On the optimization of 3D inflow-based vascular-space-occupancy (iVASO) MRI for the quantification of arterial cerebral blood volume (CBVa)

Abstract

PURPOSE:

The inflow-based vascular-space-occupancy (iVASO) MRI was originally developed in a single-slice mode to measure arterial-cerebral-blood-volume (CBVa). When vascular crushers are applied in iVASO, the signals can be sensitized predominantly to small pial arteries and arterioles. The purpose of this study is to perform a systematic optimization and evaluation of a 3D-iVASO sequence on both 3T and 7T for the quantification of CBVa values in the human brain.

METHODS:

Three sets of experiments were performed in three separate cohorts. (1) 3D-iVASO-MRI protocols were compared to single-slice iVASO, and the reproducibility of whole-brain 3D-iVASO-MRI was evaluated. (2) The effects from different vascular crushers in iVASO were assessed. (3) 3D-iVASO-MRI results were evaluated in arterial and venous blood vessels identified using ultra-small-superparamagnetic-iron-oxides (USPIO) enhanced MRI to validate its arterial origin.

RESULTS:

3D-iVASO scans showed signal-to-noise ratio (SNR) and CBVa measures consistent with single-slice iVASO with reasonable intra-subject reproducibility. Among the iVASO scans performed with different vascular crushers, the whole-brain 3D-iVASO scan with a motion-sensitized-driven-equilibrium (MSDE) preparation with two binomial refocusing pulses and an effective TE of 50 ms showed the best suppression of macrovascular signals, with a relatively low specific-absorption-rate (SAR). When no vascular crusher was applied, the CBVa maps from 3D-iVASO scans showed large CBVa values in arterial vessels, but well suppressed signals in venous vessels.

CONCLUSION:

A whole-brain 3D-iVASO-MRI scan was optimized for CBVa measurement in the human brain. When only microvascular signals are desired, a MSDE-based vascular crusher with binomial refocusing pulses can be applied in 3D-iVASO.

1. Introduction

The inflow-based vascular-space-occupancy (iVASO)^1–6 MRI is a noninvasive approach for the measurement of cerebral blood volume in arterial blood vessels (CBVa). When flow-sensitive crushing gradients are incorporated to remove fast flowing blood signals in large arteries, CBVa measured in iVASO MRI can be sensitized predominantly to microvasculature with arterial blood, here defined as small pial arteries and arterioles. As CBVa is an important indicator of tissue perfusion and the integrity of microvasculature, iVASO MRI has been applied in a number of clinical studies⁷, including stroke⁶, brain tumors^8–14, mental disorders such as schizophrenia^15,16, and neurodegenerative diseases such as Huntington’s disease, Alzheimer’s disease and Parkinson’s disease^17–20, as well as in other tissue such as muscle²¹.

In iVASO MRI, CBVa is calculated from the difference signal between an arterial blood nulled image and a second image without blood nulling (control). To account for the heterogeneity of vascular transit times, images are acquired at multiple post-inversion delay times (TI), from which absolute CBVa can be quantified². Crushing gradients can be applied to suppress signals from large arteries and thus sensitize the contrast to pial arteries and arterioles. When the blood brain barrier (BBB) becomes leaky, water exchange effects can be accounted for in the iVASO theory² using a water extraction fraction parameter (E, ranging from 0 to 1) and the Kety-Schmidt theory²². Unlike the original vascular-space-occupancy (VASO) MRI¹, the partial volume effect from cerebrospinal fluid (CSF) is minimal in the iVASO approach². This is because the circulation of CSF in the brain^23,24 is much slower than blood, and thus the CSF signals in subsequent nulling and control scans are comparable in magnitude and cancel out upon subtraction. This is confirmed experimentally by measuring the T2 value of the difference signal in iVASO. As CSF has a much longer T2 (> 1000 ms) than blood and brain tissue (< 100 ms), the measured T2 values of the difference signal² indicate that contamination from CSF, which would increase the apparent T2, is minimal in iVASO. The T2 values also indicate the predominant arterial origin of the iVASO signal², as arterial blood T2 is distinct from that of capillary and venous blood, as well as the extravascular tissue and CSF. Recently, CBVa measured by iVASO MRI has been validated using histological markers of arterial blood vessels in a mouse model²⁰.

In this study, three sets of experiments were performed to address specific challenges in iVASO MRI based on previous studies, with a common goal of optimizing and validating iVASO MRI for the quantification of CBVa. The iVASO MRI approach was originally developed in a single-slice mode using a gradient-echo (GRE) echo-planar-imaging (EPI) readout^1–6. It was later expanded to a three-dimensional (3D) mode covering most of the brain using a 3D gradient spin echo (GRASE) readout by Rane et al.²⁵, or a 3D turbo-field-echo (TFE, also known as fast gradient echo, fast GRE, or TurboFLASH) readout^15,16,26,27, which has less signal dropout and geometric distortion compared to the original EPI sequence²⁸. However, a systematic optimization and evaluation of the 3D iVASO sequence has not been conducted. In this study, we aim to optimize 3D TFE iVASO MRI and evaluate its performance on both 3T and 7T human MRI systems. Whole-brain and partial-brain 3D TFE iVASO MRI protocols were compared to the original single-slice iVASO MRI sequence. The reproducibility of whole-brain 3D TFE iVASO MRI was evaluated in healthy volunteers. Secondly, the effects from vascular crushers in iVASO were assessed. Finally, 3D TFE iVASO MRI results were compared to ultra-small superparamagnetic iron oxides (USPIO) enhanced multi-echo susceptibility weighted imaging (SWI)²⁹ in a group of patients treated for iron deficiency anemia (IDA) using intravenous (IV) infusion of Iron-Dextran (Infed^®, Allergan USA, Inc., Madison, NJ, USA). The primary goal here is to use Iron-Dextran enhanced SWI images to identify arterial and venous vessels so that the arterial origin of the iVASO signals can be assessed. Iron-Dextran is a commonly used FDA-approved USPIO to treat patients with IDA, and has been used off-label as an MRI contrast agent³⁰. Compared to gadolinium(Gd)-based contrasts, USPIO contrasts can be used in patients with impaired renal function, and have a prolonged blood-pool-phase with a plasma half-life of 14–21 hours³¹. In this MRI study on these patients, pre- and post-infusion MRI images were acquired and the results were compared to pre-infusion iVASO MRI in the same patients. The post-infusion MRI scans were performed on the same day of the infusion during the blood-pool phase (i.e. < 15 hours after infusion) when most of the iron oxide nanoparticles remain intravascular.

The overall goal of the current study is to develop an iVASO protocol suitable for clinical studies with a whole-brain coverage and a reasonably short total scan time. Note that the iVASO MRI approach is intrinsically different from the VASO functional MRI (fMRI) method³². While VASO fMRI has been widely used for high-resolution laminar functional mapping^33,34, iVASO MRI is a perfusion-based method designed for the quantification of CBVa in brain tissue, similar to arterial spin labeling (ASL) MRI. Therefore, this study was not designed for high-resolution brain mapping. As the volume of small arterial vessels is typically about 1–2% in a voxel of human brain tissue⁷, the iVASO difference signal is very small. Thus, spatial resolution comparable to those recommended for ASL MRI in clinical studies³⁵ was adopted in this study.

2. Method

2.1. Study Participants

Three cohorts of volunteers were recruited for this study. All volunteers provided informed, written consent in this Johns Hopkins Institutional Review Board (IRB) approved, Health Insurance Portability and Accountability Act (HIPAA) compliant study. Cohort 1: five healthy subjects (age = 27.2±4.4 yr; gender = 2F/3M), and cohort 2: five healthy subjects (age = 23.8±2.6 yr; gender = 3F/2M) were recruited for two sets of experiments performed on 3T respectively, as described in the next sub-section. In cohort 3, six patients (age = 65.7±8.5 yr; gender = 4F/2M) with Restless Legs Syndrome (RLS) were recruited for an add-on (in addition to the clinical studies in these patients) MRI study performed on 7T. RLS is a condition that is associated with low iron in the brain³⁶, and IV-infusion of Iron-Dextran is a common treatment for RLS to restore the body’s iron levels. Each participant made two visits for this study. The pre-infusion scans were acquired at the first visit. The IV-infusion of Iron-Dextran (Infed^®, Allergan USA, Inc., Madison, NJ, USA; standard protocol: 1000 mg, 2 hours) was performed in a separate clinic. After that, the participants returned to the imaging center for the post-infusion scans within 15 hours. Note that cohort 3 was recruited mainly because infusion of Iron-Dextran can only be performed in these patients but not in healthy subjects. This provided an opportunity to validate the predominant arterial origin of the iVASO signals by using an alternative contrast enhanced MRI method. The goal of the experiments here was not to study CBVa changes in these patients compared to controls, but to compare CBVa measured by iVASO in arterial and venous vessels identified using contrast enhanced SWI scans performed in the same subjects.

2.2. MRI Acquisition

2.2.1. Experiment 1: 3T, cohort 1, comparison of 3D and 2D iVASO scans, reproducibility.

In this study, the 3D TFE iVASO sequence^15,16,26,27 was adopted. This experiment was designed to compare the 3D TFE iVASO protocols with the original single-slice 2D GRE EPI iVASO scans, and to assess the scan-rescan reproducibility of 3D TFE iVASO. All scans in experiment 1 were conducted using a 32-channel phased-array head coil for signal reception and a dual-channel body coil for transmit on a 3T human MRI scanner (Philips Healthcare, Best, The Netherlands). Compressed-Sensing SENSitivity Encoding (CS-SENSE)^37–40 was used for parallel imaging acceleration. As CS-SENSE is a vendor supplied method, its details are not available to users. The current study was performed using CS-SENSE implemented in Philips Release 5.6 with a variable density incoherent under-sampling pattern and a default denoising level (which controls the level of wavelet regularization in CS-SENSE reconstruction and balances between lower noise level and less artificial smoothing) of 15%. The following MRI scans were performed on each subject in cohort 1:

• (1.1)3D TFE iVASO MRI with whole-brain coverage: voxel size = 3×3×6 mm³; 19 slices; flip angle (FA) = 20°; time of repetition (TR)/time of inversion(TI) = 403/191, 600/277, 800/360, 1000/437, 1500/611, 2000/757, 2700/921, 3500/1059 ms, 4 pairs of interleaved nulling and control scans were acquired at each TR/TI; k-space profile = centric (also known as low-high, i.e. the center of k-space is acquired at the first echo); multi-shot 3D TFE readout: shot number = 5; TFE factor = 29; TR_TFE = 3.7 ms; TE = 1.7 ms; TFE readout pulse train duration = 108.8 ms; CS-SENSE = 7; halfscan (partial Fourier) = 0.8; turbo direction = radial. Volume shimming (first order) was utilized in all scans in order to improve B₀ homogeneity in the entire brain (covering both the imaging volume and the inflowing arteries outside of the imaging slab). If only the imaging volume is shimmed with higher order B0 shim, the B0 field in the inflowing artery region would be very inhomogeneous. If higher order B0 shim is applied over a large region covering both the imaging volume and inflowing arteries, the shimming results are often suboptimal. The TI values were calculated based on arterial blood T1 values^41–43. The thickness of flip-back slab was 126 mm. A reference scan (TR = 20000 ms, other imaging parameters the same as iVASO scans) through the ventricle was acquired to determine the scaling factor M₀ needed to calculate the absolute CBVa values from iVASO images (see Data Analysis). The total duration for this set of scans was 10 minutes.
• (1.2)3D TFE iVASO MRI with partial brain coverage: 9 slices, shot number = 2. All the other parameters were identical as the whole brain 3D TFE iVASO scan in (1.1). The middle slice of the imaging volume was aligned between the whole brain and partial brain iVASO scans (Figure 1A). A reference scan (TR = 20000 ms, other imaging parameters the same as iVASO scans) through the ventricle was acquired with the same readout. The total duration for this set of scans was 4 minutes.
• (1.3)2D GRE EPI iVASO MRI (single-slice): this is the original iVASO sequence developed in the first iVASO papers^1,2, which has been validated recently using histology in a mouse study²⁰. It was performed here to compare to the 3D TFE iVASO sequences. Three sets of 2D single-slice EPI iVASO scans were performed at 3 different locations aligned with the top (slice #14), middle (slice #10), and bottom (slice #6) slices from the whole-brain 3D TFE iVASO scans (Figure 1B). The following imaging parameters were used in 2D EPI iVASO: single-shot 2D GRE EPI readout; TE =25 ms; and SENSE (AP direction) = 2. The imaging parameters that were kept identical in both 3D TFE and 2D EPI iVASO acquisitions include: voxel size, FA, TR/TI, thickness of flip-back slab, number of repeats, and B₀ shimming. A reference scan (TR = 20000 ms, other imaging parameters the same as iVASO) through the ventricle was acquired with the same readout. The total duration for this set of scans was 3.5 minutes.
• (1.4)A whole-brain 3D magnetization prepared rapid acquisition gradient echo (MPRAGE) scan (TR/TE/TI = 8.1/3.7/749 ms, CS-SENSE = 8, voxel size = 1 mm isotropic, 150 slices) was performed for whole-brain anatomical reference. The duration of this scan was 1 minute.
• (1.5)A low resolution MPRAGE scan with the same spatial resolution and coverage as the whole-brain 3D TFE iVASO scans was also acquired to assist the co-registration and segmentation (see Data Analysis). All the other imaging parameters were identical to those of the whole brain MPRAGE scan. The duration of this scan was 30 seconds.

Figure 1.

Illustration of the positions of imaging volumes in the (A) whole-brain (19 slices) and partial-brain (9 slices) 3D TFE iVASO; and in the (B) three single-slice (a, b, c) EPI iVASO MRI scans performed in Experiment 1.

In addition, to assess the reproducibility of the whole-brain 3D TFE iVASO scan, a repeat session was performed in each subject of Cohort 1 within 2 weeks from their initial session (average interval = 5 ± 5 days). The whole-brain 3D TFE iVASO and low resolution MPRAGE scan with identical imaging parameters as the initial scans were performed in the repeat session.

2.2.2. Experiment 2: 3T, cohort 2, effects from vascular crushers.

This experiment was designed to evaluate the effects from vascular crushers in iVASO in order to suppress fast flowing blood signals from large arterial vessels, so that the iVASO signals can be sensitized to small pial arteries and arterioles in the tissue, which are considered the primary vascular regulators of regional cerebral perfusion. In the original single-slice 2D GRE EPI iVASO, the vascular crusher was implemented using bipolar crushing gradients^2,28. In 3D TFE iVASO, it can be implemented using a motion-sensitized driven equilibrium (MSDE) preparation^28,44–46, which is a spatially nonselective Carr–Purcell–Meiboom–Gill (CPMG) based T2 preparation module with inserted motion-sensitized crushing gradients in the z-direction immediately before the 3D TFE readout. Adiabatic or binomial pulses are typically used for the refocusing pulses in the MSDE preparation. Whereas adiabatic pulses provide better spatial homogeneity, they usually have a higher power deposition compared to binomial pulses. A hyperbolic secant pulse widely available on clinical scanners was used for adiabatic refocusing. The effective TE of the MSDE preparation is also an important parameter. A longer effective TE in MSDE usually results in a better flow suppression but also leads to more signal decay mainly due to T2 relaxation. All these parameters were evaluated in experiment 2. A TE of 50 ms that was used in our previous studies²⁸ and a TE of 15 ms that is the shortest we can implement were compared. The velocity encoding cutoff V_enc was chosen to be 3 cm/s based on our previous studies². All scans in experiment 2 were conducted using the same 3T Philips system as in experiment 1. The following MRI scans were performed on each subject in cohort 2:

• (2.1)Whole-brain 3D TFE iVASO MRI without a vascular crusher: TR/TI = 532/248, 600/277, 800/360, 1000/437, 1500/611, 2000/757, 2700/921, 3500/1059 ms. Note that the shortest TR/TI (532/248 ms) was made longer here than scan (1.1) in experiment 1 to reduce the specific absorption rate (SAR) in subsequent scans in experiment 2 with increased power deposition from the MSDE preparation. The SAR values in all scans were within the FDA limit (SAR<1.95W/kg). All the other imaging parameters were the same as for scan (1.1) in experiment 1.
• (2.2)Whole-brain 3D TFE iVASO MRI with a MSDE vascular crusher implemented using two adiabatic refocusing pulses: hyperbolic secant pulses (duration = 10 ms, β = 525 rad/sec, μ = 8.4, peak B₁ = 13.5 μT, bandwidth = 1050Hz) were applied for refocusing, effective TE = 50 ms, velocity encoding V_enc = 3 cm/s, SAR < 2.91 W/kg. The imaging parameters in iVASO were the same as for scan (2.1).
• (2.3)Whole-brain 3D TFE iVASO MRI with a MSDE vascular crusher implemented using two binomial refocusing pulses: (1331) binomial pulses implemented using block pulses with an inter-pulse interval of 0.65 ms⁴⁷ were used for refocusing, SAR < 2.64 W/kg. the other parameters were the same as for scan (2.2).
• (2.4)Whole-brain 3D TFE iVASO MRI with a MSDE vascular crusher implemented using two binomial refocusing pulses with a shorter effective TE = 15 ms, SAR < 2.73 W/kg. All the other parameters were the same as for scan (2.3).
• (2.5)Single-slice 2D GRE EPI iVASO MRI with a vascular crusher implemented using bipolar crushing gradients^2,28: V_enc = 3 cm/s; SAR < 1.47 W/kg. The TR/TI values were the same as scan (2.1). All the other parameters were the same as for the single-slice iVASO scan in experiment 1.
• (2.6)The same MPRAGE scans as scans (1.4) and (1.5) in experiment 1.
• (2.7)A velocity-selective (VS) MR angiography (MRA)⁴⁸ scan with the same coverage as the whole-brain 3D TFE iVASO scans was performed to facilitate the identification of large arterial vessels in the brain: voxel size = 0.7×0.7×1 mm³; 114 slices; CS-SENSE = 8; TR/TE = 9.1/2.3 ms; scan time = 3.5 minutes.

2.2.3. Experiment 3: 7T, cohort 3, comparison with USPIO enhanced MRI

This experiment is designed to evaluate whole-brain 3D TFE iVASO scans on 7T, and to investigate the arterial origin of iVASO signals by comparing CBVa measured in large arteries and veins, which can be identified using USPIO enhanced MRI. All scans in experiment 3 were performed on a Philips 7T MRI scanner (Philips Healthcare, Best, The Netherlands). A 32-channel phased-array head coil (Nova Medical, Wilmington, MA, USA) was used for RF reception, and an 8-channel multi-transmit (MTX8) head coil was used for RF transmission. The MRCodeTool software (Tesla Dynamic Coils, Zaltbommel, the Netherlands) was used for B₀ shimming. Rectangular pads filled with high dielectric constant materials⁴⁹ were positioned on the side of the subjects’ heads to improve B₁ field homogeneity. As CS-SENSE was not available on the 7T scanner, the original SENSitivity Encoding (SENSE) method was used for parallel imaging acceleration⁵⁰. The following MRI scans were performed on each subject in cohort 3 before the USPIO administration:

• (3.1)Pre-USPIO 3D TFE iVASO MRI with whole-brain coverage: TR/TI = 1600/669, 1800/735, 2000/797, 2200/856, 2400/912, 2600/964, 2800/1013, 3000/1059 ms; the TI values were calculated based on arterial blood T1 values^41–43; voxel =3×3×6 mm³, 19 slices (same as 3T); TFE factor = 55, shot number = 2, FA = 4°, TR_TFE = 4.3 ms; TE = 2.3 ms; parallel imaging acceleration (SENSE) = 3(RL)×2(FH); SAR < 3.0 W/kg; TFE readout pulse train duration = 235.4 ms; All the other parameters were the same as the 3T whole-brain 3D TFE iVASO scan (1.1). A reference scan (TR = 20000 ms, other imaging parameters the same as iVASO) through the ventricle was acquired to determine the scaling factor M0. The total duration for this set of scans was 10 minutes.
• (3.2)Pre-USPIO SWI MRI: 3D GRE, voxel = 0.4×0.4×1 mm³, 96 slices, TR = 42 ms, TE1/TE2/TE3 = 5/10/15 ms, FA = 12°. The total duration for this scan was 10 minutes.
• (3.3)An MPRAGE scan (TR/TE/TI = 5.0/1.9/563 ms, SENSE = 2(AP)×2(RL), voxel size = 1 mm isotropic, 180 slices, scan time = 2 minutes 15 seconds) was performed for whole-brain anatomical reference.

After the USPIO administration, the following MRI scans were performed in each subject in cohort 3:

• (3.4)Post-USPIO SWI MRI: same as scan (3.2).

2.3. Data Analysis

Data analysis was performed mainly using in-house code programmed in Matlab 2017a (Mathworks, Natick, MA, USA). All iVASO images were motion corrected using the realignment routine in the statistical parametric mapping (SPM) software package (Version 12, Wellcome Trust Centre for Neuroimaging, London, UK). The difference signals between the arterial blood nulled and control iVASO images were calculated using the surround subtraction approach⁵¹. The steady state difference signal (S_diff) between the arterial blood nulled and control scans in iVASO when using a 3D TFE readout can be written as Eq. [1]. Compared to the iVASO equations derived for single-slice iVASO MRI using a 2D GRE EPI readout with a FA = 90°^1,2, a sin(FA) term was added to these equations:

$$\{\begin{array}{ll}=0,\hfill & \hfill \text{when TI}<\Delta \hfill \\ ={\text{M}}_0\cdot {\text{C}}_{\text{b}}\cdot \left(\frac{{\text{CBV}}_{\text{a}}}{{\delta }_{\text{a}}}\right)\cdot \left(\text{TI}-\Delta \right)\cdot \left(\text{1}-{\text{e}}^{-\frac{\text{TR}}{{\text{T}}_{1,\text{a}}}}\right)\cdot {\text{e}}^{-\frac{\text{TE}}{{\text{T}}_{2,\text{a}}^{*}}}\cdot \text{sin}\left(\text{FA}\right),\hfill & \hfill \text{when }\Delta <\text{TI}<\Delta +{\delta }_{\text{a}}\hfill \\ {\text{=M}}_{\text{0}}\cdot [{\text{C}}_{\text{b}}\cdot {\text{CBV}}_{\text{a}}\cdot \left(1-{\text{e}}^{-\frac{\text{TR}}{{\text{T}}_{1,\text{a}}}}\right)\cdot {\text{e}}^{-\frac{\text{TE}}{{\text{T}}_{2,\text{a}}^{*}}}+{\text{C}}_{\text{b}}\cdot \left(1-\text{E}\right)\cdot \text{f}\cdot {\text{TI}}_{\text{ex}}\cdot {\text{e}}^{{\text{TI}}_{\text{ex}}\cdot \left(\frac{\text{1}}{{\text{T}}_{1,\text{a}}}\text{-}\frac{\text{1}}{{\text{T}}_{1,\text{c}}}\right)}\cdot \left({\text{1-e}}^{-\frac{\text{TR}}{{\text{T}}_{1,\text{a}}}}\right)\cdot {\text{e}}^{-\frac{\text{TE}}{{\text{T}}_{2,\text{c}}^{*}}}+{\text{C}}_{\text{t}}\cdot \text{(}\frac{\frac{\text{E}\cdot \text{f}}{\lambda }}{\frac{\text{1}}{{\text{T}}_{1,\text{app}}}-\frac{\text{1}}{{\text{T}}_{1,\text{c}}}}\text{)}\cdot {\text{(e}}^{-\frac{{\text{TI}}_{\text{ex}}}{{\text{T}}_{1,\text{c}}}}-{\text{e}}^{-\frac{{\text{TI}}_{\text{ex}}}{{\text{T}}_{1,\text{app}}}}\text{)}\cdot {\text{e}}^{-\frac{{\text{TI}}_{\text{ex}}}{{\text{T}}_{1,\text{a}}}}\cdot \left({\text{1-e}}^{-\frac{\text{TR}}{{\text{T}}_{1,\text{a}}}}\right)\cdot {\text{e}}^{-\frac{\text{TE}}{{\text{T}}_{2,\text{t}}^{*}}}\text{]}\cdot \text{sin}\left(\text{FA}\right),\hfill & \hfill \text{when }\Delta {\text{+δ}}_{\text{a}}<\text{TI}\hfill \end{array}$$ [1]

where Δ is the time for inverted blood water spins to reach the imaging slice; δ_a is the time for inverted blood water spins to perfuse the arterial and arteriolar vessels; TI_ex=TI-Δ-δ_a; $\frac{\text{1}}{{\text{T}}_{\text{1,app}}}\text{=}\frac{\text{1}}{{\text{T}}_{\text{1,t}}}\text{+}\frac{\text{E∙f}}{\lambda }$. All the other parameters in the equation are summarized in Table S1 in the Supporting Information.

M₀ is a normalization factor that can be estimated from pure CSF signals measured in the reference scan⁵² using Eq. [2]:

$${\text{S}}_{\text{CSF}}\text{=}{\text{M}}_{\text{0}}·{\text{C}}_{\text{CSF}}\left(\text{1-}{\text{e}}^{\text{-}\frac{\text{TR}}{{\text{T}}_{\text{1,CSF}}}}\right)·{\text{e}}^{\text{-}\frac{\text{TE}}{{\text{T}}_{\text{2,CSF}}^{\text{*}}}}·\text{sin}\left(\text{FA}\right)$$ [2]

where C_CSF=1 ml water/ml CSF. When a long TR (20000 ms) and a short TE (1.7 ms in the 3D TFE readout, and 25 ms in the 2D GRE EPI readout), the equation can be further simplified to S_CSF=M₀.

Using the iVASO equations described above, CBVa can be fitted from S_diff using the Metropolis-Hastings (MH) algorithm^53–56 on a voxel-by-voxel basis. In a typical data set, the average normalized root mean squared error was around 0.1 (10%). Compared to the commonly used least-square algorithm, the MH algorithm is a Bayesian model-based method that is less likely to get stuck in local minima. It has been commonly used to fit imaging data with relatively low signal-to-noise ratio (SNR) such as ASL⁵⁷ and diffusion^54–56 MRI.

Registration between iVASO and anatomical images was conducted manually. Due to the different image contrasts between the iVASO and anatomical images, automatic registration algorithms did not provide robust results. Instead, the imaging volumes of iVASO and anatomical scans were matched and carefully prescribed before the scans. Any residual misalignment was manually corrected during analysis. The MRICloud software^58–60 was used to identify five regions of interest (ROI) on anatomical images: whole-brain grey matter (GM), cuneus, precentral gyrus (PreCG), postcentral gyrus (PostCG), and dorsal anterior cingulate cortex (dACC). These ROIs were chosen to cover different lobes and locations of the brain, and they were often investigated as target regions in previous iVASO studies. Average CBVa in each ROI were computed for each subject.

The SNR in the raw iVASO images (SNR_raw) and the difference iVASO images (SNR_diff) was calculated as: $\text{SNR=S/N×}\sqrt{\text{NSA}}$, where S is the signal, N is the noise level, and NSA is the number of signals averaged². The standard deviation of a difference image from two consecutive dynamic scans was used to estimate the noise level (N).

The generalized Sørensen-Dice similarity coefficient, which ranges between 0 and 1 with 1 indicating complete overlap, was calculated to compare spatial similarity of CBVa maps obtained from different iVASO scans.

To evaluate the reproducibility, the coefficient of variation (CV), defined as the standard deviation of the difference between scan and rescan results divided by the average⁶¹, was calculated to assess the scan-rescan reproducibility of CBVa in each ROI.

In Experiment 2, large arteries were identified on the MRA images using maximum intensity projection (MIP) from six consecutive slices to match the iVASO images. In Experiment 3, the SWI images were processed using the JHU/KKI QSMtoolbox^62–65 to obtain a quantitative susceptibility map (QSM). Large arteries were identified using MIP of the first echo (TE = 5ms) magnitude images acquired in the pre-USPIO SWI scans and minimum intensity projection (MinIP) of the third echo (TE = 15ms) post-USPIO SWI images using the methodology established in previous studies^29,66. Based on the susceptibility difference between venous blood and tissue, large veins were identified using MIP of the QSM images similar to previous studies^29,66. The MRA and SWI images were aligned with iVASO images manually similar to anatomical images. CBVa values in the large artery and vein regions were calculated for each subject.

3. Results

3.1. Experiment 1: 3T, comparison of 3D and 2D iVASO scans, reproducibility

Figure 2 shows representative images from the whole-brain and partial-brain 3D TFE iVASO scans and the single-slice GRE EPI iVASO scan. In all three iVASO protocols, the difference images (S_diff) showed an increase of inflowing blood as TI gets longer. The fitted CBVa maps showed comparable contrast among the three iVASO protocols with generalized Dice similarity coefficients above 0.9 from all comparisons (3D TFE whole-brain vs. 3D TFE partial-brain: 0.96±0.01; 3D TFE partial-brain vs. 2D GRE EPI: 0.91±0.01; 3D TFE whole-brain vs. 2D GRE EPI: 0.91±0.02) in the middle slice. The SNR values of the raw iVASO images (SNR_raw) and the difference images (SNR_diff) in the GM were comparable between 3D TFE whole-brain and 2D GRE EPI iVASO, and were slightly higher in 3D TFE partial-brain iVASO compared to the other two scans (Figure 2D). The fitted CBVa values from the three iVASO protocols in each ROI are shown in Figure 3. No significant difference (P > 0.05) was observed in CBVa in most ROIs between the whole-brain and partial-brain 3D TFE iVASO scans, and between the 3D whole-brain and 2D single-slice iVASO scans. In the reproducibility study, CBVa measured by the whole-brain 3D TFE iVASO scan showed coefficients of variation (CV) of 27–35% in the five ROIs examined (Table 1).

Figure 2.

Representative images of the difference signals in iVASO (S_diff) at each TI (shortest to longest) and the fitted CBVa maps from one subject in the middle slice (slice b in Figure 1B). In each scan, the slice corresponding to the middle slice of the whole-brain iVASO scan is shown. (A) 3D whole-brain iVASO MRI scan. (B) 3D partial-brain iVASO MRI scan. (C) 2D single-slice iVASO MRI scan. (D) Comparison of the SNR in the iVASO images. *: P < 0.05.

Figure 3.

Comparison of the fitted CBVa values in different brain regions from the whole-brain, partial-brain and single-slice iVASO MRI scans (n = 5). (A) 3D TFE partial-brain vs. 3D TFE whole-brain iVASO; (B) 2D GRE EPI single-slice vs. 3D TFE whole-brain iVASO. The same slices in the corresponding protocols were compared. Therefore, the slices from the 3D TFE whole-brain iVASO scans used in (A) and (B) are different. In (A, B), the error bars indicate inter-subject standard deviation. *: P < 0.05. (C, D) Bland–Altman plots for (A, B), respectively; the red lines indicate ± 1.96 standard deviation of the difference.

Table 1.

Scan-rescan reproducibility for CBVa measured from whole-brain 3D TFE iVASO MRI: coefficient of variation (%) by region of interest.

Region	Coefficient of variation (%)
Grey matter	31.8±1.3
Cuneus	28.0±3.5
Precentral gyrus	29.9±4.8
Postcentral gyrus	27.9±4.5
Dorsal ACC	35.7±6.6

3.2. Experiment 2: 3T, effects from vascular crushers

Figure 4 compares the SNR_raw and SNR_diff in GM from the five iVASO scans performed in this experiment. The SNR in 3D TFE whole-brain iVASO scans without vascular crusher was comparable (P > 0.1) to the same scans performed in Experiment 1 (Figure 2). Among the four 3D TFE whole-brain iVASO scans performed in Experiment 2, the scans without vascular crusher showed the highest SNR, followed by the scans with MSDE crushers with a shorter effective TE (15 ms), which were both greater than the SNR in the scans with MSDE crushers with an effective TE of 50 ms (P < 0.05). SNR_raw from 3D TFE iVASO scans with MSDE crushers implemented using adiabatic was slightly lower than the ones with binomial refocusing pulses when the effective TEs are the same (P < 0.05). The SNR in 2D GRE EPI iVASO scans with bipolar gradients was slightly lower (P < 0.05) than the 2D GRE EPI iVASO scans without vascular crusher performed in Experiment 1 (Figure 2). When comparing 3D TFE and 2D GRE EPI iVASO scans, the SNR in 2D GRE EPI iVASO scans with bipolar gradients was comparable to the 3D TFE iVASO scans with MSDE crushers with an effective TE of 50 ms (P > 0.1).

Figure 4.

SNR comparison in GM in iVASO scans performed in Experiment 2. *: P < 0.05.

Figure 5A compares the CBVa values measured from the four 3D TFE whole-brain iVASO scans with different vascular crushers. The 3D TFE iVASO scans without crusher showed the greatest CBVa in most ROIs, the values of which were consistent with Experiment 1. The CBVa values measured from the 3D TFE iVASO scans with MSDE crushers implemented using adiabatic and binomial refocusing pulses with an effective TE of 50 ms were comparable in all ROIs. The CBVa values measured from the 3D TFE iVASO scans with MSDE crushers with a shorter effective TE (15 ms) were slightly larger than the scans with a longer effective TE (50 ms) in most ROIs. Figure 5B compares the CBVa values measured from the 2D GRE EPI iVASO scans with bipolar gradients and 3D TFE whole-brain iVASO scans with MSDE crushers. CBVa values measured from the 2D GRE EPI iVASO scans with bipolar gradients were comparable to the 3D TFE iVASO scans with MSDE crushers with an effective TE of 15 ms in all ROIs.

Figure 5.

Comparison of CBVa results from iVASO protocols with different vascular crushers. (A) CBVa values in different brain regions acquired using 3D TFE whole-brain iVASO scans with different vascular crushers. *: P < 0.05. (B) Comparison of CBVa values between 3D TFE whole-brain and 2D single-slice iVASO with vascular crushers. Every 3D scan was compared to the 2D scan, but no comparisons among 3D scans were performed in (B). The same slices in the corresponding protocols were compared. Therefore, the slices from the 3D TFE whole-brain iVASO scans are different in (A) and (B). (C) Macro- and micro-vascular contributions in the 3D TFE whole-brain iVASO scan without a vascular crusher are shown in different brain regions. CBVa values from the 3D TFE iVASO scan with a MSDE vascular crusher using adiabatic pulses were considered to be predominantly from the microvascular compartment. The difference between this scan and the 3D TFE iVASO scan without vascular crusher was used to estimate the macrovascular contribution. (D) Representative images to show the vascular crushing effects in different iVASO scans. A TOF MRA scan was used to identify the middle cerebral artery (MCA, red arrows), and CBVa maps from iVASO scans with different vascular crushers are compared.

Among all the ROIs, the vascular crushing effects were particularly significant in large arteries. In Figure 5C, the contribution from macrovessels in the 3D TFE iVASO scan without vascular crushers was estimated using the CBVa difference between such a scan and the 3D TFE iVASO scan with a MSDE vascular crusher using adiabatic pulses. In all five GM ROIs, the macrovascular contribution was approximately 35–50%. In large arterial vessels, the macrovascular contribution was about 70%. Figure 5D demonstrates an example of the vascular crushing effects in large arteries in iVASO images acquired with the five iVASO scans in this experiment. The middle cerebral artery (MCA) was identified using an MRA scan. The iVASO scan performed without vascular crusher showed large CBVa values in the artery region, whereas the iVASO scans performed with vascular crushers showed much suppressed CBVa signals in the artery region.

3.3. Experiment 3: 7T, comparison with USPIO enhanced MRI

The SNR in GM from 3D TFE whole-brain iVASO MRI scans without vascular crushers performed on 7T was SNR_raw = 101.8±11.2 and SNR_diff = 45.7±4.9. The 7T SNR results cannot be directly compared to the 3T SNR results mainly due to the different parallel imaging approaches used. The CS-SENSE technique used on 3T was not available on 7T, and therefore the conventional 2D SENSE technique was used for 7T scans, which led to differences in a number of imaging parameters such as shot number, TFE factor, TR_TFE/TE, and FA between iVASO scans performed on 3T and 7T.

Figure 6A shows the CBVa values measured from the 3D TFE whole-brain iVASO scan without vascular crusher on 7T, and compares them to the same iVASO scan performed on 3T (to provide the reference range only; the experiments were not designed to compare 3T and 7T results directly for the reasons stated in the previous paragraph). The CBVa values measured at 3T and 7T were in the same range, but showed small difference in some ROIs. As no vascular crusher was applied, the CBVa values in large arteries were significantly greater than the other regions (P < 0.001). In addition to the ROIs examined in all 3T scans, large veins were identified from SWI images on 7T. The CBVa values in large veins were significantly smaller than arteries (P < 0.001) and all the other ROIs in the GM (P < 0.05). Figure 6B demonstrates an example of CBVa values measured in large arteries and veins in iVASO images. The posterior cerebral artery (PCA) was identified using the pre-USPIO (hyperintensity) and post-USPIO (hypointensity) SWI images, while the internal cerebral vein (ICV) was identified using the pre-USPIO SWI QSM images (hyperintensity). The CBVa map from the 3D TFE whole-brain iVASO scan without vascular crusher performed on 7T showed large CBVa values in the artery, but much suppressed CBVa signals in the vein.

Figure 6.

Results from 3D TFE whole-brain iVASO scans on 7T. (A) CBVa values acquired from 3D TFE whole-brain iVASO MRI without vascular crushers in different brain regions. *: P < 0.05. (B) Representative images in areas with large arterial and venous vessels. Large arterial vessels are expected to show hyperintensity on Pre-USPIO SWI short-TE magnitude MIP images and hypointensity on post-USPIO SWI long-TE MinIP images (red arrows in the first row). Large venous vessels are expected to show hyperintensity on pre-USPIO SWI QSM MIP images (yellow arrows in the second row).

4. Discussion

In the current study, a 3D iVASO MRI sequence was optimized and evaluated on both 3T and 7T human MRI systems for the quantification of CBVa values. The proposed whole-brain 3D TFE iVASO sequence showed consistent CBVa measurement with the original single-slice iVASO MRI sequence on both 3T and 7T¹⁷. The single-slice iVASO was used as a reference standard here as it has recently been validated using histological markers of arterioles in a mouse model²⁰. In 3D TFE iVASO, the imaging slab volume did not show significant effects on the measured CBVa values. Therefore, this 3D iVASO protocol can be tailored to different coverage for various applications. The CBVa values measured with the whole-brain 3D TEF iVASO scans showed reasonable intra-subject reproducibility in healthy volunteers, comparable to previous ASL and VASO reproducibility studies⁶¹, especially in small brain regions with high physiological noise.

Regional parenchymal perfusion in the brain is primarily regulated by smaller pial arteries and arterioles with diameters up to 100–150 μm^2,67–69. Therefore, it is important to differentiate large arteries from smaller pial arteries and arterioles when measuring CBVa. In iVASO, magnetic field gradients can be added in both arterial blood nulled and control scans to suppress signals in large arteries with greater velocity of blood flow^7,70–72. In the original single-slice 2D GRE EPI iVASO, such vascular crushers were implemented as bipolar gradients in a 2D GRE EPI sequence². In this study, 2D GRE EPI iVASO with bipolar gradients was also performed, and the measured CBVa values in healthy subjects were consistent with previous studies when the same velocity encoding cutoff V_enc (3 cm/s) was used². In 3D TFE iVASO, vascular crushers can be implemented with a MSDE preparation^28,44–46. In our data, the flow suppression effects from vascular crushers were confirmed by comparing iVASO signals with and without vascular crushers in large arteries identified from MRA scans. The measured CBVa values in large arteries showed approximately 60% decrease when MSDE based vascular crushers with an effective TE of 50 ms were added. The residual iVASO signals in the large artery regions may be due to incomplete flow suppression or partial volume effects from the surrounding parenchyma (iVASO voxel = 3×3×6 mm³). Furthermore, in this study, the MSDE based vascular crushers in 3D TFE iVASO were evaluated with two different types of refocusing pulses (adiabatic versus binomial pulses) or two different effective TEs (15 ms versus 50 ms). MSDE with binomial refocusing pulses had lower SAR compared to MSDE with adiabatic refocusing pulses, whereas the SNR and CBVa measured from these two scans were comparable. Therefore, although adiabatic pulses have a reduced sensitivity to B₁ inhomogeneities, binomial pulses seemed to work fine possibly due to the self-compensating nature when even number of refocusing pulses are used in the MSDE module and due to the use of a body coil for excitation. When a shorter effective TE (15 ms) was used in MSDE, the SNR of iVASO images was improved largely owing to a shorter T2 decay. However, the flow suppression effects diminished with a shorter effective TE as the measured CBVa values were larger compared to iVASO scans performed using an MSDE with a longer effective TE (50 ms). Together, our results suggest the use of an MSDE preparation with two binomial refocusing pulses and an effective TE of 50 ms when a CBVa measure in predominantly small pial arteries and arterioles is desired. Our data also showed that the difference between iVASO scans with and without vascular crushers can be used to estimate the large artery contribution in each voxel.

It is important to validate the arterial origin of the iVASO signals. In the original iVASO study, the arterial origin was validated by the T2 and T2* values of the difference iVASO signal, which are highly sensitive to blood oxygenation levels². In this study, we provided further evidence that the iVASO signals originate predominantly from arterial vessels. Using USPIO-enhanced SWI images, arterial and venous vessels can be identified in the brain^29,66. The CBVa values measured with the whole-brain 3D TFE iVASO scans without vascular crusher showed large signals in arterial vessels but much smaller signals in venous vessels. The residual iVASO signals in venous vessels may be largely due to strong partial volume effects from the parenchyma regions around the veins given that the spatial resolution of iVASO images (voxel = 3×3×6 mm³) was coarser compared to the size of typical venous vessels in the brain.

Several technical limitations should be discussed. First, the blood nulling times were calculated at the first echo (center of k-space) in the 3D TFE readout. Thus, MR signals acquired at the other echoes in the readout, corresponding to higher spatial frequencies in the MR images, may have incomplete blood nulling. Therefore, it is important to keep the readout pulse train as short as possible. In the current study, multi-shot 3D TFE and parallel imaging approaches were used for this purpose. Also, when comparing different iVASO protocols, the duration of readout pulse train should be kept identical, as we did in the current study. Second, B₁ field inhomogeneity is still a substantial source of error for CBVa quantification in iVASO, similar to many other quantitative MRI methods. On 3T, as the body coil was used for RF transmit, B₁ field in the brain is usually relatively homogeneous. On 7T, dielectric pads were used to improve B₁ homogeneity. The M0 was estimated from the ventricular CSF in the middle of the brain, which usually has the most homogeneous B1 field in the brain. For the rest of the brain, since CBVa was calculated from the iVASO difference signal, assuming B₁ inhomogeneity affects the nulling and control scans similarly, the effect should be largely cancelled out upon subtraction. However, a B1 mapping scan can be added to the protocol for B1 field correction in future studies. Besides, we are currently working on parallel transmit based methods to improve B₁ homogeneity. Thirdly, this study was not designed to compare iVASO MRI at 3T and 7T directly. Instead, it focused on the optimization and validation at each field strength by comparing the results to the single-slice iVASO sequence proposed originally. The results from 3T and 7T scans in the current study cannot be compared directly as various parameters, such as parallel imaging approaches, shot number, TFE factor, readout pulse train duration, TR_TFE/TE, and FA were different due to the intrinsic differences between the two systems. Besides, the MR scans on 3T and 7T were performed in different study participants. A follow-up study is warranted to rigorously compare iVASO MRI on 3T and 7T in the same cohort. Fourth, the coefficients of variation (CV) measured from the repeat scans in the current study, although in the same range as previous reproducibility studies performed at our site⁶¹, were relatively high. Finally, although this is a technical study, the sample size is relatively small. Therefore, the lack of significant difference in the results should be treated with caution.

5. Conclusions

The results indicate that a whole-brain 3D TFE iVASO MRI scan with an MSDE preparation using two binomial refocusing pulses and an effective TE of 50 ms is an optimal sequence to achieve CBVa measurements for predominantly small pial arteries and arterioles in the brain. Results from this technical study will serve as the basis for future clinical studies applying iVASO MRI in various brain diseases.

Supplementary Material

Supinfo

Table S1. Physiological and physical parameters used in the iVASO equations.