Velocity-selective arterial spin labelling bolus duration measurements: Implications for consensus recommendations

Arterial spin labelling (ASL) MRI provides a non-invasive measurement of perfusion. It is being used to further our understanding of physiological mechanisms in both the brain (Haller et al., 2016; Liang et al., 2013; Mattsson et al., 2014; Satterthwaite et al., 2014; Segerdahl et al., 2015; Staffaroni et al., 2019) and body (Taso et al., 2023). ASL consists of labelling water protons in arterial blood and waiting for the labelled water to perfuse into the tissue of interest before imaging. The original and predominant ASL method labels water protons based on spatial position, outside of the imaging volume (spatially-selective ASL) (Alsop et al., 2015; Dai et al., 2008; Detre et al., 1992; Edelman et al., 1994; Kim, 1995; Williams et al., 1992; Wong et al., 1998). This approach is dependent on waiting long enough for the labelled blood to flow from the labelled arteries into the tissue of interest, a time defined as the arterial transit time (ATT).

Subsequently, an alternative ASL method was proposed, termed velocity-selective ASL (VSASL), whereby water protons are labelled based on their velocity, rather than their spatial position (Wong et al., 2006). The main advantage of this approach is that blood is labelled much closer to the tissue of interest, significantly reducing or eliminating the ATT (Qin et al., 2022). This insensitivity to ATT makes VSASL a good candidate for studying situations where ATT is long, or may change, which would result in a biased measurement of perfusion for spatially-selective ASL. Long arterial transit times have been observed in the ageing brain (Mahroo et al., 2024; Scheel et al., 2000), in vascular pathologies (Bokkers et al., 2009; Bolar et al., 2019; Fan et al., 2017; Qiu et al., 2012; Wang et al., 2013) and in gliomas (Qu et al., 2022). Changing ATT is also a source of bias in cerebrovascular reactivity (CVR) measurements made with spatially-selective ASL (Zhao et al., 2021). Therefore, there is a large range of applications that could benefit from the ATT insensitivity of VSASL.

VSASL consists of a velocity-selective labelling module followed by a vascular crushing module. The velocity-selective labelling module labels spins moving above a cut-off velocity (v_cutoff). After a period where the labelled arterial blood spins perfuse into tissue, a vascular crushing module suppresses signal from spins moving above v_cutoff, leaving the signal from perfused spins which decelerated as they perfused into the tissue. Here, we define the time between the velocity-selective label and the vascular crushing module to be the Label-to-Crusher Time (LCT). The labelling module can either be formed of a train of saturation or inversion pulses. Labelling efficiency is higher with the inversion method and the SNR approaches that of pCASL (Qin et al., 2022). A Fourier transform velocity selective inversion (FT-VSI) labelling method (Qin & van Zijl, 2016) was chosen for this study, with FT-VSI recently demonstrating good test-retest reliability for measuring CBF in healthy participants (Xu et al., 2023). A recent consensus paper recommended LCT = 1.4 s and v_cutoff = 2 cm/s (Qin et al., 2022).

The bolus duration is the time from when the leading edge of the labelled blood enters the tissue of interest until the trailing edge of the bolus arrives. As no further fresh labelled blood arrives, the kinetic curve decays according to T₁. Models used to quantify perfusion with a single LCT often assume that the bolus duration is longer than the LCT. If the bolus duration is shorter than the LCT, then these models will overestimate the duration of the labelled bolus and thus underestimate perfusion. Therefore, knowledge of bolus duration will improve specificity of perfusion measurements and will help to inform choice of LCT based on the bolus duration of the population to be studied. Bolus duration has been measured previously (Guo & Wong, 2015) as 2.03 ± 0.08 s using velocity-selective saturation ASL in five young healthy volunteers. However, that study did not report the spatial characteristics of bolus duration or the influence of elevated CBF, as observed in some pathologies with fast flow or during cerebrovascular reactivity (CVR) measurements. A recent study raises questions over bolus duration (Xu et al., 2024), with FT-VSI ASL-based measurements of cerebrovascular reactivity (CVR) being underestimated by 40% when an LCT of 1.52 s was used, but when a shorter LCT of 0.5 s was used CVR was much closer to a phase-contrast based CVR measurement. The observed underestimation in CVR is consistent with the bolus duration being shorter than 1.52 s, especially during hypercapnia, where flow velocity increases. Previous studies generally consider bolus duration to be equivalent to LCT (the time between labelling and vascular crushing modules), implying that the leading edge of the bolus arrives in the tissue instantaneously, or with minimal delay, and that the trailing edge of the labelled bolus is cut off by the vascular crushing module. In this context, LCT and bolus duration are used interchangeably and are often referred to as $τ$⁠. This assumes that the trailing edge does not reach the tissue of interest before the vascular crushing module. The present study aims to measure bolus duration in healthy participants at rest and during sustained hypercapnia and place these measurements in the context of previous studies and recommended values of LCT.

Data were acquired on a Siemens 3T Prisma MRI system with body transmit and a 32-channel head receive-only coil. An MPRAGE T₁-weighted anatomical scan was acquired for image registration and tissue segmentation (TR/TE/TI = 2100/3.19/850 ms; flip angle 8°; 1.1 x 1.1 x 1.0 mm, sagittal orientation (224 x 224 x 176 matrix); phase partial Fourier 7/8).

A Fourier transform velocity selective inversion (FT-VSI) ASL sequence (Qin & van Zijl, 2016) was implemented with regional pre-sat (2 s pre-delay) using the water suppression enhanced through the T₁ effects (WET) technique (Ogg et al., 1994), and a sym-BIR-8-based vascular crushing module. v_cutoff was defined as per Qin et al. (2022) and values of 2, 3 and 4 cm/s were used in Experiment 1, while 3 cm/s was used in Experiment 2. The FT-VSI pulse train was 62 ms in length with composite refocusing pulses, velocity-compensated control, and a 7.8 ms gap between each 20° hard pulse; the rise/fall time for the gradients in the FT-VSI pulse train was 0.35 ms. The gradient amplitude was altered linearly to achieve the desired v_cutoff, with an amplitude of 18.75 mT/m used to achieve a 2 cm/s v_cutoff in the inferior-superior direction.

Three background suppression pulses were timed to ensure the perfusion contrast was proportional to the magnitude subtraction between tag and control conditions at all LCTs, avoiding the need for complex image subtraction. The timings of the background suppression pulses were interpolated from the results presented in Liu et al. (2020), and inversion times are given in Table 1.

A 2D EPI readout was used, with 3.3 x 3.3 mm in-plane resolution (64 x 64 matrix), 15 slices (6 mm thickness, 3 mm slice gap), TE = 12 ms, and 6/8 partial Fourier. TR was set as short as possible, ranging from 3.5–6.3 s from the shortest to longest LCT. A 50 ms delay was applied between the vascular crushing module and the readout to reduce eddy currents, and the post-label delay increased by 42.5 ms for each ascending slice. Four tag/control pairs were acquired for each LCT with an M₀ acquired at the start of each set. LCT values will be specified in each respective experimental section below.

The M₀ included the sym-BIR-8-based vascular crushing module to cancel out T₂ weighting in the perfusion quantification. The first M₀ in each block was preceded by a pause of more than 30 s, so it was free from T₁ weighting and was used for quantification.

Thirteen healthy participants (8 female, 5 male; aged 18–52, median 22, mean ± stdev 28 ± 11 years) gave written informed consent. The School of Psychology, Cardiff University Ethics Committee approved this study.

To calculate bolus duration, 11 LCT values of [0.55, 0.66, 0.8, 0.96, 1.15, 1.39, 1.67, 2, 2.4, 2.9, 3.5] s were acquired in a total time of 7:20. This was repeated for three v_cutoff values of 2/3/4 cm/s, with the v_cutoff order balanced across participants (as there were 13 participants, 5 participants had an order 2/3/4, whereas 4 participants each underwent orders of 3/4/2 and 4/2/3 cm/s).

Nineteen age-matched healthy participants (10 female, 9 male; aged 18–45, median 22, mean ± stdev 26 ± 9 years) gave written informed consent. The School of Psychology, Cardiff University Ethics Committee approved this study. The sample size for Experiment 2 was powered a-priori based on the mean grey matter bolus duration measurements from Experiment 1 and anticipating a reduction in bolus duration of 0.1 s with hypercapnia (based on a similar reduction in bolus arrival time measured in pCASL and PASL previously) (Donahue et al., 2014; Ho et al., 2011; Su et al., 2017). For an effect size of 0.7, alpha = 0.05, and power = 0.8, a total sample size of 19 was required (G*Power 3.1).

The gas delivery system used was introduced previously by our group (Germuska et al., 2016), with mass flow controllers (MKS Instruments, Wilmington, MA, USA) delivering either 25 L/min of medical air (normocapnia) or 5% CO₂ (hypercapnia). Participants wore a close-fitting face mask (Quadralite, Intersurgical, Wokingham, Berkshire, UK) attached to a breathing circuit following the design of Tancredi and colleagues (Tancredi et al., 2014). Exhaled gas was monitored throughout, with a sampling line from the face mask to a rapidly responding gas analyzer (PowerLab, ADInstruments, Sydney, Australia).

To calculate bolus duration, a set of eight LCT values of [0.55, 0.66, 0.8, 0.96, 1.15, 1.39, 1.67, 2] s were acquired in a total time of 4:45. This set was repeated four times, [normocapnia – hypercapnia – normocapnia – hypercapnia]. To increase participant compliance, the amount of time a participant was at hypercapnia in each set was minimised by omitting the last three LCT values (2.4, 2.9, and 3.5 s) acquired in Experiment 1. Each set only started once end-tidal PCO₂ stabilised to a steady state (~1 min following the transition), such that each condition lasted approximately 6 min. An M₀ image was acquired at the start of each set, more than a minute after the preceding acquisition, so it was free from T₁ weighting. A 3 cm/s v_cutoff was chosen to reduce cerebrospinal fluid (CSF) signal contamination, which is a potential problem for hypercapnic acquisitions, where global vasodilation displaces CSF volume (Blockley et al., 2011; Thomas et al., 2013).

FT-VSI ASL data were concatenated across LCT values for each dataset, then motion correction was performed (MCFLIRT, FSL (Jenkinson et al., 2002)). Bolus duration and CBF were estimated using a two-compartment perfusion model adapted from Buxton et al. (1998) (‘general kinetic model’), assuming instantaneous arrival of the bolus in the tissue (Equation 1). An additional macrovascular component, as per Chappell et al. (2010), was included to account for any remaining signal in the arterial or venous vessels that is not completely removed by the BIR-8 vascular crushing module (Equation 2). Model fitting was performed in MATLAB via unconstrained minimisation (Quasi-Newton method). A multi-start implementation (with multiple starting values for bolus duration and macrovascular bolus duration) was used to reduce the likelihood of fitting to a local minimum.

Equation 1

$τ$ = bolus duration when $τ≤LCT$⁠, otherwise $τ=LCT$

Equation 2

Where ∆M is the signal difference between the label and control images. $τ$ = LCT for the case where LCT $≤$ bolus duration. However, $τ$ = bolus duration in the case where LCT > bolus duration (i.e., the image is acquired after arrival of the trailing edge of the labelled bolus). $τmacro$ is the macrovascular bolus duration, and macroBV is the apparent macrovascular blood volume fraction (blood volume signal remaining after application of the vascular crushing module). t is the time between labelling pulse and readout (i.e., LCT + post-label delay for each slice), f is the cerebral blood flow (ml/g/s), α is the inversion efficiency (0.56), α_BGS is the additional signal attenuation caused by the three background suppression pulses (0.86), $λ$ is the blood-tissue-partition coefficient (0.9 ml/g), M₀ is the signal intensity of the proton density image, T_1,blood = 1.6 s is the T₁ value of blood (Lu et al., 2004), and tissue T₁ = 1.3 s (Wansapura et al., 1999).

M₀ scans were coregistered to the MPRAGE using FSL epi_reg (Greve & Fischl, 2009; Jenkinson et al., 2002), and the MPRAGE was coregistered to MNI152 space (Fonov et al., 2011) using FSL FNIRT (Jenkinson et al., 2012).

Grey matter maps were segmented using FSL BET (Smith, 2002) and FAST (Zhang et al., 2001) on the MPRAGE. Grey matter partial volume estimate maps were realigned into native ASL space. All grey matter analyses were performed using a mask of voxels with a grey matter partial volume estimate of at least 50%. An additional threshold was applied to the M₀ of greater than 50% of the maximum signal to avoid areas where the EPI signal had significant signal drop-off (intra-voxel dephasing close to air spaces).

The MNI Structural Atlas (Mazziotta et al., 2001), which is provided in FSL in MNI152 space, was realigned into native ASL space. Regional analyses were performed within the grey matter mask defined in the previous paragraph by defining each region based on a 25% threshold of that region’s partial volume estimate (Craddock et al., 2012). All region analyses were calculated by taking the median across voxels. Summary data were reported as mean and standard deviation across participants.

Group average bolus duration and CBF maps were calculated in MNI152 space by realigning parameter maps into MNI152 space at 2 mm isotropic resolution, then taking the mean across participants. For physiological consistency, estimates of bolus duration and macrovascular bolus duration were assigned a value of zero when the respective CBF or macrovascular blood volume estimates were zero or less.

In Experiment 1, results are first presented in the context of how bolus duration can inform the choice of sequence timings (LCT, the labelling to vascular crushing delay time) at rest, presenting data from all three v_cutoff values in parallel. Secondly, the effects of small changes in v_cutoff are considered in the context of designing CVR experiments. Finally, results are presented from Experiment 2, examining the impact of hypercapnia on the choice of sequence timings for CVR experiments or in populations with fast flow.

Figure 1 shows group-averaged maps of ∆M in MNI space for each LCT acquired with a v_cutoff of 2 cm/s (data for 3 and 4 cm/s can be found in Supplementary Figure S1).

Fitting the kinetic model of Equations 1 and 2 to the average grey matter signal reveals a bolus duration that is shorter than the maximum LCT. Figure 2 shows example model fits for three subjects, demonstrating short (1.73 s), medium (2.20 s), and long (2.61 s) bolus durations. In many subjects, there is evidence of partial voluming with a macrovascular blood signal with a shorter macrovascular bolus duration, suggesting incomplete suppression of macrovascular blood by the BIR-8 crushing module. There was no discernible evidence of a bolus arrival time in grey matter, consistent with the idea that the leading edge of the velocity selective bolus is created close to the tissue. However, we were unable to verify this observation in white matter due to the significantly reduced SNR.

Averaging voxelwise fits across grey matter and participants provides bolus duration estimates (mean ± standard deviation across participants) of 2.20 ± 0.35 s/2.22 ± 0.53 s/2.05 ± 0.34 s at rest (2/3/4 cm/s v_cutoff). Grey matter CBF values were 57 ± 8 ml/100 g/min/56 ± 8 ml/100 g/min/54 ± 9 ml/100 g/min (2/3/4 cm/s v_cutoff). The observed macrovascular blood volume was, on average, less than 0.2% with a macrovascular bolus duration of 1.22 ± 0.17 s/1.22 ± 0.11 s/1.21 ± 0.13 s (2/3/4 cm/s v_cutoff). All participants’ grey matter CBF, bolus duration, macrovascular blood volume, and macrovascular bolus duration, with median and interquartile range are shown in Figure 3. Where macrovascular blood volume was zero, it was not possible to characterise macrovascular bolus duration, so these voxels were not included in grey matter averages.

Bolus duration maps in Figure 4A show spatial heterogeneity, with anterior and temporal regions having shorter bolus duration than posterior regions. The low white matter macrovascular bolus duration values in Figure 4B are an artefact of being unable to characterise macrovascular bolus duration in voxels with zero macrovascular blood volume. Macrovascular bolus duration was set to zero for voxels with zero macrovascular blood volume, before transformation to group space and averaging across participants. Maps of macrovascular blood volume (Fig. 4D) are mostly consistent with regions with known macrovascular anatomy. However, in contrast to PASL experiments (Chappell et al., 2010), the maps also show some venous contrast, which is likely associated with venous outflow during the 2D readout.

The spatial heterogeneity is further investigated by subdividing the brain into 9 MNI regions (Mazziotta et al., 2001), including the four brain lobes, cerebellum, and four subcortical structures. The shortest bolus duration was found in the caudate, 1.58 ± 0.24 s/1.60 ± 0.34 s/1.77 ± 0.51 (2/3/4 cm/s v_cutoff), while the longest bolus duration was observed in the thalamus, 2.35 ± 0.59 s/2.75 ± 0.76 s/2.24 ± 0.57 (2/3/4 cm/s v_cutoff). Table 2 summarises the regional parameter estimates for both Experiments 1 and 2 (which is discussed in detail in Section 3.3). All regions in Experiment 1 have a bolus duration longer than the consensus recommended LCT = 1.4 s (Qin et al., 2022), suggesting that the consensus recommendation is, indeed, appropriate for the cohort investigated at rest (normocapnia).

Reviewing the above results in the context of minimising signal contamination from CSF spaces, Figure 1 and Supplementary Figure S1 demonstrate that the apparent perfusion signal from CSF spaces is reduced from 2 cm/s to 3 cm/s v_cutoff and reduced to a lesser extent from 3 cm/s to 4 cm/s v_cutoff. While Figures 3–4 suggest that if there are any differences in CBF and bolus duration measured at each v_cutoff, then they are small compared to intra- and inter-participant variation. A one-way ANOVA testing for a change in grey matter bolus duration across v_cutoff was not significant, F(2,36) = 0.72; p = 0.50. Therefore, for CVR experiments, where CSF spaces are apparently displaced by vasodilation, a higher v_cutoff of 3 cm/s was chosen. This is higher than the consensus recommendation 2 cm/s (Qin et al., 2022) but is likely to be more appropriate in this context.

The hypercapnic challenge caused an 8.3 ± 2.4 mmHg increase in P_ETCO₂. Figure 5 shows group-averaged maps of ∆M in MNI space for each LCT acquired at rest and during hypercapnia.

Fitting the data to the kinetic model (Equations 1 and 2) produced an estimated grey matter CBF increase of 45 ± 23% during hypercapnia and a CVR of 6.7 ± 6.2%/mmHg (see Table 2). Figure 6 shows that the median increase in perfusion signal with hypercapnia is reduced for longer LCTs. Visual inspection of the CVR maps suggests a large underestimation of CVR when using an LCT above approximately 1.2 s. GM CVR across LCTs was 6.8 ± 11.9, 6.2 ± 5.2, 6.7 ± 6.0, 6.8 ± 11.3, 6.9 ± 10.4, 5.0 ± 7.3, 2.8 ± 5.5, and 3.4 ± 5.8%/mmHg for LCTs 550/660/800/960/1150/1390/1670/2000 ms. Thus, we find that CVR is reduced by approximately 25.4%, 58.7%, and 50.0% for LCTs of 1390 ms/1670 ms/2000 ms compared to LCTs from 550 to 1150 ms.

The grey matter bolus duration measured in Experiment 2 was shorter both at rest (normocapnia) 1.65 ± 0.20 s and during hypercapnia 1.43 ± 0.24 s (see Table 2). The shorter bolus duration at rest is undoubtedly a result of the limited sampling of longer LCTs during Experiment 2 (a requirement of the experiment to maintain subject comfort and compliance during prolonged hypercapnia). Truncating the data in Experiment 1 (Vcutoff = 3 cm/s) at a maximum LCT of 2.0 s reduces the mean GM bolus duration estimate from 2.22 ± 0.53 s to 1.69 ± 0.12 s. However, the mean CBF₀ estimate is not significantly affected with values of 56 ± 8 ml/100 g/min and 61 ± 9 ml/100 g/min. The histograms in Supplementary Figure S2 demonstrate how bolus duration estimates are truncated at 2.0 s (the longest LCT) in Experiment 2 but not Experiment 1, with many estimates hitting the 2.0 s limit during normocapnia. Therefore, sampling of LCTs was not sufficient to quantify the bolus duration accurately. However, during hypercapnia the increased blood flow velocity appears to reduce the bolus duration significantly, with far fewer estimates hitting the 2.0 s limit. Indeed, with the modal peak appearing around 1.4 s, it appears that the sampling of LCTs is sufficient to capture the hemodynamics during hypercapnia, with only a small negative bias expected from group-averaged results.

Figure 7 shows the group-averaged maps of the kinetic model fits, demonstrating the significant reduction in bolus duration during hypercapnia, with shorter bolus duration in the frontal and temporal regions compared to posterior regions during hypercapnia. The average bolus duration map during normocapnia is more uniform due to the 2.0 s upper LCT limit. There is no observable change in the macrovascular bolus duration between normocapnia and rest, while there appears to be an increase in the macrovascular blood volume, with GM macrovascular blood volume being 0.10 ± 0.07 at normocapnia and 0.15 ± 0.07% at hypercapnia (mean ± std over participants).

Expanding the regional analysis to the MNI lobes (Table 2), we can see that the minimum average bolus duration is 1.15 ± 0.30 s in the caudate, with many regions including both the frontal and temporal lobes having a bolus duration less than 1.4 s (the consensus recommended LCT). Therefore, consistent with the results shown in Figure 6, the bolus duration measurements suggest that an LCT of less than 1.4 s is needed to accurately capture CVR. Indeed, an upper limit of approximately 1.2 s appears to be appropriate in this adult cohort.

This work presents measurements of bolus duration in the healthy adult brain for Fourier-transform velocity selective inversion ASL. We show that while grey matter average bolus duration is 2.2 s at rest, there is a large spatial variability, with anterior and temporal regions having shorter bolus durations than posterior regions. We extend these findings by measuring a reduction in bolus duration during hypercapnia in an experiment designed to measure CVR, providing an example where high flow velocity causes a reduction in bolus duration. These results help to understand the origin of reported underestimation of CVR with FT-VSI ASL when using conventional LCTs (Xu et al., 2024), and enable an upper bound to be placed on LCT for such experiments.

The motivation for longer LCT is that more labelled blood can perfuse into tissue, increasing signal. Therefore, the optimal LCT is a balance of specificity (shorter LCT) and sensitivity (longer LCT). Future VSASL studies can take these results to inform the choice of LCT based on bolus duration for their population and region of interest. For example, in the healthy, predominantly female cohort studied in Experiment 1, the consensus recommendation LCT of 1.4 s appears appropriate for all anatomical regions at rest and is expected to provide unbiased CBF estimates across the brain. However, for CVR experiments an LCT of 1.2 s (or less) appears to be more appropriate, being the lower limit of the bolus duration measured during hypercapnia, as it would provide an unbiased measurement of CBF change across all cortical and sub-cortical regions. In cohorts where fast flow velocity can occur during normocapnia, for example, anaemia (Gottesman et al., 2012), subarachnoid haemorrhage (Uryga et al., 2022), and traumatic brain injury (Kelly et al., 1996), then it is likely that a shorter LCT will also be required to avoid CBF underestimation.

Both at rest and during hypercapnia a clear pattern of reduced bolus duration in frontal and temporal regions was observed and is consistent with shorter ATT in the anterior circulation, compared to the posterior circulation, that has been observed using pCASL (Pinto et al., 2023; Zhao et al., 2021). This pattern of reduced regional bolus duration is also evident in the results of Xu et al. (2024), where the largest CVR reduction was observed in these regions with a LCT of 1.52 s. The reduction in GM bolus duration during Experiment 2, (CVR acquisition) was approximately 13%, which is comparable with ATT reductions recently reported with pCASL during hypercapnia (Pinto et al., 2023). However, this is likely to be an underestimate as bolus duration measurements are limited to 2.0 s in our hypercapnia experiments, which is insufficient for characterising the bolus duration at rest.

Secondary to measuring bolus duration, Experiment 1 also investigated the effect of v_cutoff on the FT-VSI ASL signal. Minimizing v_cutoff brings the label into smaller arteries and arterioles, closer to the capillary bed. This extends the leading edge of the labelled bolus and minimises the contribution of small blood vessels to the subtracted ASL signal. However, signal contamination from CSF is present with low v_cutoff values. Thus, the choice of v_cutoff is somewhat a trade-off between CSF contamination and the contribution of small blood vessels to the subtracted ASL signal. While there appears to be a small, non-significant reduction in bolus duration with v_cutoff changing from 2 to 4 cm/s (Fig. 3), this is small compared to intra- and inter-participant variability in bolus duration, while CBF measurement was not affected by v_cutoff. However, increasing v_cutoff above 2 to 3 cm/s resulted in reduced CSF signal contamination with a smaller reduction apparent moving from 3 to 4 cm/s (Supplementary Fig. S1). Therefore, considering the trade-off between CSF contamination and the desire to minimise blood vessel contribution, we chose to use a v_cutoff of 3 cm/s for Experiment 2, where CSF contamination could confound measurement of CBF CVR, where CSF spaces are displaced by global vasodilation (Blockley et al., 2011; Thomas et al., 2013). In the case where a patient group may have atrophy, or for CVR studies (Zhao et al., 2021), our results support using a v_cutoff of 3 cm/s to reduce CSF contamination, while not significantly reducing the size of the labelled bolus.

While visible in the lateral ventricles, CSF also borders cortical regions, so it will form a small fraction of many of the voxels classified as grey matter, contributing to the signal of these voxels. Therefore, a consideration of the nature of the CSF signal is important to understand if CSF signal contamination could bias the bolus duration calculation in cortical regions. The CSF signal has large variability due to its sensitivity to physiological noise, especially cardiac pulsations. This high variability, combined with multiparametric fitting to the limited number of LCT datapoints, meant that we were unable to fit a CSF component in the multi-compartment model. However, the T₁ of CSF is long compared to the perfusion kinetics, so CSF signal is relatively constant across the range of LCT values acquired here, as demonstrated in Figure 1. The contribution of physiological noise in the CSF signal adds a large degree of variation across measurements, but there is no dependence of this variability on LCT.

A limitation of our method for measuring bolus duration is that we assume a sharp trailing edge of the labelled bolus which does not account for any dispersion. Dispersion would smooth out the kinetic curve. The effect of dispersion in the data, not accounted for in the model kinetic curve, would be to overestimate bolus duration. Additionally, we chose to fit a macrovascular compartment to minimise any bias from residual macrovascular signal that was not completely supressed due to the macrovascular crushing gradients being applied in only one direction (foot-head). It is possible that this component is not a pure macrovascular signal but also contains some contribution from diffusion attenuation effects that are not completely matched between the velocity-selective label and control. The effect of any diffusion contribution to this compartment would be to introduce error into the macrovascular bolus duration. However, it is unlikely that such errors would have a significant influence on the principal parameters of interest, CBF and bolus duration.

A further consideration is the use of slab-selective pre-saturation, rather than global pre-saturation. Slab-selective pre-saturation has the potential benefit of increased SNR for CVR ASL experiments, where we wish to minimise the TR so that the acquisition is sensitive to CBF changes over a short time scale. The SNR benefit is due to the assumption that all arterial blood is fresh blood and, therefore, has fully relaxed longitudinal magnetisation, whereas the magnetisation is reduced by a factor of 1 – exp(-Tsat/T_1,blood) when using global pre-saturation (Tsat = saturation time). A regional pre-saturation scheme was also used in the work of Xu et al. (2024), and further investigation is needed to determine if this has an influence on the bolus duration under conditions of fast flow velocity.

In conclusion, bolus duration in FT-VSI reduces significantly with increased flow velocity, as demonstrated with a hypercapnic intervention in a cohort of healthy adults. Thus, the use of a shorter LCT will be important when studying groups or states where bolus duration is short or expected to change, such as cerebrovascular reactivity mapping, anaemia, subarachnoid haemorrhage, and traumatic brain injury. Our data suggest that an LCT of 1.2 s or less will minimise any bias in CVR measurements in healthy adult populations, although at the expense of reduced signal-to-noise in superior and posterior brain regions that exhibit a longer bolus duration. We also found that increasing v_cutoff from 2 to 3 cm/s did not have a significant effect on either bolus duration or CBF quantification. However, there is a clear reduction in CSF contamination, which, if present, could bias cerebrovascular reactivity measurements, or confound CBF quantification in certain pathologies. Therefore, we recommend increasing FT-VSI v_cutoff to at least 3 cm/s for such studies.

Participants only gave ethical approval for anonymised data to be shared, so we are unable make the raw ASL or MPRAGE data publicly available. Instead, we provide bolus duration and CBF parameter maps for each participant, in group (MNI152) space, and summary grey matter and Harvard-Oxford atlas region-averaged data, available at https://doi.org/10.17605/OSF.IO/6SJNQ.

The analysis code for estimating CBF and bolus duration is available online https://doi.org/10.5281/zenodo.13755864.

Ian D. Driver: Conceptualisation, Methodology, Formal analysis, Investigation, and Writing—Original Draft. Hannah L. Chandler: Investigation, Writing—Review & Editing. Eleonora Patitucci: Investigation, Writing—Review & Editing. Emma L. Morgan: Investigation, Writing—Review & Editing. Kevin Murphy: Funding acquisition, Writing—Review & Editing. Stefano Zappala: Investigation, Writing—Review & Editing. Richard G. Wise: Methodology, Funding acquisition, and Writing—Review & Editing. Michael Germuska: Conceptualisation, Methodology, MRI sequence development, Formal analysis, Investigation, Funding acquisition, and Writing—Review & Editing.

The authors have no competing interests to declare.

This research was funded by a Wellcome Trust Strategic Award [104943/Z/14/Z]. For the purpose of open access, the author has applied a CC BY public copyright licence to any Author Accepted Manuscript version arising from this submission.

I.D.D., H.L.C., and K.M. are supported by the Wellcome Trust [WT224267].

M.G. is supported by the Wellcome Trust [220575/Z/20/Z] and the Engineering and Physical Sciences Research Council [EP/S025901/1].

R.G.W. is supported by

European Union-NextGenerationEU-Italian Ministry of University and Research (MUR), National Plan for Recovery and Resilience (PNRR) and Projects of National Relevance (PRIN), Project Code: P2022ESHT4, Project Title: “Advancing MRI biomarkers of brain tissue microstructure and energetics in Multiple Sclerosis.” Funding call No. 1409 of September 14, 2022, Concession decree No. 1367 of September 1, 2023 adopted by MUR, ERC Panel LS5 “Neuroscience and Disorders of the Nervous System” CUP: D53D23019210001.

European Union—NextGenerationEU under the National Plan for Recovery and Resilience (PNRR), Mission 4 Component 2—M4C2, Investment 1.5—Call for tender No. 3277 of December 30, 2021 Italian Ministry of Universities Award Number: ECS00000041, Project Title: “VITALITY—Innovation, digitalization and sustainability for the diffused economy in Central Italy,” Concession Decree No. 1057 of June 23, 2022 adopted by the Italian Ministry of University and Research. CUP D73C22000840006.

Supplementary material for this article is available with the online version here: https://doi.org/10.1162/imag_a_00506.