T2 -oximetry-based cerebral venous oxygenation mapping using Fourier-transform-based velocity-selective pulse trains.

PURPOSE: To develop a T2 -oximetry method for quantitative mapping of cerebral venous oxygenation fraction (Yv ) using Fourier-transform-based velocity-selective (FT-VS) pulse trains.
METHODS: The venous isolation preparation was achieved by using an FT-VS inversion plus a nonselective inversion (NSI) pulse to null the arterial blood signal while minimally affected capillary blood flows out into the venular vasculature during the outflow time (TO), and then applying an Fourier transform based velocity selective saturation (FT-VSS) pulse to suppress the tissue signal. A multi-echo readout was employed to obtain venous T2 (T2,v ) efficiently with the last echo used to detect the residual CSF signal and correct its contamination in the fitting. Here we compared the performance of this FT-VS-based venous isolation preparations with a traditional velocity-selective saturation (VSS)-based approach (quantitative imaging of extraction of oxygen and tissue consumption [QUIXOTIC]) with different cutoff velocities for Yv mapping on 6 healthy volunteers at 3 Tesla.
RESULTS: The FT-VS-based methods yielded higher venous blood signal and temporal SNR with less CSF contamination than the velocity-selective saturation-based results. The averaged Yv values across the whole slice measured in different experiments were close to the global Yv measured from the individual internal jugular vein.
CONCLUSION: The feasibility of the FT-VS-based Yv estimation was demonstrated on healthy volunteers. The obtained high venous signal as well as the mitigation of CSF contamination led to a good agreement between the T2,v and Yv measured in the proposed method with the values in the literature.

INTRODUCTION

Regional mapping of cerebral venous oxygenation fraction (Yv) is required for hemodynamic assessment of brain oxygen consumption. MRI techniques for voxel‐wise Yv mapping are mainly based on the fact that hemoglobin changes its magnetic property from diamagnetic to paramagnetic when dissociated with oxygen, such as asymmetric spin echo, , respiratory‐calibrated fMRI, , , , , quantitative BOLD (qBOLD), , , , , , quantitative susceptibility mapping (QSM) with microvascular , , or microvascular models, , , the combination of qBOLD and QSM, , , and quantitative venous T2 (T2,v) methods. , , , Compared to the other methods that rely on sophisticated signal models of the extravascular tissue and the deoxygenated venous blood, T2‐oximetry methods such as QUIXOTIC , and VSEAN , were developed with direct isolation of the local venous blood signal and the measurement of T2,v.

In contrast to the more widely used T2‐oximetry methods that obtain global Yva from large cerebral veins, , , regional Yv mapping techniques are inherently SNR‐limited due to the limited amount of venous blood in each voxel, with a volume fraction of only about half of the total cerebral blood volume (CBV) and 2%–3% of the cerebral parenchymal volume. , ,

The QUIXOTIC method first utilized a “sinc” modulated velocity‐selective saturation (VSS) pulse train followed by a NSI pulse to null the arterial blood signal, which also attenuated the venous signal substantially during the outflow. The second VSS module was applied right before the acquisition with interleaved label and control pulse trains to subtract out the signal from the static tissue, which can introduce considerable error from the much larger tissue background signal. To increase SNR, the VSEAN method applied a slab‐selective inversion under the imaging plane to null the incoming arterial blood while retaining the venous blood signal in the imaging plane. However, this approach is sensitive to the arterial transit time, which is not ideal for a large spatial coverage or patients with ischemic regions. For better suppression of static tissue signal, VSEAN used a “sine” modulated VS module twice to specifically excite the slow‐moving spins without control‐label acquisitions and subtraction. Although this unique technique would alleviate the background noise from static tissues, the double VS module application results in additional signal attenuation. Furthermore, VSEAN requires a phase‐based signal projection procedure for flow signal separation, which is less straightforward than subtraction‐based methods.

Based on the laminar flow distribution within the vessels, the “sinc” modulated VSS pulse trains saturate the magnetization of spins flowing above a cutoff velocity (Vcut) while preserving the magnetization of spins flowing below Vcut. VSS‐based label and control modules have been employed for MR angiography (MRA) , , as well as measuring cerebral blood flow, , total CBV, and venous CBV (vCBV). Conversely, the emerging Fourier‐transform–based velocity‐selective (FT‐VS) pulse trains preserve magnetization of spins flowing above a Vcut and saturate (FT‐VSS) or invert FT‐velocity selective inversion (FT‐VSI) the magnetization of spins flowing below Vcut, and have been applied to MRA , , , , , , as well as measuring cerebral blood flow. , , , , , , When quantifying total CBV, FT‐VSS–based label and control modules have been shown to achieve more effective suppression of the tissue background noise than VSS preparations. Recently, a technique for estimating vCBV was proposed by using FT‐VSI and NSI pulses for arterial‐nulling and FT‐VSS–based label and control modules for tissue suppression. Compared to the VSS‐based arterial‐nulling as used in QUIXOTIC, FT‐VSI can potentially reduce the attenuation of the venous signal during the outflow and achieve higher SNR.

Obtaining images at multiple echo times within each TR has been demonstrated to be feasible for a more efficient T2,v fitting. A further consideration for cerebral Yv mapping is to mitigate CSF contamination, which arises from either improper labeling due to the pulsatile effect or unmatched diffusion weightings between VS label and control modules. Because the T2 value of the CSF (T2,CSF) is much longer than T2,v, images acquired at a long TE can be employed to account for the CSF partial volume effects as proposed previously. ,

In this work, we propose an FT‐VS–based cerebral Yv mapping method by combining the FT‐VS–based preparations to isolate the venous signal with high SNR and a multi‐echo acquisition scheme to measure T2,v and correct for CSF contamination. , The T2,v and Yv quantifications were compared between the FT‐VS–based and the existing VSS‐based (QUIXOTIC) Yv mapping preparations among healthy volunteers at 3 T.

METHODS

Pulse sequence

The sequence diagram for FT‐VS–based Yv mapping is shown in Figure 1A. It has the same preparation modules and timing as the sequence for vCBV mapping. A slab‐selective presaturation pulse train is used at the start for resetting any magnetization from the previous history, followed by a postsaturation delay of 2500 ms. A spatially nonselective arterial‐nulling module, which consists of an FT‐VSI pulse train plus an NSI pulse, is followed by an outflow time of 1050 ms based on the arterial T1 at 3 Tesla (T) of 1.89 s for a hematocrit (Hct) of 0.42 , and an FT‐VSI + NSI inversion efficiency of 0.86. Interleaved scans with FT‐VSS label and control pulse trains applied at the end of TO and before imaging are acquired to effectively subtract out the signal of static tissue. Two sets of FT‐VS pulse train parameters were compared. One set of the FT‐VSS and FT‐VSI pulse train blocks have a duration of TVS = 96 ms with triangular gradient lobes (GVS = 29 mT/m; length: 1.2 ms; ramp time: 0.6 ms; foot–head direction), which yielded a Vcut = 0.5 cm/s when defined as the half‐width–half‐maximum point of the main lobe of the velocity response of the FT‐VS pulse train. Note that, although the same RF and gradient configurations are used, the previous description of Vcut = 0.7 cm/s was defined based on the gradient first moment. The definition of Vcut for FT‐VS pulse trains adopted in this study allowed closer matching of the labeling with the VSS pulse trains when assuming a laminar flow distribution. The other set of FT‐VS pulse trains was applied with a duration of 88 ms with triangular gradient lobes (17 mT/m; length: 0.8 ms; ramp time: 0.4 ms), yielding a Vcut = 1.4 cm/s for comparison.

FIGURE 1

Diagrams of the (A) FT‐VSS and (B) VSS–based (QUIXOTIC) venous oxygenation (Yv) mapping methods, including a slab‐selective presaturation with a postsaturation delay of ((A) 2500 ms; (B) 2800 ms); an arterial‐nulling module ((A) FT‐VSI + NSI with outflow time TO = 1050 ms; (B), VSS‐TO1‐NSI‐TO2, TO1/TO2 = 400/325 ms); label/control pulse trains ((A) FT‐VSS; (B) VSS) for tissue suppression, fat‐suppression; and (C) a multi‐echo readout with 8 pairs of HT refocusing pulses (total of 16 echoes, echo spacing = 15 ms). Simulated magnetization evolution of the arterial (red), capillary (black), and venous (blue) blood during the preparation for both methods are displayed with the T2‐induced signal drop at the FT‐VSI/FT‐VSS (TVS = 96 ms, Vcut = 0.5 cm/s) and VSS (TVS = 25 ms, Vcut = 1.0 cm/s) pulse trains included. (D) Location of the imaging FOV (dashed red box) and the presaturation (yellow‐shaded rectangular area).FT‐VSI, Fourier‐transform–based velocity‐selective inversion; FT‐VSS, Fourier‐transform–based velocity‐selective saturation; HT, hyperbolic tangent; IJV, internal jugular vein; NSI, nonselective inversion; QUIXOTIC, quantitative imaging of extraction of oxygen and tissue consumption; TO, outflow time; tSNR, temporal SNR; V

The sequence diagram for VSS–based Yv mapping is shown in Figure 1B, which has the same preparation modules and timing as the QUIXOTIC studies. , The VSS pulse trains had a duration of TVS = 25 ms with trapezoid gradient lobes (GVS = 16 mT/m; length: 2.0 ms; ramp time: 0.5 ms) yielding a Vcut = 1.0 cm/s when defined as the first zero‐crossing of the velocity response under the assumption of laminar flow. Note that this Vcut is with regard to the mean blood velocity in the vessels, whereas the previous description of Vcut = 2.0 cm/s , referred to the maximal blood velocity, which is twice the mean velocity.

The simulated magnetization evolutions of the blood spins flowing into the arterioles and venules are illustrated under each sequence diagram. Unlike previous studies, , the T2 attenuation effects of both the capillary and venous blood during the first and second VS pulse trains were taken into account in the present simulation as done for vCBV mapping : assuming Hct of 0.42; Yv of 0.6 for venous blood; and a modeled capillary oxygen distribution, T2,v = 71 ms, T2,c = 118 ms, for FT‐VS and VSS pulse trains with 6 ms and 12.5 ms interecho spacing, respectively. The corresponding T2 attenuation effects were calculated as −0.62 (0.62 after NSI) for capillary blood and 0.61 for venous blood during the FT‐VSI and FT‐VSS (Figure 1A), and as 0.79 for capillary blood and 0.70 for venous blood during the first and second VSS pulse trains (Figure 1B), respectively. Hence the estimated magnetization of venous blood after the second VS pulse trains for FT‐VS and VSS‐based preparations were 0.42 and − 0.27, respectively.

A multi‐echo acquisition was employed to obtain T2,v efficiently. The last echo (TE = 240 ms) was used to detect the residual CSF signal, which was then used to correct its contamination of the T2 fitting for venous blood.

Experiments

Experiments were conducted on a 3 T scanner (Ingenia, Philips Healthcare, Best, The Netherlands) using the body coil for RF transmission (maximum amplitude 13.5 μT) and a 32‐channel head coil for signal reception. The maximum strength and slew rate of the gradient coils were 45 mT/m and 200 mT/m/ms, respectively. The protocol was approved by the institutional review board of Johns Hopkins University School of Medicine, Baltimore, MD. Six healthy volunteers (2 females, 4 males, 34 ± 7 years old) participated in this study, and all provided written informed consent. The FT‐VS–based Yv mapping sequence with Vcut of 0.5 cm/s and the VSS‐based one with Vcut of 1.0 cm/s were applied for each subject. The FT‐VS–based sequence using Vcut of 1.4 cm/s was applied to 4 of them.

For both the FT‐VS‐ and VSS‐based Yv mapping methods, a total of 16 echoes were acquired with an echo spacing of 15 ms in each TR, and a 2D single‐shot EPI was performed as the readout in each echo: a single slice of 10 mm thickness, FOV = 252 × 252 mm2 with the acquisition matrix of 32 × 32 and resolution = 7.9 × 7.9 mm2, EPI factor = 13, SENSE factor = 2.5. Hyperbolic tangent pulses (4.0 ms, frequency sweep of 11 500 Hz) were used as refocusing pulses with a M.H. Levitt (MLEV)‐16 phase‐cycling pattern through the 16 echoes. To further alleviate the spurious signal from the imperfect refocusing, excitation pulses of the successive dynamic scans were phase‐cycled with 0 and π alternatively. TR was 4.5 s, and 32 dynamic scans were obtained with a total scan time of 5.2 min.

To obtain voxel‐wise T2,CSF maps for correcting CSF partial volume for each subject (detailed in Data Analysis below), a sequence modified from a previous work (Supporting Information Figure S1) was conducted, in which a 600 ms T2prep module was applied before the multi‐echo acquisition to suppress all other signals except CSF. A total of 32 echoes with echo spacing of 15 ms were acquired with the same acquisition parameters as the readout used in T2,v measurements (TR = 10 s; 1.1 min).

With the same FOV and resolution, 2 double inversion recovery (DIR) images were collected to visualize gray matter (TI1 = 3.58 s; TI2 = 0.48 s) and white matter (TI1 = 4.05 s, TI2 = 0.77 s), respectively (TR = 10 s; 0.9 min). The blood T1 and T2 were measured at the internal jugular vein , to quantify individual Hct and global Yv , with about 1.0 min for each scan.

Data analysis

MatLab (MatLab R2018b, MathWorks, Natick, MA) was used for data processing. Only even echoes were fitted with a mono‐exponential decay model as a function of TEs for both T2,v and T2,CSF values. The T2,CSF maps were first derived from the multi‐echo CSF images: For T2,v quantification, both the T2,CSF map and the difference image between label and control scans of the last echo (SIdiff,16) were used to estimate the intensity of remaining CSF in the difference image of the nth echo (SIdiff,n,CSF, n = 1,2,…15) : where ∆TE = 15 ms. Hence the CSF‐corrected difference image of the nth echo (SIdiff,n,CSF−corrected) is obtained by subtracting out the CSF contamination (SIdiff,n,CSF): The corrected images of all 8 even echoes were then smoothed with a 10 mm FWHM Gaussian kernel, then fitted voxel‐by‐voxel using nonlinear least square algorithm to obtain the T2,v map: where Sv is the obtained signal intensity of venous blood at each voxel and S0 is the fitted amplitude before T2 decay. Note that a coefficient 0.935 was multiplied to the fitted T2,v values to correct the overestimation induced by the 4 ms pulse length of hyperbolic tangent refocusing pulse during the 15 ms refocusing interval (Supporting Information Figure S2). The standard errors of the fitted T2,v were reported.

The obtained T2,v map was converted to the Yv map using the previously developed T2‐Y calibration model with the Hct value estimated from the blood T1 measured at internal jugular vein (IJV). , Global Yv was obtained from the blood T2 measured at internal jugular vein with the same calibration model. The temporal SNR (tSNR) of the second echo through all the dynamics was also computed for each voxel. The DIR images for gray matter (GM) and white matter (WM) were used to build a mask to calculate averaged Yv in GM and WM, respectively.

RESULTS

One subject's control and label difference images at 4 different echoes (SIdiff,n, n = 2, 4, 8, 16) in FT‐VS and VSS‐based scans are shown in Figure 2A. The corresponding difference images after removing the CSF signal (SIdiff,n,CSF−corrected) are shown in Figure 2B. The fraction of remaining CSF in the original difference images at the second echo (SIdiff,n,CSF / SIdiff,n) is displayed in Figure 2C for the 3 sequences. Figure 2D shows that the fitted T2,v with CSF correction (shown in red) reduces the overestimation bias (shown in black). Both FT‐VS–based sequences contained less CSF contamination than the VSS‐based one (Figure 2C) (Supporting Information Table S1).

FIGURE 2

Difference images of the second, fourth, eighth, sixteenth echoes (A) before and (B) after CSF correction of a representative subject for the FT‐VS–based methods with Vcut = 0.5 cm/s and 1.4 cm/s; and the VSS‐based method with Vcut = 1.0 cm/s. (C) CSF fraction in the original difference images at the second echo. (D) T2 fitting of 1 representative voxel (red arrow at the second echo) before (black) and after (red) CSF correction; separately acquired (E) CSF‐weighted and (F) DIR‐based GM‐weighted images.DIR, double inversion recovery; GM, gray matter.

Figure 3 arrays another subject's signal intensity and tSNR of the CSF‐corrected difference images at the second echo acquired by the 3 sequences, and the maps of quantified T2,v along with the fitting errors (T2,v,error). Unlike the difference images (Figure 2B), which have a similar contrast as DIR GM, the fitted T2,v maps and the corresponding Yv maps show more uniform intensities between GM and WM (Figure 4), as expected for brain OEF. For each sequence, the higher tSNR yielded lower T2,v,error in GM than WM. The tSNR improvement of FT‐VS over VSS methods is more notable in GM. Figure 4 shows the DIR GM images, the maps of fitted T2,v and converted Yv for all 6 subjects, created by applying FT‐VS–based protocol using Vcut of 0.5 cm/s.

FIGURE 3

Signal intensity and tSNR of the difference images at the second echo after CSF correction of another subject by the FT‐VS–based methods with Vcut = 0.5 cm/s, 1.4 cm/s; VSS‐based method with Vcut = 1.0 cm/s; and corresponding maps of fitted T2,v and fitting errors. DIR‐based GM‐weighted image is shown as the anatomical reference

FIGURE 4

DIR‐based GM‐weighted images, fitted T2v based on the FT‐VS–based method with Vcut = 0.5 cm/s, and converted Yv maps for all 6 subjects

Table 1 compares the normalized signal intensity and tSNR of the CSF‐corrected difference images at the second echo averaged across the slice for each subject, their T2,v with T2,v,error, and converted Yv values with comparison to the global Yv measured at internal jugular vein. The averaged signal intensities of the difference images by the FT‐VS protocol using Vcut = 1.4 cm/s and the VSS protocol using Vcut = 1.0 cm/s were 31% and 58% lower than those measured by the FT‐VS protocol using Vcut = 0.5 cm/s (1.00 vs. 0.69 and 0.42). The tSNR in FT‐VS experiments was more than twice that in the VSS experiment (5.9 ± 0.9 and 4.8 ± 1.2 vs. 2.3 ± 0.7), which led to 17% and 11% lower T2,v,error comparing FT‐VS with VSS experiments (5.3 ± 0.5 and 5.7 ± 0.5 vs. 6.4 ± 0.8). Meanwhile, the whole‐slice averaged Yv values measured in the 2 FT‐VS–based (Vcut = 0.5 cm/s and 1.4 cm/s) and 1 VSS‐based (Vcut = 1.0 cm/s) experiments were close to the global Yv values, as 0.61 ± 0.02, 0.60 ± 0.03, 0.58 ± 0.02, and 0.60 ± 0.03, respectively.

TABLE 1

Whole‐slice averaged signal intensity, tSNR of the difference image at the second echo, the whole‐slice averaged T2,v measured by the FT‐VS (Vcut = 0.5 and 1.4 cm/s) and VSS (Vcut = 1.0 cm/s) experiments, the fitting errors T2,v,error, and the correspondingly converted Yv compared with the global Yv measured at IJV

	FT‐VS (0.5 cm/s)					FT‐VS (1.4 cm/s)					VSS (1.0 cm/s)
Subject	Signal a	tSNR	T2,v (ms)	T2,v,error (ms)	Yv	Signal a	tSNR	T2,v (ms)	T2,v,error (ms)	Yv	Signal a	tSNR	T2,v (ms)	T2,v,error (ms)	Yv	Global Yv
1	1	5.2 ±2.1	56 ±11	5.7 ±4.1	0.60 ±0.06						0.42	2.7 ±1.4	54 ±12	6.7 ±3.4	0.59 ±0.07	0.63
2	1	7.2 ±2.9	63 ±11	4.8 ±2.9	0.60 ±0.05						0.43	3.4 ±1.7	58 ±12	5.7 ±3.6	0.57 ±0.06	0.56
3	1	4.9 ±2.0	59 ±12	4.8 ±2.8	0.62 ±0.07	0.56	3.6 ±1.4	61 ±15	6.2 ±4.0	0.62 ±0.08	0.40	1.6 ±0.9	51 ±12	5.4 ±2.5	0.57 ±0.08	0.63
4	1	6.0 ±2.6	60 ±11	5.3 ±3.4	0.63 ±0.05	0.67	6.1 ±2.5	59 ±14	6.1 ±4.2	0.62 ±0.09	0.36	2.2 ±1.0	59 ±15	6.0 ±3.6	0.62 ±0.09	0.62
5	1	6.6 ±3.3	56 ±11	5.0 ±3.5	0.58 ±0.06	0.83	5.4 ±2.3	56 ±14	5.3 ±3.4	0.56 ±0.09	0.45	1.8 ±1.0	54 ±13	7.2 ±4.7	0.55 ±0.09	0.60
6	1	5.6 ±1.8	59 ±10	6.1 ±4.5	0.62 ±0.05	0.71	4.0 ±1.6	59 ±16	5.3 ±4.8	0.60 ±0.11	0.48	2.1 ±1.2	54 ±14	7.2 ±4.4	0.59 ±0.09	0.57
Mean ±Std	1	5.9 ±0.9	59 ±3	5.3 ±0.5	0.61 ±0.02	0.69 ±0.11	4.8 ±1.2	59 ±2	5.7 ±0.5	0.60 ±0.03	0.42 ±0.04	2.3 ±0.7	55 ±3	6.4 ±0.8	0.58 ±0.02	0.60 ±0.03

The averaged signal intensity of the CSF‐corrected difference images at the second echo measured at different experiments normalized by those measured by the FT‐VS protocol using Vcut = 0.5 cm/s for each volunteer.

FT‐VS, Fourier‐transform–based velocity‐selective; IJV, internal jugular vein; tSNR, temporal SNR; Vcut cutoff velocity; VSS, velocity‐selective saturation; Yv, venous oxygenation fraction.

DISCUSSION

In this work, an improved technique is introduced for quantifying regional Yv by embedding an FT‐VS–based preparation module to suppress the static tissue and arterial blood signal and a multi‐echo acquisition scheme for T2 measurement and CSF correction. Compared to the VSS‐based preparation (QUIXOTIC), this method yielded higher venous signal intensity (and tSNR) and less CSF contamination. The obtained Yv values are comparable to those reported in the literature , and have good agreement with the global Yv measured individually (Table 1).

Higher SNR of local venous blood signal would reduce the error of multi‐echo fitting for T2,v. Based on the theoretical estimation, the FT‐VS–based preparation can intrinsically improve the magnetization of venous blood by 56% over VSS‐based approach (0.42 vs. 0.27) (Figure 1), largely driven by employing FT‐VSI plus immediate NSI for arterial nulling instead of using VSS followed by NSI after a delay. As explained in the original QUIXOTIC paper, the labeled venous volume depends on both the venous outflow time TO between the 2 VS pulse trains and the employed Vcut (the trailing edge). Our experimental results showed 157% and 109% improvement of tSNR of the isolated venous blood signal by the FT‐VS preparations with both the higher and lower Vcut (0.5 cm/s, 1.4 cm/s) over the VSS‐based method (1.0 cm/s). The in vivo tSNR advantage could be in part due to the 45% longer outflow time for the labeled venous bolus, with TO = 1050 ms between the FT‐VSI and FT‐VSS pulse trains and 725 ms between the first and second VSS pulse trains. Note that to null the arterial blood at 3 T, the TO following the FT‐VSI / NSI pulses is fixed to be 1050 ms, during which the leading edge of the venous bolus should be still within or close to the voxels of the capillary source, and the measured T2,v values reflect the local venous oxygenation in order to ensure the spatial specificity of the Yv maps. This condition is supported by the close resemblance between the DIR GM images and the venous blood signal obtained at the second echo in this study (Figures 2, 3) or the vCBV maps derived from the same preparation modules in a prior study. Meanwhile, FT‐VS experiments using Vcut of 0.5 cm/s had 23% higher venous blood signal (5.9 vs. 4.8) (Table 1) than using Vcut of 1.4 cm/s, indicating that lower Vcut could increase the venous blood signal being detected. As discussed before, the chosen Vcut determines the trailing edge of the venous bolus and ideally should be just above capillary blood velocities (close to 0.1 cm/s ) to achieve the maximum available venous signal. However, utilizing a smaller Vcut value on clinical scanners is limited by the gradient maximal strength and slew rate as well as potential artifacts caused by eddy current or diffusion mismatch between label and control scans.

CSF contamination is another factor that affects the accuracy of T2 measurement. To correct CSF contamination, Stout et al. applied a fixed magnetization fraction of CSF in all GM voxels (0.112) obtained from a separate calibration experiment. Instead, our protocol observed about ∼20%–40% CSF fractions contained in the difference images through direct estimation (Supporting Information Table S1). The measured average CSF T2 for all our volunteers was 1.2 ± 0.2 s, which is comparable to previously measured T2 values of CSF in brain tissue. , , Therefore, instead of employing an extra scan to map CSF T2, it would be faster and simpler to just use 1.2 s as a fixed CSF T2 value for all voxels.

To accurately measure Yv, conversion of venous blood T2 value to Yv is also very important. The T2,v value depends highly on the interval between the refocusing pulses (τ). Currently, however, blood T2‐Yv calibration curves , are available for only a limited set of τ values. This induces the inconvenience of choosing the interval between refocusing pulses, especially for our multi‐echo acquisition design in which the interval of refocusing pulses is limited by other acquisition parameters. Recently, a blood T2‐Yv model has been proposed that can calculate blood T2 based on B0, τ, and Hct with good agreement with previous experimental results. A website is also available to conveniently convert blood T2 to Yv. Note that Hct measured from blood T1 values is an efficient noninvasive approach that not only facilitates T2‐Yv calibration but also assists evaluation of the between‐subject variations for perfusion and functional MRI studies.

Despite of achieving higher SNR than VSS‐based method, to ensure acceptable levels of accuracy and precision for the Yv estimation per voxel, the current FT‐VS–based T2‐oximetry method only acquired a single slice Yv map with a low spatial resolution (voxel size of 7.9 × 7.9 × 10.0 mm3) over 5 min. Although the current spatial resolution is comparable to the range in [15O]‐gas PET studies, further technical development is still warranted to extend its implementation from 2D single‐slice imaging to 3D whole‐brain coverage using segmented acquisitions to ensure wider utility of this method for clinical use. Denoising with deep‐learning approach , can also be employed for obtaining higher spatial resolution.

CONCLUSION

Employing advanced velocity‐selective pulse trains, an improved venous oxygenation mapping technique is proposed with minimal sensitivity to arterial transit delay, high venous signal, and mitigation of CSF contamination, which are important to improve the accuracy and precision of T2,v measurements. The estimated Yv values are similar between GM and WM and comparable to the values measured globally.

CONFLICT OF INTEREST

Dr. van Zijl is a paid lecturer for Philips Healthcare, has research support from Philips, and has technology licensed to Philips. This has been approved by the Committee on Conflict of Interest of Johns Hopkins University.

Figure S1: (a) The diagrams of the 2D CSF T2 mapping method, including a slab‐selective pre‐saturation with a post‐saturation delay (9000 ms), a T2 preparation (600 ms) to suppress other tissues with short T2s, fat‐suppression, and a multi‐echo readout with 16 pairs of hyperbolic tangent (HT) refocusing pulses (total of 32 echoes, echo spacing = 15 ms).

Figure S2: Calculation (a) and demonstration (b) of a multiplication coefficient to correct the T2 overestimation due to the 4 ms pulse length of HT refocusing pulses during the 15 ms refocusing interval.

Table S1: The whole‐slice averaged remaining CSF fraction in the difference images at the 2ndecho by FT‐VS (Vcut = 0.5 and 1.4 cm/s) and VSS (Vcut = 1.0 cm/s) experiments.

Click here for additional data file.