Arterial cerebral blood volume weighted functional MRI using pseudo-continuous arterial spin Tagging (AVAST)

Abstract

Purpose:

Neurovascular regulation, including responses to neural activation that give rise to the blood oxygenation level dependent (BOLD) effect, occurs mainly at the arterial and arteriolar level. The purpose of this study is to develop a framework for fast imaging of arterial cerebral blood volume (aCBV) signal suitable for functional imaging studies.

Methods:

A variant of the pseudocontinuous arterial spin tagging technique was developed in order to achieve a contrast that depends on aCBV with little contamination from perfusion signal by taking advantage of the kinetics of the tag through the vasculature. This technique tailors the tagging duration and TR for each subject. The proposed technique, called AVAST, is compared empirically with BOLD imaging and standard arterial spin labeling (ASL) technique, in a motor-visual activation paradigm.

Results:

The average Z-scores in the activated area obtained over all the subjects were respectively 4.25, 5.52 and 7.87 for standard ASL, AVAST and BOLD techniques. The aCBV contrast obtained from AVAST provided 80% higher average signal-to-noise ratio and 95% higher average contrast-to-noise ratio compared to that of the standard (perfusion weighted) ASL measurements.

Conclusion:

AVAST exhibits improved activation detection sensitivity and temporal resolution over the standard ASL technique, in functional MRI experiments, while preserving its quantitative nature and statistical advantages. AVAST particularly could be useful in clinical studies of pathological conditions, longitudinal studies of cognitive function, and studies requiring sustained periods of the condition.

Introduction

Cerebral blood volume (CBV) is a very important measurement for the study of a number of neurological disorders that involve vascular dysregulation. These include hypertension, stroke (1,2), small brain infarcts (3), Alzheimers disease (4), astrocytoma and other brain tumors (5,6) and cerebral stenoses (7). The ability to separate the arterial compartment from the venous compartment of CBV may provide additional information not obtainable by imaging the total CBV (e.g. in steno-occlusive disease of the internal carotid artery (8)). It may also provide insights into vascular control mechanisms in both healthy and diseased tissue (6,8,9).

Cerebral perfusion is regulated at the arteriolar level. Thus, measurement of aCBV can be an essential indicator of vascular reactivity (9) and thus can also be used to study the clinical conditions during which the normal relation between cerebral blood flow (CBF), CBV and blood oxygenation is disrupted. Arterial and venous vessels show considerably different responses during neuronal activation (10,11), and thus imaging of arterial cerebral blood volume (aCBV) can also provide important information for the study of human brain function and cognition. Indeed, functional magnetic resonance imaging (fMRI) using physiological parameters such as CBF or CBV, unlike BOLD fMRI, provides a quantifiable contrast and is also more closely related to neural activity (13–15). These methods can also be much less sensitive to the local susceptibility artifacts due to the use of shorter TEs or by employing imaging techniques that are less sensitive to the field inhomogeneity (e.g. fast spin echo). This would have specific use in areas of the brain that lie near air spaces such as frontal and low brain regions which is an issue plaguing the BOLD effect (16). Furthermore, dynamic measurement of aCBV can provide critical information about the mechanism of the blood oxygen level-dependent (BOLD) contrast itself (12).

Because of these potential advantages, CBF, CBV and aCBV based functional MRI methods have recently received attention as alternative and complementary techniques to BOLD fMRI. These methods, however, generally have been hampered in terms of signal to noise ratio (SNR), temporal resolution and multi-slice imaging capability.

Different methods have been proposed for non-invasive measurement of aCBV in animals (17) and humans (18–21). Most prominently, iVASO (21) uses the difference in T1 relaxation time of blood and tissue to assess the aCBV. This method however, is based on a set of assumptions about the arterial arrival time and T1 values, which requires further verification (22). Alternatively, aCBV can also be estimated from the signal obtained from variants of arterial spin labeling (ASL) techniques. Some techniques obtain ASL images at different post labeling delays and quantify the aCBV by modeling the collected signal (18–20). Several other techniques take differencing ASL signals with and without flow-crusher gradients in order to isolate the arterial signal (18,22–24). Another technique, called MOTIVE (17), uses magnetization transfer along with ASL method to modulate the tissue and blood signals and estimate the aCBV in animals. While recent improvements in imaging and labeling techniques such as 3D GRASE readout (25), background suppression (26) and optimized labeling schemes (27–30) can reduce practical issues of these techniques and despite the fact that all these methods have been shown to be useful in various applications, many still are not practical for clinical studies or cognitive psychology experiments. These experiments are often more demanding in terms of spatial coverage and resolution as well as temporal resolution, and hence may not permit long acquisition times or additional measurements of physiological constants.

The purpose of this study is to develop a simple, novel, dynamic aCBV measurement technique based on pseudocontinuous arterial spin labeling (pCASL) (30,31) that is suitable for functional imaging experiments. We refer to our proposed measurement as Arterial Volume using Arterial Spin Tagging (AVAST). AVAST tailors the tagging duration and TR for each subject to achieve a contrast that depends on aCBV with little or no contamination from perfusion signal. AVAST also allows for a fast acquisition rate (TR ~2.2 s), making it suitable for dynamic fMRI experiments.

AVAST offers sensitivity to brain activation that is on par with BOLD imaging and superior to CBF ASL, while maintaining many advantages of CBF ASL imaging such as reduced sensitivity to susceptibility artifact, improved localization (13), quantifiable signal and reduced sensitivity to the MR scanner signal drift (32). We demonstrate AVAST using numerical simulations and in vivo experimental data. We then compare it with standard ASL and BOLD techniques in terms of sensitivity, reproducibility, locus of activation and temporal response for fMRI experiments.

Methods

Theory

Our strategy is to design the timing parameters of a continuous ASL sequence such that the signal is acquired when the tagged spins are primarily still in the arteries before they have filled the capillaries. By adjustment of the timing parameters, the contribution from tagged spins in the capillary and tissue compartments is the same in both the control and tagged images and therefore can be subtracted out. We consider a given voxel to be consisting of three separate compartments through which blood water flows: arterial compartment, capillary and tissue compartment, and a venous compartment. In this manuscript we will simply refer to the capillary and extra vascular tissue as “tissue”. In an ASL experiment, the tagged spins enters the voxel via the arterial compartment and exchanges through the capillaries/tissue on its way to the venous compartment, from which it leaves the voxel. At 3 Tesla and below, T1 decay prevents the tagged spins from retaining their tag by the time they reach the venous compartment, so it can be safely neglected.

Therefore, for our purposes, the observed ASL signal can be split into an arterial and a tissue compartment (S_ASL = S_a + S_t) and modeled separately. Arterial signal (S_a), in a continuous ASL experiment can be expressed by adapting the general kinetic model (33) for magnetization difference between tagged and control images as follows:

$$S_a\left(t\right)=\Delta M_a\left(t\right)=2\alpha .M_{a0}.\mathit{\text{aCBV}}.c\left(t\right)$$ (Eq. 1)

$$c\left(t\right)=\mathit{\text{rect}}\left(t,w\right)*[k\left(t-{\tau }_1\right).m\left(t\right)]$$ (Eq. 2)

$$k\left(t\right)=\left[\frac{S^{(1+SP)}}{\Gamma (1+\text{SP})}{t}^{SP}{e}^{-St}\right]$$ (Eq. 3)

$$m\left(t\right)=e^{-t/T_{1a}}$$ (Eq.4)

where M_a0 is the equilibrium magnetization of arterial blood and M_a(t) describes the magnetization of arterial blood at time t. aCBV is the arterial cerebral blood volume fraction in a given voxel, α is the inversion efficiency of the ASL pulse and rect(t,w) describes a rectangular tagging function of duration w. The mean time it takes for the tagged spins to reach the voxel is τ₁, or “arterial arrival time”. The bolus of tagged spins experiences longitudinal relaxation, as well as dispersion of velocities as it moves up the arterial network. m(t) represents the longitudinal magnetization relaxation function defined as the fraction of original longitudinal magnetization tag carried by the water molecules that remains at a time t after their arrival in the voxel and is characterized by a decaying exponential with time constant T_1a (T₁ decay of blood in artery). The dispersion of the tag during travel to the imaging plane can be approximated by convolving the input function with a Gamma kernel k(t) (34,35) where s and p characterize the “sharpness” and the “time-to-peak” of the kernel.

The arterial tag in turn serves as an input to the tissue compartment. The tissue signal (S_t), can be expressed as a second convolution

$$S_t\left(t\right)=S_a\left(t\right).f*\left\{e^{-\frac{f}{\lambda }\left(t-{\tau }_2\right)}.e^{-\left(t-{\tau }_2\right)T_{1t}}\right\}$$ (Eq. 5)

Here, the second term in the convolution captures the fraction, f (perfusion) of tag from the tissue after a short delay or “tissue transit time”, τ₂, and the decay of the tag’s longitudinal magnetization in the tissue compartment with time constant T_1t (T₁ decay of blood in tissue). λ is brain-blood partition coefficient for water in the whole brain (38).

Intuitively, this framework describes the process by which the tag is taken up and washes out of the arterial compartment and then the tissue compartment of a given voxel. The goal of the following simulations is to find a range of pulse sequence timing parameters that produces an ASL signal that is primarily dominated by the arterial compartment and to characterize the observed signals.

Simulations

We simulated the general kinetics of the ASL signal at different TRs and tagging durations using equations 1–5 in order to explore the effect of the pulse sequence timing parameters with respect to multiple arterial arrival and tissue transit times. We assumed that image acquisition happens instantly after the tagging period (i.e. no post-inversion delays). We used the following constants in the model (f = 0.015 ml/g/s, M_0a = 3000, aCBV = 2 ml/100 ml, T_1a = 1.6 s (36), T_1t = 1.4 s (37), α = 0.8, λ = 0.9 ml/g). M_0a is an arbitrary signal intensity representing the equilibrium magnetization of arterial blood. We also assumed inversion efficiency (α) of the tagging pulse to be 80% (28). To account for the dispersion of tagged blood, we convolved the input function with input function with the Gamma kernel presented in Eq. 3 using the parameters suggested in (35) (s=0.38, P=0.11). Arterial arrival time (τ₁) and tissue transit time (τ₂) were 1 s and 0.5 s (39).

Figure 1 shows a simulation of the ASL signal arising from the arterial and tissue compartments separately (recall that the ASL signal consists of the subtraction of a tagged image from a control image) acquired with several different timing parameters. The contribution of each compartment to the ASL signal depends on the pulse sequence’s timing parameters along with each subject’s arterial arrival time, τ₁, and local arterial delay or tissue transit time, τ₂. If the tagging duration is relatively long (>1 second) and a delay greater than the arterial arrival time is introduced between the tagging and acquisition portions of the sequence, the tag clears the arterial compartment and fills the tissue compartment. Thus, the arterial contribution to the ASL signal can be eliminated using a long post inversion delay (Figure 1.a) (40,41). Without a post inversion delay (Figure 1.b), the ASL signal is a mixture of arterial and tissue signals. Note that changing the tagging duration and the TR determines the amount of tag in the arterial and tissue compartments when the control and tagged images are collected (Figure 1.b, c and d). As can be seen in (Figure 1.c), by choosing the proper TR and tagging duration it is possible to have the same contribution from tissue in the control and tagged images. In this case the tissue signal subtracts out in the ASL signal and the remaining observed signal arises from the arterial compartment alone, which is our goal for this technique.

Figure 1.

Simulation of the kinetics of ASL the signal timeseries arising from arterial and tissue compartments in a given voxel using different timing parameters. The tagging function, arterial and tissue signals and total ASL signal time courses are shown using green, red, blue and black curves, respectively. The timing of the acquisition of first pair of control and tagged images is shown using dotted vertical lines. Tissue contribution and arterial contribution in the subtraction of tagged from control image are also indicated on the right side of each curve by blue and red arrows, respectively: (a) ASL signal in a standard CBF measurement experiment with long post inversion delay (large tissue signal, no arterial signal); (b) ASL signal without using a post inversion delay (tissue signal, large arterial signal); (c) ASL signal without using a post inversion delay with optimal tagging duration for aCBV measurement (no tissue signal, large arterial signal); (d) ASL signal without using a post inversion delay with shorter than optimal tagging duration for aCBV measurement(negative tissue signal, large arterial signal).

We used this framework to compute the contributions of the arterial and tissue compartments to the ASL signal as a function of tagging duration. Thus, we were able to identify a range of tagging durations for which the net signal is mostly arterial with small tissue contribution to the signal.

The contribution of the arterial and tissue compartments to the ASL signal as function of tagging duration is shown in Figure 2, assuming that the image acquisition happens at the end of tagging period. The highlighted points (red crosses) on the black curve in Figure 2 correspond to the timing parameters shown in Figure 1.b,c, and d. As can be seen in Figure 2, there are several tagging durations for which the tissue contribution to the arterial signal is zero. At these “zero-crossings” of the tissue contribution curve (blue), there is still ~50–75 % of arterial contribution to the ASL signal. Among these points, the location indicated in the Figure 2 by a red rectangle suggests a very stable range of tagging times for which the net signal is mostly arterial (i.e., the tissue contribution to the signal is less than 5%). Consequently, using this tagging duration, the fractional ASL signal change upon activation during a functional study is related to the local changes in the arterial blood volume. We refer to this timing combination as the “aCBV point” and use it to collect aCBV-weighted images.

Figure 2.

Simulated ASL signal (control minus tagged images) as a function of tagging duration considering no post inversion delay. Arterial arrival time assumed to be 1s. TR was adjusted dynamically to accommodate the increasing tagging duration. Points highlighted with red crosses corresponds to the kinetic curves shown in Figure 1.b, c and d.

In practical terms, the first task in the AVAST techniques is to identify the timing parameters of the aCBV point by testing a range of tagging durations with arterial suppression gradients, in order to isolate the contribution of the tissue signal. Once the aCBV point is identified in a preliminary scan, a time series of dynamic aCBV images can be collected by fixing the pulse sequence parameters to those of the aCBV point and turning off the flow suppression gradients.

Sensitivity to timing variations

The location of the aCBV point is determined by the subject’s specific arterial arrival and tissue transit times relative to the image acquisition parameters. Therefore, the best selection of parameters will vary from subject to subject and over brain regions. To evaluate the sensitivity of AVAST to these timing variations, we conducted a simulation study. We simulated the ASL signal for a case in which the aCBV sequence was calibrated for a specific region with a given set of physiological parameters (arterial arrival time of 1 sec and transit time of 0.5 sec (39)). Keeping the acquisition parameters the same, in figure 3, we calculated what happens to the signals when the voxels’ arterial arrival and tissue transit times differ up to 0.5 second from the original calibration in order to illustrate the amount of tissue contamination that can be potentially found using this method. In figure 3, we see that if we are wrong by 50% in the arterial arrival time or tissue transit time, the contribution from the tissue-tagged spins to the aCBV signal is ~ 10 %.

Figure 3.

Simulation study of the sensitivity of AVAST to the arterial arrival time and tissue transit time variabilities across different area of the brain or under different conditions (e.g activation). aCBV acquisition is optimized for a voxel where the arterial arrival time (τ₁) the tissue transit time (τ₂) are 1 and 0.5 (s) respectively. Contribution of arterial and tissue compartments in the ASL subtraction signal (control minus tagged images) is estimated in voxels whose arterial arrival time and tissue transit times differ from the parameters used for optimization by a broad range (± 0.5 s). These plots show the total (arterial + tissue) and tissue signals estimated in voxels with different arterial arrival time (left) and tissue transit time (right).

In addition to spatial variability, tissue transit time also changes during the neuronal activation. It has been shown that neuronal activation is accompanied by a local reduction in the tissue transit time of approximately 0.150+/− 0.045 s (42). As a result of this reduction, there will be a change in the total signal between rest and activation conditions. Figure 4 shows the relative change in the total signal between rest and activation as a function of the reduction in the tissue transit time during neuronal activation. Since the vasodilation happens locally at the area of activation, in this simulation we assumed that the neuronal activation mainly affects the tissue transit time and not the arterial arrival time. It is important to note that this change is due to the shift in the aCBV point and not due to the actual change in aCBV due to activation. It can be seen that there could be up to ~10% increase in the signal because of the activation-induced reduction in the transit time.

Figure 4.

Simulated change in the signal between rest and activation due to reduction in the transit time during neural activation.

Quantification

Using equations 1–4 it is possible to quantify the aCBV signal obtained using AVAST. To this end, it is necessary to estimate the arterial arrival time from the tagging plane at the neck to the imaging plane. We estimated the arterial arrival time by varying the tagging duration and TR and following the changes in the ASL signal without using flow crusher gradients (42,43). The signal can then be quantified using the following equations:

$$\mathit{\text{aCBV}}=\frac{\Delta M(TR)}{2\alpha .M_{a0}.c(TR)}$$ (Eq. 6)

$$M_{a0}=\frac{M_a(TR)}{\left(\text{sin}\theta .e^{-TR/T_{1a}}\right)}$$ (Eq. 7)

in which θ is the flip angle of 3D imaging pulse and M_a is the equilibrium magnetization of the arterial blood measured in the sagittal sinus measured from the mean of the control images.

In Vivo Experiments

3D image acquisition

We employed a 3D imaging pulse sequence with multi-shot (segmented) readouts for imaging the aCBV signal. The readout is a 3D “stack of spiral trajectories” implemented using RF-spoiling. Besides a modest increase in SNR relative to 2D (44), the 3D acquisition ensures the same acquisition delay for the entire imaging volume relative to the tagging period and therefore provides us with the capability to measure the aCBV in a volume rather than a single slice (44). If we were to use sequential 2D imaging, each slice would be acquired at a different post-inversion delay and consequently, the optimality of the aCBV acquisition would be slice-dependent.

The 3D gradient echo spiral images in this study were obtained with pulse TE=0.004 s, field of view = 24cm, slab thickness = 11cm, and matrix size = 64 × 64 × 22. Planes along the k_z- dimension were sampled starting at the center of k-space and moving outward alternating between negative and positive k_z increments using 22 excitations, each taking 0.023 s. We used a cubic flip-angle schedule from 15 to 90 degrees as we traversed k-space along the slab-select direction, in order to reduce blurring along the slice encoding dimension (44). Due to use of a low flip angle we did not consider spin history effects and assumed those to be negligible. The TR and tagging durations used for each set of experiments are explained in the subsequent sections.

aCBV calibration

Our simulations suggest that the timing parameters producing the aCBV images is determined by each subject’s specific arterial arrival and tissue transit times relative to the image acquisition parameters. Therefore, the best selection of parameters varied from subject to subject, although our simulations indicate that the technique is not very sensitive to errors in transit times. For each subject, we identified the timing parameters of the aCBV point by running a calibration sequence before the functional experiment, as follows. The calibration sequence tests the ASL signal over a range of TRs in the presence of arterial suppression gradients (similar to the blue curve in Figure 2). Arterial signals were suppressed by using a pair of bipolar crusher gradients (b = 4 s/mm2) along the Z-direction providing a cut off velocity of 1cm/s. Guided by the simulation result and by previous experience (42,43,45), TR was varied from 1.5 to 2.5 s, steps of 0.1 s, while keeping the tagging duration 0.5 s shorter than TR to allow the acquisition of 22 slices using for the calibration sequence. For each TR we collected 8 tag/control images and averaged them together during the post processing. Total scan time for the aCBV calibration using these parameters was 352 s.

The images were reconstructed, pairwise subtracted and averaged for each TR. The TR and tagging duration yielding the lowest ASL signal in the presence of flow crusher gradients, was selected as optimal for aCBV (i.e. aCBV point highlighted with a red box in Figure 2).

Functional experiments

Using the obtained aCBV parameters, we carried out a functional aCBV experiment and compared it to the result of functional CBF and BOLD experiments. Ten subjects were scanned using a 3.0 T Signa Excite scanner (General Electric, Waukesha, WI). We employed pseudocontinuous tagging pulses (30,31) for functional CBF and aCBV experiments. The tagging efficiency of the pCASL pulse sequence was optimized by correcting for the field off-resonance using the method proposed in (28). Image acquisition was carried out employing the above-mentioned segmented 3D spiral acquisition pulse.

Subjects performed a simultaneous motor (sequential finger opposition) and visual (8Hz flashing checkerboard) activation task while being scanned using BOLD weighted imaging, perfusion weighted ASL, and AVAST, as described below. The activation paradigm consisted of five cycles of alternating rest (30 seconds) and activation (30 seconds). Total functional scan time for all methods was 300 s.

The first activation map was obtained using the functional aCBV scheme of AVAST (i.e. TR/tagging duration obtained from the calibration scan, no post inversion delay). The second one was carried out using a functional CBF scheme employing standard ASL technique (TR = 4 s, tagging duration = 2 s, post-tagging delay = 1.5 s). A BOLD fMRI study was also conducted (Single shot gradient-echo reverse spiral pulse sequence (46), TE=0.03 s, TR = 2 s, flip angle = 90 degree, same FOV, slice location and resolution as aCBV and CBF experiments) for comparison purposes. Four dummy scans were collected in the beginning of all functional scans to make sure that steady state was reached. To compare the reproducibility of each method, we repeated the functional experiments on the same day, using each method for all subjects. All datasets were reconstructed, surround subtracted and analyzed by estimation of standard general linear model (GLM) using home written software FASL01 (http://fmri.research.umich.edu/resources/software/shared_code.php).

SNR and CNR calculations

We characterize our functional experiment by a linear model as previously described in (47,48).

$$y_t={\beta }_0+{\beta }_1{x}_{1t}+{\epsilon }_t$$ (Eq. 8)

for each time point t=1,…,n. The first regressor and its coefficient parameter, ${\beta }_0$, indicate the baseline signal for the BOLD experiment and baseline subtraction ASL signal for the aCBV and CBF experiments. The second regressor x_1t with its regression coefficient, ${\beta }_1$, describes the signal (BOLD or subtraction ASL signal) difference due to activation and thus the amplitude of ${\beta }_1$ is indicative of contrast of the signal between rest and activation. The square root of the estimated variance of the effects of interest (${\sigma }_{{\beta }_0}$ and ${\sigma }_{{\beta }_1}$) also corresponds to noise. Therefore temporal SNR and CNR of the functional experiment conducted using each method can be calculated using Eq.9 and Eq.10.

$$SNR=\underset{{\mathit{\text{v}}ϵ\mathit{\text{V}}}_{\mathit{\text{brn}}}}{\text{Mean}}\left(\frac{{\beta }_{0v}}{{\sigma }_{{\beta }_{0v}}}\right)$$ (Eq. 9)

$$CNR=\underset{{\mathit{\text{v}}ϵ\mathit{\text{V}}}_{\mathit{\text{act}}}}{\text{Mean}}\left(\frac{{\beta }_{0v}}{{\sigma }_{{\beta }_{0v}}}\right)$$ (Eq. 10)

in which, V_brn and V_act respectively represent total brain voxels and activated brain voxels in each experiment.

Results

aCBV Calibration

Figure 5 shows the ASL signal acquired at different TRs over a wide range of tagging durations (0.6–3 s) with and without the flow crusher gradients in an ROI containing 50 voxels of gray matter around the motor cortex area from one of the subjects. The black curve corresponds to the ASL signal without the flow crusher gradients and contains both arterial and tissue signals. The blue curve corresponds to the ASL signal with the flow crusher gradients and therefore it arises from the tissue signal (CBF). The tissue curve was used to estimate the optimal aCBV parameters by finding the aCBV point (pointed to by the red arrow). In agreement with our simulations (see figure 2), by incrementing the tagging duration and TR together, we were able to isolate the arterial contribution to the ASL signal. Note the temporal shift of the tissue-only ASL signals relative to the whole signals (arterial plus tissue). This is to be expected since these spins must traverse the branching arteries and arterioles before they reach the capillary bed where they can exchange. This delay corresponds to tissue transit time in our model. The range of TRs for the aCBV point estimated for our subjects was from 2.0 to 2.5 s (mean/std=2.26/1.8 s). Also note that, in practice, only the tissue curve over a narrower range of tagging durations (1.5–2.5 s) is necessary to conduct an AVAST experiment.

Figure 5.

Mean ASL signal measured for different tagging durations with (blue) and without (black) flow crusher gradients. For each measurement, TR was also adjusted to accommodate the increasing tagging duration.

Functional Study Results

All experiments produced activation in visual cortex, motor cortex and supplementary motor area (SMA) using a Z-threshold of 5 (Z>5). Figure 6 shows the activation maps from a representative subject obtained using standard ASL, AVAST and BOLD fMRI overlaid on the corresponding average image. Active active areas in the AVAST map were more focal (less scattered) than BOLD in all subjects. AVAST also produced wider activated areas compared to standard ASL.

Figure 6.

Orthogonal slices illustrating activation maps (Z>5) and their corresponding signal time course for a single representative subject obtained using: (a) BOLD, (b) Standard ASL (CBF contrast) and (c) AVAST (aCBV contrast) during visual and unilateral motor stimulation. Each activation map is overlaid on the average image corresponding to the technique used to generate that map. Time courses are averaged over the top 1% of the activated voxels with the highest Z-scores for each technique. Error bars indicate the standard deviation across top 1% activated voxels. Dark lines under the time courses show periods of stimulation.

We compared the hemodynamic response of these methods by averaging the time course of the top 1% of voxels with the highest Z-score for each technique across the 5 cycles and subjects (N=10). It is important to note that each method measures different physiological parameters and therefore the locus of the activation area detected using each method is different. Thus, to compare the CBF, aCBV and BOLD time courses, it was not possible to choose an anatomical ROI that fairly represented the activated area for all three methods. Moreover, since these methods have quite different sensitivities, comparing the results by choosing a fixed Z threshold leads to different number of voxels for comparison and will not be statistically equivalent for all methods. Therefore, we compared the most significant voxels of each method by choosing the voxels with the top 1% Z-scores for each method. Additionally, since TR for AVAST experiments is not the same for different subjects, all aCBV time courses were first linearly interpolated to TR=2s before averaging.

Figure 7 shows the comparison of the hemodynamic response obtained using BOLD, Standard ASL (CBF weighted contrast) and AVAST (aCBV weighted contrast) methods. It can be seen that the CBF and BOLD responses have a delayed time to peak and slower return to baseline than aCBV. During post-stimulus period. A post stimulus undershoot in the aCBV time courses was observed, but it returned to the baseline faster than the BOLD signal. The average signal changes relative to the BOLD signals of aCBV and CBF were 85% and 46% respectively

Figure 7.

Average hemodynamic response averaged across 5 cycles for all subjects (N=10) for aCBV, CBF and BOLD fMRI techniques. Error bars indicate the standard devation across subjects. Dark lines show the period of stimulation.

To compare the sensitivity of each method, we used the mean of top 1% Z-scores again. Figure 8 shows the calculated mean Z-scores for all ten subjects. Average Z-scores for all the subjects were 4.25, 5.52 and 7.87 for standard ASL, AVAST and BOLD, respectively, indicating the superior activation detection sensitivity of AVAST compared to standard ASL.

Figure 8.

Distribution of Mean Z-score values averaged over the top 1% voxels for BOLD, AVAST and standard ASL for all subjects.

In order to understand the spatial relationship between activation maps obtained with these methods in terms of locus of activation, we examined the overlap of the top 1% of the activated voxels between AVAST-BOLD, ASL-AVAST and ASL-BOLD. Figure 9.a shows the calculated number of overlapping voxels for all ten subjects. The average overlaps over all subjects were 54%, 36% and 46% respectively, which was expected considering the different physiological origin of each signal. In order to compare the reproducibility of each technique, we also calculated the overlap of the top 1% of the activated voxels between two runs for each method (Figure 9.b). The results indicate that AVAST method has superior reproducibility compared to standard ASL but still is outperformed by BOLD.

Figure 9.

Number of overlapping activated voxels detected using the three methods, AVAST, standard ASL and BOLD: (a) between two runs of different methods (b) between two runs of the same method.

CNR and SNR of each method were calculated using Eq. 9 and 10 and averaged over all subjects (Table 1). It is important to note that the total scan time is the same for all techniques. However, since these techniques have different TRs, the number of measurements is not the same for these measurements. To account for the differences in TR, we also calculated SNR and CNR of a single measurement for the techniques. Our calculations indicate that AVAST outperformed the standard ASL in terms of CNR and SNR, both in single measurement and the whole scan. Due to significantly better temporal resolution of AVAST compared to ASL, this difference in more pronounced in the whole scan. BOLD however provides the best CNR and SNR among all techniques. Note that the subtraction of control and tag images in AVAST and ASL experiments lead to much smaller signal values compared to BOLD signal that cause a significant lower SNR for AVAST and ASL.

Table 1.

Estimated SNR (in the whole brain) and CNR (in the activated area) averaged over all subjects for each method. CNR and SNR of a single measurement are also presented. For AVAST and Standard ASL, one pair of tagged and control images considered as a single measurement.

	AVAST	Standard ASL	BOLD
CNR	1.7	0.9	3.2
SNR	12.9	6.6	1594
CNR of single measurement	0.026	0.024	0.021
SNR of single measurement	0.195	0.178	10.626

aCBV Quantification

Figure 10.a shows an example of aCBV maps estimated in the resting state using AVAST. The aCBV values over the brain were 1.1+/−1 mL/100 mL. Figure 10.b shows the histogram of the obtained values in GM. The large spread in GM aCBV values is likely related to different partial volume effects with vessels, white matter and CSF.

Figure 10.

Representative aCBV map estimated using AVAST (a) and its corresponding histogram of aCBV values in Gray matter (b) for a representative subject.

Discussion

Inspecting both simulated and experimental kinetic curves (figure 2), one may notice that there are two other zero crossing points (~ 0.7 and ~1.1 s in figure 2) which can potentially be used for obtaining aCBV weighted signal. Although shorter TR of these two points are very desirable, rapid changes in total ASL around these 2 points would make them very sensitive to transit timing variations. For that reason the third point was used in this study. In this regime, the transit time variations lead to small contamination from the tissue signal, since the rate of change in of the ASL signal with respect to transit time is much slower (aCBV point).

To attain an aCBV-weighted signal with minimum contribution from the BOLD effect, the shortest TE possible is used in AVAST. While at the TE of choice in our experiment the sensitivity of the BOLD signal is approximately less than 10% of that of the typical BOLD experiments (49) it may still contribute to the error of the measurement. If this contribution is severe, one can include the expected BOLD response in the general linear model as a nuisance regressor, as in (11,50), and eliminate any possible BOLD contribution from the analysis. In this study, however, we have chosen to keep preprocessing to a minimum in order to show a fair comparison among techniques.

The model for ASL signal used in this study is a well-established model that frequently has been used and tested in the literature (33). Although more complicated models for ASL signal have been proposed in the literature (35,51,52), the consistency of our experimental results with the simulation results indicate that the complexity of model we used for our simulation is appropriate for this application.

AVAST requires a calibration scan before the fMRI scan. Once aCBV parameters found, however AVAST does not need multiple measurements during the actual fMRI experiment, which is a critical advantage of this method for fMRI applications compared to other ASL based techniques proposed for aCBV estimation (18–20,22–24). One drawback of estimating the aCBV point by finding the zero-crossing point is the lower SNR of measurements around the zero crossing points. We circumvented this problem by averaging 8 tag/control images for each TR. Another solution could be to collect a larger range of TR’s on the calibration curve and estimating the aCBV point from a fit of the above kinetic model.

A new variant of VASO (53) dubbed iVASO has recently been proposed in order to isolate the arterial component of CBV (21). By inverting the spins in a region outside the volume of interest and collecting the images at the time that the arterial blood’s magnetization is at its null point, iVASO limits the blood nulling to the incoming blood spins, which leads to an increase in SNR and CNR compared to VASO. AVAST and iVASO pulse sequences are very similar in terms of tagging the inflowing blood. The differences are that AVAST uses continuous tagging while iVASO uses a pulsed tagging scheme and, most importantly, the contrast in iVASO is based on the nulling of arterial blood while AVAST measures the amount of label in the arteries. However the use of iVASO in its current implementation faces a number of challenges for fMRI applications including slice coverage, CSF contamination and, spatial variations in inversion efficiency and tissue/blood T1. Some of these issues have been solved in recent improvements of VASO technique (54) and can potentially be implemented for iVASO as well. Having said that, AVAST does not face these problems. Since AVAST tailors the aCBV parameters for each subject, it is potentially more reliable compared to iVASO. However, a thorough, empirical, side-by-side comparison is required to assess the advantages and disadvantages of these two methods fairly and that is beyond the scope of the present article.

There is great variability in the literature regarding the relative aCBV change arising from cerebral activation. Yan et al. (20) reported 67.2% and 49.7% aCBV signal change using FAIR and QUIPSS II techniques respectively. Brookes et al. (19) reported 33% change in aCBV singal measured using LL-EPI-STAR technique. The range of values reported by Hua et al. using iVASO technique was 58±7% (21). Lee et al. (10) reported 79% change in the aCBV signal. While these results are lower than the numbers we report in this study (85%), they are within error of our results. Having said that, these discrepancies can be mostly explained by the transit time effects discussed earlier and, to a lesser degree, by different contributions from the capillary signals in each method. Differences in spatial resolution, species, presence or absence of anesthesia, stimulation paradigm and age could also contribute to the discrepancy between our observations and previous works.

It should be noted that in both the present study as well as the mentioned previous works, the relative CBF and aCBV signal changes are different than resting state CBF-CBV coupling. This is to be expected because arterial and venous vessels have been shown to have considerably different responses during neuronal activation and vascular challenges (e.g. hypercapnia challenge) (10,55).

Although it has been shown that activation sites detected using aCBV weighted signal are in general more localized in cortical layers (13), the spatial resolution of our study does not allow for investigating the exact location of the activation sites with respect to cortical layers and confirm this hypothesis.

The challenges to AVAST, as currently implemented, are its reduced slice coverage (compared to BOLD) and the need to acquire a resting-state calibration curve to determine the optimal sequence timing parameters. Also, reduction in the local transit time, can muddle quantification of aCBV signal during activation periods because it exaggerates the observed signal change as the result of less T1 decay. As an example, if the total transit time is reduced from 1500 to 1300 ms without an increase in aCBV, the observed signal will be ~10% greater. While this is detrimental for the quantification process, it benefits the statistical power of detecting active regions. This effect is similar to our previous observations in measuring CBF using Turbo-CASL technique (43). This effect introduces the need for a correction scheme for the overestimation of aCBV signal change during activation. One possible solution for this problem could be an approach similar to that used in (43). Considering these facts, the application of AVAST at this current stage is confined to quantification of aCBV at rest, and detecting activation-induced aCBV changes qualitatively

The reduction in the arterial arrival time during neural activation is also accompanied by an increase (10–20%) in the local mean blood velocity. If this local change were to lead to a considerable change in carotid arteries at the tagging plane it would cause a reduction in the tagging efficiency of the pCASL technique (56). In that case, this reduction in the tagging efficiency needs to be taken into account for quantification of the results. However, since the vasodilation happens locally at the area of activation it is very unlikely that the neural activation would cause any detectable change in the blood velocity at the tagging plane. Also notable is that the tagging efficiency term in the model is multiplicative, meaning that changes in the efficiency would affect the amplitude of the curve, but would not shift the aCBV point.

It should also be noted that we employed a 3D stack-of-spirals acquisition scheme in order to eliminate differences in sensitivity to transit times across slices. Although it is multishot sequence, since it applies the shots along the K_z direction, it does not lead to a slice-timing shift. Instead it produces minor blurring along the z-axis due to T1 effects and also due to inflow spins. We minimized the blurring effects by employing a cubic flip-angle schedule from 15 to 90 degrees, as demonstrated in (44).

Conclusion

We have presented a new method, called AVAST that uses a pseudo-continuous labeling scheme to provide an aCBV weighted signal suitable for functional imaging experiments. Our initial results, presented in this article, suggest that AVAST provides superior activation detection sensitivity and temporal resolution over the standard perfusion-weighted functional ASL methods, closer to those of the BOLD technique, while retaining the more desirable properties of ASL techniques. Namely, it yields a readily quantifiable physiological parameter, and because it is a subtraction technique, its noise is mostly white and not auto-regressive. Furthermore, minimal echo times can be employed in this scheme, resulting in images that are less sensitive to susceptibility artifacts compared to BOLD weighted imaging. AVAST can be particularly useful in longitudinal studies of cognitive function and studies requiring sustained periods of the condition or state of interest. This technique can also have a central role in the study of areas of the brain that lie near air spaces such as frontal and low brain regions which is quite challenging using BOLD effect. The proposed method can also be helpful in study of the pathological conditions during which the normal relation between CBF, CBV and blood oxygenation is not valid.