end{array}$$ [2]

where α_label, α_control, α are the label, control and combined labeling efficiency respectively. M_z is the magnitude of the remaining longitudinal magnetization of blood and M_z0 is the equilibrium magnetization of blood. Note that, in Eq. [2], the denominator is 2M_zo instead of M_zo, as we are calculating our labeling efficiency to be comparable to the efficiency of other inversion-based ASL methods. Since pTILT is a saturation technique, we lose a factor of 2 in our efficiency compared to these other techniques. Three groups of pTILT labeling parameters were investigated: labeling slice thickness/RF pair spacing (T_ps) = (I) 10 mm/20 ms, (II) 10 mm/30 ms, (III) 20 mm/20 ms.

Study 2: Labeling slice profile artifacts

Experiments were carried out to compare the level of flow contamination from static tissues from the labeling slice profile effects between pTILT and TILT. For each sequence, five axial slices were acquired twice, by placing the labeling slab below and above the imaging slices. The gap between labeling and imaging slabs was kept the same for all acquisitions (10 mm). Here, the gap was defined as the distance between the centers of the labeling slice and the bottom imaging slice minus half their nominal widths. By placing the labeling slab above the imaging slices, we assume that the flow-related signal was insignificant and the estimated flow map represented static tissue contamination from the labeling pulse.

Study 3: Perfusion acquisition

Two sets of perfusion measures were carried out on volunteer subjects to show reliable measures of global perfusion, flexible labeling geometries, and sensitivity of the technique to task-induced changes in flow. The first scan was used to acquire resting perfusion maps for whole brain coverage with pTILT. A similar geometry as existing pCASL was used in this study, by labeling at the level of main feeding arteries, e.g., internal carotid arteries (Figure 1b) and using a post-labeling delay of 1 s.

In the second scan, all subjects were instructed to perform two functional tasks, a visual task and a motor task. In the visual task, subjects passively viewed a blocked visual presentation that consisted of 30 s of display of a fixation cross followed by 30 s of a checkerboard reversing at 8 Hz. Five repeats of the task/fixation blocks were employed per functional run with a total duration of 5 minutes. Four coronal imaging slices covering the visual cortical areas were acquired with pTILT, and the labeling slice was placed anterior to the imaging slab with a 10 mm gap (Figure 1c). In the motor task, subjects were asked to perform sequential finger-to-thumb tapping for 30 s with five repeats, interleaved with a 30 s rest. The total duration was also 5 minutes. Six imaging slices covering the hand area of the primary motor cortex were acquired, and the labeling slice was placed 10 mm below the edge of imaging slab (Figure 1d). A post-labeling delay of 0.5 s was used for these two functional acquisitions.

Activation maps were formed using FSL 4.1.4 (FMRIB Software Library; http://www.fmrib.ox.ac.uk/fsl). The time series was high-pass filtered with sigma = 60 s. Spatial filtering was performed with a Gaussian kernel with a full width at half-max of 5 mm. The signal was modeled by multiplication of a label/control regressor with a block model of the task that was convolved with a gamma-function hemodynamic response function. Z (Gaussianised T/F) statistic images were thresholded using clusters determined by Z≥2.3 and a (corrected) cluster significance threshold of P=0.05.

RESULTS

Study 1: Simulation of labeling efficiency

Figure 2 depicts simulated label, control and the combined labeling efficiency for a range of velocities. The results indicate that pTILT has good and stable efficiencies for both fast and slow flow spins. For comparison, the blue dashed curve in Figure 2a shows the shape of the combined efficiency for a flow driven adiabatic inversion pCASL method (12). Note that the inversion-based pCASL methods benefit from a factor of 2 in labeling efficiency due to the use of inversion. Adiabatic flow driven pCASL provides high efficiency for fast spins (≥15cm/s), but lower efficiency for slow spins (≤10 cm/s). The reason for the reduced efficiency is that slow spins do not satisfy the flow-driven adiabatic condition very well.

Figure 2.

Simulated label (dashed line), control (dotted line) and combined (solid line) efficiency as a function of blood velocity with the pTILT sequence. (a) Efficiency comparison between different RF pair spacings (T_ps=20 ms, red, and T_ps=30 ms, black) with the same labeling slice thickness 10 mm. (b) Efficiency comparison between different labeling slice thickness (10 mm, red, and 20 mm, black) with the same RF pair spacing (T_ps=20 ms). T1 and T2 relaxations of blood are not considered in this simulation. The blue dashed line in (a) denotes the labeling efficiency of flow-driven adiabatic inversion pCASL methods for comparison (12).

Study 2: Labeling slice profile artifacts

The flow-weighted images (ΔM/M_o) by TILT and pTILT are shown in Figure 3. When the labeling slice is placed below the slab of imaging slices, these images contain both flow-related signal and labeling slice profile induced artifacts. When the labeling slice is placed above the imaging slab, only label-induced static tissue artifacts are expected. Therefore, the image amplitude in Figure 3b and 3d represents the slice profile errors for TILT and pTILT, respectively. TILT showed large artifact signals in the imaging slices close to the labeling slab, and smaller artifact signals in slices further away (Figure 3a and 3b). pTILT showed consistently reduced artifactual signals throughout the imaging region. Figure 3c shows the perfusion-weighted images from pTILT without significant slice profile errors and consistent sensitivity to flow across the multi-slice acquisition.

Figure 3.

Flow-weighted images (ΔM/M_o) acquired by using TILT labeling applied (a) below and (b) above the imaging slices, and using pTILT labeling applied (c) below and (d) above the imaging slices. Note that no flow signal is expected when labeling above the imaging slices.

Study 3: in vivo perfusion acquisition

A typical set of perfusion-weighted maps from global pTILT is shown in Figure 4a. Figure 4b and 4c are the localized perfusion maps acquired during visual (coronal slices) and motor (axial slices) functional tasks, respectively. Example z-score activation maps by pTILT and blood oxygenation dependent level (BOLD) are shown in Figure 4d (visual) and 4e (motor). The BOLD data are co-acquired with the pTILT data. The activation results were overlaid on a high-resolution 2D T2 turbo-spin echo acquisition taken with slices at the same location as the functional acquisitions.

Figure 4.

(a) An example of multi-slice, whole-brain perfusion-weighted images with pTILT. (b) Coronal perfusion-weighted images of visual cognitive area with localized pTILT. (c) Axial perfusion-weighted images of motor cognitive area with localized pTILT. (d) Z-score activation maps by co-acquired BOLD (1^st column) and pTILT (2^nd column) during a visual stimulus task. (e) Z-score activation maps for the motor task, same arrangement as in (d).

Perfusion maps were quantified by Eq. [1]. Mean CBF values and mean temporal SNR in ROIs were calculated for each subject. For resting global pTILT perfusion maps, a gray matter ROI for each subject was obtained in a similar way to previous literature (20), by calculating quantitative perfusion maps and applying a threshold for each voxel with CBF ≥ 40 ml/100 g/min. A potential drawback to this approach is that it might induce a bias toward high perfusion values and overestimate the average perfusion in gray matter. For functional perfusion acquisitions, the ROI was taken as the activated voxels (those that exceeded the cluster-level statistical threshold as explained above). The temporal SNR of single repetition is calculated by dividing the mean flow-weighted signal in each voxel by the standard deviation through the time course (30 repetitions) followed by spatially averaging over each ROI (20,21). All the values are listed in Table 2, along with a comparison with literature values of temporal SNR from other ASL techniques.

Table 2.

CBF (mL/100 g/min) and temporal SNR (per image) values in four subjects: for comparison, average SNR values for several other techniques from the literature are included.

Subjects		1	2	3	4	Mean	SD
Global pTILT	CBF	70.1	70.4	70.5	67.4	69.6	1.3
	SNR	1.0	1.0	0.9	0.9	1.0	0.1
Localized pTILT (visual)	CBF	78.7	63.8	64.9	72.4	70.0	6.0
	ΔCBF%	92.3	94.5	63.5	98.5	87.2	13.9
	SNR	2.0	1.8	2.0	1.7	1.8	0.2
Localized pTILT (motor)	CBF	74.1	76.3	72.8	79.7	75.7	2.6
	ΔCBF%	94.4	92.6	42.0	56.8	71.5	22.7
	SNR	2.3	1.7	1.7	1.9	1.9	0.3
VSASL*	SNR					1.1	0.2
PICORE/QUIPSS II*	SNR					2.1	0.3
pCASL*	SNR					4.9	2.4

^*The temporal SNR values of VSASL, PICORE/QUIPSS II and pCASL are from references (20), (20) and (21), respectively.

pTILT gives much higher SNR when applied as a localized perfusion acquisition scheme (1.8±0.2 for posterior area and 1.9±0.3 for superior area) than when applied as a whole-brain coverage scheme (1.0±0.1). However, the CBF values obtained with these two different geometries are in similar range. Figure 4d and 4e indicate a good overlay of detected areas by pTILT and co-acquired BOLD during functional stimuli. In the visual cognitive task, the measured average CBF in the activated voxels of four subjects is 70.0±6.0 mL/100 g/min, and the perfusion change is 87.2±13.9%. In the finger tapping experiment, the average CBF value in the activated voxels is 75.7±2.6 mL/100 g/min, and the average flow change is 71.5±22.7%. These results match the literature reports for flow changes in similar functional tasks (22,23).

DISCUSSION

Labeling slice profile artifacts

Negligible labeling slice profile errors are induced in pTILT by using a small labeling slice and small gap, thus permitting pTILT to label close to tissues of interest. There is a relationship between the thickness of the labeling slice and the required spatial gap to ensure insignificant contributions to the flow signal from the saturation profile of the labeling pulse. As shown in Figure 3, a smaller labeling slice in pTILT introduces much smaller slice profile error in the imaging region. By using a thinner labeling slice, we can make the gap between label and imaging slab smaller without causing slice profile errors. In addition, given the high labeling efficiency of pTILT to slow flows, closer labeling will not result in a reduction of labeling efficiency.

Labeling efficiency

In pTILT, we can adjust the sequence timing and geometry parameters to achieve maximal labeling efficiency for different velocities. For example, in Figure 2a, a higher label efficiency is achieved for faster flowing spins by using a shorter RF pair temporal spacing with the same labeling slice thickness (10 mm), while Figure 2b illustrates that with the same RF pair spacing (T_ps= 20 ms), a wider labeling slice thickness generates higher label efficiency for faster flows.

Roughly speaking, the label efficiency increases for faster flows with labeling slice thickness, and decreases with increased RF pair spacing. The critical velocity of pTILT (i.e., the velocity at which the label efficiency begins to decrease quickly) can be estimated by dividing the labeling slice thickness by the RF pair spacing. For example, for labeling slice 10 mm and T_ps of 30 ms, the critical velocity is around 30 cm/s as shown in Figure 2a. Another potential effect on labeling efficiency is flow-induced phase changes due to the labeling slab selection gradients. Although flow-induced phase changes do not play a major role in our current experimental setup, consideration must be given to the impact on efficiency from flow-induced phase changes for other parameter settings.

Based on our simulation results, wider labeling slice thickness and shorter RF pair spacing lead to higher labeling efficiency for faster flows, however, the choice of these two parameters are limited by other considerations. For example, the cost of using a wider labeling slice is that we need a bigger gap between labeling and imaging slabs to avoid slice profile errors and contamination of the ASL perfusion signal by saturated static tissues. If we use smaller RF pair spacing, then a larger number of RF pairs need to be implemented in order to keep the same total labeling duration. This will increase specific absorption rate (SAR) for the subject.

Better sensitivity and labeling efficiency with CASL and pCASL relies on a higher pulse power, however, the maximum pulse power is constrained by the magnet’s SAR limits, especially in high magnetic field (≥ 3 T) (24). While in the pTILT sequence, we chose relatively long RF pulses (2.56 ms) which have a 45° flip angle and the RF pulse pairs are spaced by 30 ms, resulting in an effective duty cycle of 17.1%, much lower than that by CASL and pCASL (24). Although as many as 100 repetitions of the RF pairs were carried out, the measured SAR levels of the 3 T pTILT sequence, as determined by the manufacture’s power supervision, never exceeded 20% of the SAR limit for the protocols given in Table 1.

SNR analysis

Theoretically, global pTILT would have half the temporal SNR of that by flow-driven based pCASL approaches due to the use of saturation for labeling in pTILT instead of inversion. As seen in the experimental results in Table 2, global pTILT provides SNR as high as Velocity-Selectively ASL (VSASL) (20), which also uses saturation for labeling. The SNR with global pTILT is around half the value by the pulsed ASL method of (PICORE/QUIPSS II) (20) and around a fifth of that of a pCASL SNR measurement in the literature (21). There are multiple possible reasons for the lower SNR with global pTILT than the expected value: first, the mean and the peak velocities through internal carotid arteries (ICAs) are reported in the range of 30 cm/s (25) and 75 cm/s (26) in the literature, respectively. The critical velocity in our pTILT protocal was around the mean blood flow velocity in ICAs, but smaller than the peak velocity. Therefore, we may lose labeling efficiency for fast spins (> 30 cm/s). The sensitivity to faster flowing spins can be adjusted through geometric and timing parameters in the sequence. Second, pTILT labeling is sensitive to off-resonance effects due to the time spacing between the paired RF pulses (τ, as shown in Figure 1d). Strong field inhomogeneity (ΔB_o) was observed around the labeling sites during our scans, and it worsens the experimental SNR with global pTILT. A correction method proposed previously by our group may be used to address some of this loss in SNR (27). We note that other balanced and unbalanced pCASL methods have also been reported to be very sensitive to this effect (12,28). Third, B1-inhomogeneity may play a role in reducing the labeling efficiency with pTILT as well, since non-adiabatic pulses were implemented in pTILT and their performance is known to be B1-dependent, especially at high magnetic fields (≥ 3 T).

Localized pTILT is expected to benefit from a gain of 1.3 times higher SNR than that of global pTILT by implementing a shorter post-labeling delay (PLD) of 500 ms, while the experimental SNR with localized pTILT was 1.9 times higher than global pTILT. There are several possible explanations for this improvement. First, in localized pTILT, the labeling slice was placed further downstream in the arterial pathway, at a location in which the blood flow velocity should be slower than the critical velocity (30 cm/s) and thus high labeling efficiency can be achieved. Second, the labeling slice of localized pTILT was located in more superior regions in the brain, which have more homogeneous magnetic field map than that at the ICA sites where global pTILT labels. This results in increased effectiveness of the labeling pair of RF pulses and less signal loss due to field inhomogeneity. Third, it was observed in our studies that repetitive applications of labeling RF pairs could relieve the B1-dependence of the non-adiabatic RF pulses, resulting in a better saturation profile. For the same labeling slice thickness and RF pair spacing, slower flowing spins labeled in localized pTILT could be exposed to a higher number of repeated saturation RF pairs, resulting in more complete labeling.

According to Table 2, localized pTILT provides around twice the SNR reported for VSASL (20), nearly the same SNR as pulsed ASL (20) and 0.4 times the SNR reported for pCASL (21). Note that the comparisons of SNR between these different methods is only approximate, since different imaging parameters were used in these studies, such as the voxel size, echo time and post-labeling delays. However the comparison provides a basic idea about the SNR level achieved by pTILT.

Asymmetry in the perfusion signal (left versus right hemisphere) was observed in some of the subject data, as shown in Figure 4a. Asymmetry indices were calculated for each subject, defined as one minus the ratio of MIN(CBF_left, CBF_right) over MAX(CBF_left, CBF_right) with CBF_left the mean CBF in the left ICA territory. The mean indices for all four subjects are 0.09±0.03 and 0.07±0.04 for global and localized pTILT, respectively. The higher asymmetry index in global pTILT may be due to differences in the magnetic field from magnetic susceptibility effects in the more inferior labeling planes. For the superior brain regions that are in the labeling plane for localized pTILT, shimming results in much smoother magnetic field maps. The improvement in magnetic field inhomogeneity results in more complete label and control preparation from the concatenated RF pulse pairs.

Transit time in pTILT

Compared to pCASL and PASL acquisitions, in which a typical PLD of 1000 to 1500 ms is used (8,12,13,24), there are several reasons as to why we propose that localized pTILT enables the use of a shorter (500 ms) PLD without reducing microvascular sensitivity. (i) In our experiments, a long labeling duration (3 s) with pTILT helped weight the flow-weighted signal significantly more towards microvasculature and a PLD of 500 ms allows fast flowing spin to clear. (ii) Arterial transit times (ATT) between the labeling and imaging slabs are reduced by placing the thin labeling slab close to the imaging volume. A significant amount of transit time is incurred in the large vessels when labeling at the ICA. According to the literature, the average transit time is in the range of 400–700 ms for the labeled spins traveling from the carotid arteries to the Circle of Willis (29,30). (iii) In addition to delays in the large arteries, a recent study showed an increase of 63.5 ms in ATT was encountered between slices in a multi-slice ASL acquisition when the slice thickness was 9 mm (31). In our localized pTILT, we are applying the labeling plane at a distance of approximately 4 cm closer to the imaging slice than the Circle of Willis. This potentially leads to a further reduction in ATT by 280 ms. In summary, we expect that the shorter post-labeling delay of 500 ms implemented in our localized pTILT does not compromise the signal sensitivity towards perfusion. Further experiments are necessary to confirm this behavior. This is a unique feature of localized pTILT, as pCASL and PASL methods target the large arteries either for maintaining high labeling efficiency or for producing a large labeling volume. A short PLD in these other ASL acquisitions would result in reduced sensitivity towards perfusion.

By labeling close to the tissue with localized pTILT, we would expect to gain 1.3 times the SNR of global pTILT, by reducing the PLD from 1 s to 500 ms. The minimal PLD would be limited by the quality of the RF slice profile and the minimum gap required to avoid contamination of the flow signal with static spins. The potential SNR gain becomes higher in pathologies like stroke, in which the arterial transit time for global pTILT (or other pCASL methods) can be longer than T1 relaxation time of blood (32). For example, in the case of stroke, a theoretical gain of 2.0 times higher SNR (than global pTILT) could be obtained from a reduction in the PLD from a value of around 1.68 s (T1 of blood at 3 T) to 500 ms using localized pTILT. In conclusion, for tissues with long ATTs, localized pTILT can be a good solution to reduce ATT sensitivity by labeling very close to the tissue of interest and gaining SNR due to reduced signal loss by using a shorter but reasonable PLD.

There are two other flow measurement techniques under development which are also less vulnerable to the arterial transit time effects, similar to localized pTILT. The first one is VSASL (20) proposed by Wong’s group, in which the labeling is only velocity-selective but not spatial selective. The second method is flow-enhanced signal intensity (FENSI), also proposed by our group (33), in which the labeling slice is placed in the middle of imaging slice to avoid dependency on flow history.

Future work

In our localized pTILT (10 mm labeling slice and 30 ms RF pair spacing), the labeling efficiency is high for both fast and slow spins. There is a possibility that fast spins get tagged and do not fully exchange into tissue after a short PLD of 500 ms. The resulting perfusion maps could be contaminated by intravascular signals. There are two ways to exclude the inflowing signal, one is to use bipolar crusher gradients (34), and the other specific solution for pTILT is to decrease the critical velocity by using thinner labeling slice and longer RF pair spacing. Implementation of Shinnar-Le Roux (SLR) RF pulses (35), which is well-known for providing analytical tradeoffs for slice profile performance, could further reduce slice profile errors in the imaging slab and allow for closer spacing between the labeling slice and the imaging slices.

Due to the lack of a competing pCASL sequence, we only made an approximate comparison between the SNR of pTILT and that of pCASL techniques reported in the literature. Future work will be carried out to investigate and compare these two methods in the same study.

CONCLUSIONS

In this work, we propose to make the original TILT method into a novel pCASL technique, named pseudo-continuous transfer insensitive labeling technique (pTILT). To our knowledge, pTILT is the first pCASL method that is not based on a flow-driven adiabatic mechanism. The features of pTILT include: (1) significantly reduced slice profile errors compared with TILT; (2) equivalent labeling efficiency for both fast and slow flow spins. These two features enable pTILT to label at major feeding arteries as well as at smaller vessels very close to tissues of interest. Perfusion maps obtained during both resting state and functional tasks are demonstrated with pTILT. Our studies indicate that, in terms of signal-to-noise ratio and insensitivity to transit time, pTILT’s localized labeling may be beneficial in measuring localized perfusion maps in tissues that would otherwise suffer from long arterial travel distances and long transit times.