Advertisement
Browse Subject Areas
?

Click through the PLOS taxonomy to find articles in your field.

For more information about PLOS Subject Areas, click here.

  • Loading metrics

Open Access

Peer-reviewed

Research Article

Regional Reproducibility of BOLD Calibration Parameter M, OEF and Resting-State CMRO2 Measurements with QUO2 MRI

  • Isabelle Lajoie ,

    isabelle.lajoie@mail.mcgill.ca

    Affiliations Département de physiologie moléculaire et intégrative, Institut de génie biomédical, Université de Montréal, Montreal, Quebec, Canada, Department of Neurology and Neurosurgery, Montreal Neurological Institute, McGill University, Montreal, Quebec, Canada

    ⨯
  • Felipe B. Tancredi,

    Affiliation Departamento de Radiologia, Centro de Pesquisa em Imagem, Hospital Israelita Albert Einstein, São Palo, SP, Brazil

    ⨯
  • Richard D. Hoge

    Affiliation Department of Neurology and Neurosurgery, Montreal Neurological Institute, McGill University, Montreal, Quebec, Canada

    ⨯

Abstract

The current generation of calibrated MRI methods goes beyond simple localization of task-related responses to allow the mapping of resting-state cerebral metabolic rate of oxygen (CMRO2) in micromolar units and estimation of oxygen extraction fraction (OEF). Prior to the adoption of such techniques in neuroscience research applications, knowledge about the precision and accuracy of absolute estimates of CMRO2 and OEF is crucial and remains unexplored to this day. In this study, we addressed the question of methodological precision by assessing the regional inter-subject variance and intra-subject reproducibility of the BOLD calibration parameter M, OEF, O2 delivery and absolute CMRO2 estimates derived from a state-of-the-art calibrated BOLD technique, the QUantitative O2 (QUO2) approach. We acquired simultaneous measurements of CBF and R2* at rest and during periods of hypercapnia (HC) and hyperoxia (HO) on two separate scan sessions within 24 hours using a clinical 3 T MRI scanner. Maps of M, OEF, oxygen delivery and CMRO2, were estimated from the measured end-tidal O2, CBF0, CBFHC/HO and R2*HC/HO. Variability was assessed by computing the between-subject coefficients of variation (bwCV) and within-subject CV (wsCV) in seven ROIs. All tests GM-averaged values of CBF0, M, OEF, O2 delivery and CMRO2 were: 49.5 ± 6.4 mL/100 g/min, 4.69 ± 0.91%, 0.37 ± 0.06, 377 ± 51 μmol/100 g/min and 143 ± 34 μmol/100 g/min respectively. The variability of parameter estimates was found to be the lowest when averaged throughout all GM, with general trends toward higher CVs when averaged over smaller regions. Among the MRI measurements, the most reproducible across scans was R2*0 (wsCVGM = 0.33%) along with CBF0 (wsCVGM = 3.88%) and R2*HC (wsCVGM = 6.7%). CBFHC and R2*HO were found to have a higher intra-subject variability (wsCVGM = 22.4% and wsCVGM = 16% respectively), which is likely due to propagation of random measurement errors, especially for CBFHC due to the low contrast-to-noise ratio intrinsic to ASL. Reproducibility of the QUO2 derived estimates were computed, yielding a GM intra-subject reproducibility of 3.87% for O2 delivery, 16.8% for the M value, 13.6% for OEF and 15.2% for CMRO2. Although these results focus on the precision of the QUO2 method, rather than the accuracy, the information will be useful for calculation of statistical power in future validation studies and ultimately for research applications of the method. The higher test-retest variability for the more extensively modeled parameters (M, OEF, and CMRO2) highlights the need for further improvement of acquisition methods to reduce noise levels.

Citation: Lajoie I, Tancredi FB, Hoge RD (2016) Regional Reproducibility of BOLD Calibration Parameter M, OEF and Resting-State CMRO2 Measurements with QUO2 MRI. PLoS ONE 11(9): e0163071. https://doi.org/10.1371/journal.pone.0163071

Editor: Jeroen Hendrikse, Universitair Medisch Centrum Utrecht, NETHERLANDS

Received: July 14, 2016; Accepted: September 1, 2016; Published: September 20, 2016

Copyright: © 2016 Lajoie et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.

Data Availability: Our local ethics committee does not allow the deposition of raw imaging data into publicly accessible repositories. Anonymized raw data will nevertheless be made available upon request, subject to review and approval by the Comité d'éthique de la recherche vieillissement-neuroimagerie. Please address requests for data access to data.requests@bic.mni.mcgill.ca.

Funding: This study was supported by the Canadian Institutes of Health Research MOP 123492 to RDH, Canada Foundation for Innovation 17380 to RDH, Natural Sciences and Engineering Research Council of Canada R0018142 to RDH, Fonds Québécois de la Recherche sur la Nature et les Technologies (CA) Mrs. Isabelle Lajoie Le Consortium québécois sur la découverte du médicament (CQDM) to RDH. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

Competing interests: The authors have declared that no competing interests exist.

Introduction

Mapping of resting metabolism in the brain is of considerable interest for diagnostic and research applications. Until recently, positron emission tomography (PET) using a triple injection of radio-labeled O2 was the only imaging method for measuring cerebral metabolic rate of O2 consumption (CMRO2) [1]. The PET method requires exposure to ionizing radiation, arterial sampling, and access to an on-site cyclotron to produce the short-lived 15O-labeled tracers, limitations that have led to the development of magnetic resonance imaging (MRI) techniques to measure O2 consumption [2–4].

The approach proposed by our team, Quantitative O2 (QUO2) MRI is based on respiratory calibration of the BOLD signal, in which the oxygen extraction fraction (OEF) at rest is determined using hypercapnia (HC) and hyperoxia (HO). During the respiratory manipulation, we monitor end-tidal O2 (ETO2) levels and use dual-echo ASL to measure BOLD and cerebral blood flow (CBF) simultaneously. ETO2, BOLD and CBF then serve as inputs to the generalized calibration model (GCM) described in Gauthier and Hoge [5], which yields a system of two equations with solutions for the BOLD calibration parameter M, i.e. the maximum BOLD signal increase when venous O2 saturation approaches 100%, and resting OEF. Multiplication of OEF by baseline CBF and arterial O2 content (estimated from ETO2 monitoring and, optionally, blood testing) gives resting CMRO2 in absolute units, e.g. μmol/100 g/min.

An initial proof-of-concept of the QUO2 method was presented in Gauthier and Hoge [6] to obtain individual and group maps of BOLD calibration parameter M, resting OEF and CMRO2. While valid regional and group-averaged estimates of the latter parameters were obtained, individual OEF maps showed signs of instability characterized by large fluctuations in the modeled values and a lack of solution in certain regions. The stability of individual solution maps generated using this method depends on accurate and robust measures of end-tidal O2 and maps of fractional changes in BOLD and CBF during the respiratory manipulation. In an attempt to improve the stability and avoid circumstances where the QUO2 model cannot be solved, we have adapted the imaging and respiratory protocols used in several ways. Instead of performing two separate respiratory scans for hyperoxia and hypercapnia, we have adopted the 18-minute respiratory sequence that alternates between periods of hypercapnia and hyperoxia [2] and during which the total time dedicated to each gas manipulation is increased compared to the original protocol, improving statistical power. We developed a simple breathing circuit allowing a better control over fractional concentration of inspired gas and thus yielding improved stability of end-tidal values compared with the simple oxygen masks used in our earlier studies [7]. Combining a dual-echo Pseudo-Continuous ASL sequence (de-pCASL), the respiratory protocol and breathing circuit mentioned above, we assessed the test-retest reliability in the respiratory responses and in CBF and BOLD responses within GM [8]. In the present study, we attempted to further improve the voxel-wise image quality in single-subject maps by performing an integrated analysis on the dual-echo pCASL data. The novel analysis strategy (further described in Materials and Methods) involved 1) motion correcting the interleaved echo series using the same transformation for the two echo times; 2) applying a more holistic general linear model on the motion-corrected series to extract baseline parameters and gas responses using one regressor per combination of echo and tag/control; 3) employing both echoes information to estimate R2* rather than approximating it from the second echo; and 4) employing a more sophisticated approach to mitigate solution instabilities from isolated non-parenchymal voxels using 3D median filtering.

The aim of this study was to assess, in a small cohort of healthy individuals, the variability of the optimized QUO2 measures across and within subject in different brain regions. These results, based on the precision of the method, will help guide future developments and research application of the method. The following estimates were obtained: BOLD calibration parameter M, OEF, O2 delivery and resting-state CMRO2. The impact of systematic and random errors on the accuracy and precision of such estimates was also evaluated. Furthermore, these estimates were compared with CBF and BOLD-based reproducibility estimates derived from the same group of subjects and same enhanced analysis.

Materials and Methods

Eight healthy adults were enrolled in this study (4 females, mean age: 30.5 ± 5.7 years). All participants gave written informed consent and the project was approved by the Comité mixte d’éthique de la recherche du Regroupement Neuroimagerie/Québec. They were scanned twice (referred to as Test A and Test B), within 24 hours, using the same imaging procedures and respiratory manipulation. To minimize effects of diurnal fluctuation in blood flow, all sessions was acquired between 2 PM and 6 PM [9]. The participants were asked to abstain from caffeine 3 hours prior to scanning.

Respiratory Manipulation

For the respiratory manipulation, we adopted the gas sequence described by Bulte et al [2] with a total duration of 18 minutes. This involves two 2-min periods of hypercapnia (HC) and two 3-min periods of hyperoxia (HO). HC was followed by a 1-min normocapnic period and then the 3-min hyperoxic stimulus. HO was followed by a 3-min period of normoxia.

HC and HO blocks were induced by supplying participants with gas mixtures enriched with CO2 or O2. Participants breathed the gas mixtures through the breathing circuit developed in-house [7]. An automated system, also developed in-house, was used to deliver the gas mixtures with a reproducible time course in all scanning sessions. The system comprises 4 flow controllers (SideTrak® 840, Sierra instruments, L. Monterey, CA, USA), a ~25 mL mixing chamber, a digital interface (Sierra, FloBoxTM 954) to send commands to the flow controllers and a laptop computer to automate the gas mixture processes and collect behavioral data. The system’s output is connected to the breathing circuit via 10 meters of plastic tubing (BIOPAC, AFT31-MRI). Three gases were input to the flow controllers: medical air, oxygen and a 5% CO2 and air mixture. During the hyperoxia periods, subjects breathed a mixture of 50% pure oxygen balanced with air, yielding a fix inspired O2 concentration of 60% O2. Otherwise, participants were given medical air to breathe. Gas mixtures were administered at a rate of 20 L/min, except during transitions in inspired concentrations, during which the flow rate was increased to 50 L/min for 5.4 seconds in order to accelerate transitions.

Respiratory gases were continuously monitored using the CO2100C and O2100C modules of a BIOPAC MP150 system (BIOPAC Systems Inc., CA, USA). Gases were sampled via a 10m segment of rigid tubing (AFT31-XL, from BIOPAC System Inc.) in series with a bacterial filter (#2200/01, GVS filter technology, LA, UK) and 1’ segment of oxygen tubing attached to the sampling port of the respiratory circuit.

Image Acquisition

Images were acquired on a clinical 3 T scanner (Siemens TIM TRIO, Siemens Medical Solutions, Erlangen, Germany) using the vendor’s 32-channel receive-only head coil. The scan session included a 5-minute anatomical acquisition (1 mm3 MPRAGE with TR/TE/flip angle = 2.3 seconds/3 msec/9°, 256x240 matrix, GRAPPA factor = 2), and an 18-minute functional scan using dual-echo pseudo-continuous ASL sequence (de-pCASL) [10] in order to acquire simultaneous measures of BOLD and CBF. The de-pCASL parameters were: TR/TE1/TE2/alpha = 4.12 seconds/8.4 msec/30 msec/90°, labeling duration = 2 seconds using Hanning window-shaped RF pulse with duration/space = 500 μsec/360 μsec, flip angle = 25°, slice-selective gradient = 6 mT/m, label offset = 100 mm below the center of image slab, nominal and average post-labeling delay (PLD) = 0.9 and 1.44 seconds respectively. The readout consisted of a GRE-EPI with GRAPPA factor = 2, partial sampling of k-space = 7/8, in-plane resolution of 4.5 x 4.5 mm2, 21 slices with 4.5 mm thickness and 0.45 mm gap.

Respiratory Data Analysis

Analysis of the respiratory data was carried out using an in-house program developed in Matlab (MathWorks, Natick, MA, USA). An automatic extraction of the end-tidal (ET) and fixed inspired (FI) points from the continuous O2 and CO2 traces was performed. It was observed that the filter placed in series with the sampling line added an extra resistance causing an effect of low-pass filtering to the respiratory waveform. This resulted in an offset of both the ET and FI monitored pressures, i.e. an attenuation of the peak-to-peak amplitude of the waveform, which was dependent on the participant’s breathing pace. Each ET point was corrected using the average of the differences between the observed and expected FI points surrounding it. ET values were also corrected to account for an expired partial pressure of water of 47 mmHg [11]. The resting ET and changes in ET during HC and HO periods were determined by applying the linear model previously described [12]. The model is composed of a third-degree polynomial term and four regressors to represent responses to the hypercapnic and hyperoxic blocks. The offset term served to estimate the baseline ET whereas the effect size of each response regressor yielded an estimate of the associated ET change. Final ETO2 change to periods of HO was obtained by averaging the two ETO2 changes to HO. The same method was employed to compute the final ETO2 change to periods of HC, and ETCO2 responses to both gases.

The average values of ETO2 at baseline and during both respiratory stimuli were used to compute arterial O2 content (ml O2/ml blood) and change in the venous deoxygenated fraction ([dHb]/ [dHb]0) as in Chiarelli et al [13] and Gauthier et al [3]. The latter quantities are integrated to obtain the BOLD calibrated value M, resting OEF and CMRO2 as specified below.

Imaging Data Analysis

Preprocessing.

Analysis of functional scans was performed using in-house software implemented in C. The interleaved echo series was motion corrected with consecutive first and second echo frames sharing the same transformation matrix. The resultant series was spatially filtered (8mm FWHM 3D Gaussian kernel), had extra-cerebral voxels removed and was intensity normalized (brain mean 100). The fMRI data were then fit to a GLM to extract the label and control series of both echoes during baseline, hypercapnia and hyperoxia periods. The model used four regressors per conditions to account for both echoes label and control points, and a third-degree polynomial with an offset term representing signal drifts. We used a single-gamma HRF function with 20 seconds time-to-peak and 40 seconds width, which yielded near-exponential transitions to account for the slow response of the arterial partial pressures to the inspired gas [14]. ASL (S0) and BOLD (R2* or 1/T2*) control and label series at baseline and during gas manipulations were computed using both echoes information. ASL flow series were computed from subtraction of S0 control to S0 label series, whereas BOLD series were isolated averaging the control and label R2* series. A 3D median filtering (radius of 1 voxel) was applied on the resultant maps to minimize the impact of non-parenchymal voxels such as those containing large blood vessels.

The functional maps produced by the above analysis were then used to further reduce the impact of voxels not meeting the assumptions of the QUO2 model: Baseline T2* maps served to exclude voxels in regions degraded by susceptibility artifact (lower threshold of 30ms). Voxels in which ΔR2*HO was positive were assumed to be dominated by susceptibility artifacts from adjacent nasal cavity due to the paramagnetic effect of molecular O2. Additional voxels with positive ΔR2*HC were considered as non-parenchymal and were also excluded from the analysis. The ASL signal was converted into physiological units of flow (mL/100 g/min) as in Wang et al [15] using the constants recommended by the ISMRM Working Committee [16] and an adjusted PLD to account for slice acquisition time (PLD range for 21 slices of 900-1960ms).

During hyperoxic manipulation, the T1 of blood is altered due to an increase in plasma concentration of paramagnetic O2 [17]. To account for this change in blood T1, which would bias our Δ%CBFHO estimation, we applied a corrective factor using the approach described in Chalela et al [18] and Zaharchuk et al [19]. The T1 of blood during hyperoxic intervals was estimated individually using the R1 and PaO2 relationship in rats’ blood reported in Pilkinton et al [17].

Computation of metabolism.

For each gas challenge, the changes in the venous deoxygenated fraction, along with the change in BOLD (ΔR2*) and CBF were used as inputs to the generalized calibration model (GCM) described in Gauthier et al [5]. This yields a system of two equations with two unknowns: the BOLD calibration parameter M (extrapolated maximum BOLD fractional signal increase when venous O2 saturation approaches 100%) and OEF (the fraction of delivered oxygen that is consumed). Absolute CMRO2 was then determined by multiplying OEF by O2 delivery, computed as the product of resting CBF by arterial O2 content. In the absence of intersection in between the HC and HO curves, the voxel is said to have no solution and will later be excluded from any ROI or voxel average in M, OEF and CMRO2. Because of the low CNR of the ASL hyperoxic response, the GM-averaged value obtained from the post-T1-correction Δ%CBFHO was used as an estimate of the whole-brain post-T1-correction Δ%CBFHO. Previous studies [3,13] also report using a whole-brain estimate of Δ%CBFHO, with the difference that, in the current paper, the value was computed for each participant. In the equation defining M (Gauthier and Hoge 2012, equation 7), the parameter α, which expresses the relationship between changes in blood flow and blood volume, was assumed to be 0.18 [20] while β, defining the non-linear dependence of changes in R2* on deoxygenated hemoglobin, was set to 1.5 [21]. The hemoglobin concentration [Hb] was assumed to be 15 g Hb/dl blood, although this can be readily measured using a venous blood draw. It was also assumed that oxygen consumption remained constant during periods of hypercapnia and hyperoxia (CMRO2HC and CMRO2HO). The sensibility of QUO2 model-derived estimates to the assumed parameters was also evaluated, as detailed in the section ‘Accuracy of QUO2 model-derived estimates—sensitivity to systematic errors’.

Tissue segmentation

Automated segmentation of gray matter (GM) from the anatomical scans was carried out using the FMRIB Software Library (FSL) [22]. Structural images were extracted from T1-weighted scans using the brain extraction tool (FSL’s BET). Then, a binary mask delineating the brain was created along with a probability mask of GM employing the automated segmentation tool (FSL’s FAST). Both were resampled to the resolution of the functional EPI scans.

Regions of Interest.

In addition to the whole brain gray matter (GM), six ROIs in ICBM space were selected from OASIS-TRT-20 in three-dimensional mode [23]. The selected ROIs are located in parietal, occipital or temporal lobes, and are known to be implicated in conditions such as Alzheimer’s disease [24–29]. ROI’s, presented in Fig 1, include left and right: inferior parietal (IP), superior parietal (SP), precuneus (PRE), hippocampus (HIP), anterior (caudal and rostral) cingulate (AC) and posterior cingulate (PC). Each ROI was registered to the resolution of the functional EPI scans before being conjoined with the individual’s GM probability mask excluding voxels with a GM probability lower than 50% and non-parenchymal voxels identified previously. The resultant ROI probability masks were used to perform weighted averaging of the different metrics. Voxels where no solution was found for M and OEF were excluded when performing the ROI analysis of M, OEF and CMRO2.

thumbnail
Download:
Fig 1. Regions of interest (ROIs).

QUO2 parameters were evaluated in GM and in six ROIs selected from the OASIS-TRT-20 atlas.

https://doi.org/10.1371/journal.pone.0163071.g001

Registration.

Individual BOLD, CBF, M, OEF and CMRO2 maps were non-linearly normalized to the ICBM152 template using the CIVET software package [30] via the CBRAIN tool [31] with 12 degrees of freedom. Test A and B average maps of BOLD and CBF were computed as arithmetic means using in-house software. On their part, average maps of M, OEF and CMRO2 were obtained excluding from the average any voxels where no solution was found.

Statistical Analysis

Within each measurement, i.e. CBF, BOLD (R2*), M, OEF and CMRO2, Test A and Test B were averaged and compared across ROIs, considering P < 0.05 level of significance, correction for multiple comparisons.

Statistical tests were performed on the data to ensure it satisfied the criteria for conducting a reproducibility analysis. For ETO2 and the other parameters listed above, the distribution of differences was tested for normality using the Shapiro-Wilk W-test, and the independence between the magnitude of differences and mean of measurements was verified using a rank correlation coefficient (Kendall’s τ). If the differences distribution appeared to deviate from a normal distribution, or if the magnitude of differences increased with the mean of measurements, the data were transformed on the log10 scale and the verification was repeated. In cases where the log10 scaled data satisfied the criteria, the reproducibility was assessed on these scaled values. Otherwise, assessment of reproducibility was based on the original values along with appropriate annotation [32–34]. Additionally, to determine whether there was an order effect between the two tests, we performed a two-tailed paired t-test on each set of ROI-averaged values, considering a P < 0.05 level of significance.

ROI-averaged reproducibility was evaluated using Matlab to compute metrics that give complementary information on the agreement between repeated measures and population variance:

  1. dSD, the standard deviation of the differences between Test A and B measurements.
  2. wsSD, the within-subject standard deviation, equals dSD/√2 considering two measurements.
  3. wsCV, the within-subject (or intra-subject) coefficient of variation, as used in Floyd et al [33] and Chen et al [35]. wsCV = √ [mean of the (wsSD/subject mean)2]. wsCV provides an unbiased reproducibility measurement expressed as a percent of the mean with a low wsCV indicating a high reproducibility. When data were on the log10 scale, wsCV was approximated by antilog(wsSD)-1 [36].
  4. CR, the coefficient of repeatability [37] = 1.96*√2*wsSD or 1.96*dSD. CR gives an estimate of the range of values one would obtain in a retest measurement. Thus, 95% of repeated measures for the sample will lie between the interval mean differences ± CR (⍺ = 0.05).
  5. bsCV, the between-subject (or inter-subject) coefficient of variation as computed in Tjandra et al [38]. bsCV = SDpooledData / meanpooledData * 100.

Accuracy of QUO2 model-derived estimates—sensitivity to systematic errors

The QUO2 derived estimates accuracy rely on assumed physiological parameters such as α, β, [Hb], as well as assumed normalized CMRO2 changes during HC and HO. As sources of systematic errors, the assumed values won’t affect the reproducibility/precision analysis outcomes, however they can lead to individual inaccuracy of M, OEF and CMRO2. Using the group-averaged GM experimental Test A data, the sensitivity of QUO2 estimates to assumed parameters was evaluated by independent variation in α, β, [Hb], CMRO2HC and CMRO2HO. The values used in the previous analysis were: α = 0.18, β = 1.5, [Hb] = 15g Hb/dl blood, while isometabolism during hypercapnia and hyperoxia was considered (CMRO2HC and CMRO2HO = 1.0). Explored ranges of α and β were respectively 0.15 to 0.45 and 1.0 to 1.5, matching those in Chiarelli and al. [13]. The span of [Hb], i.e. from 11 to 17 g Hb/dl blood, was chosen as in Mark et al. [39] to take into account differences in gender and presence of anemia or polycythemia [40]. Evaluated ranges of change in CMRO2 during HC and HO were determined as in Merola et al [41]: i.e. a change of ± 1% in CMRO2 for 1 mmHg and 40 mmHg change in end-tidal CO2 and O2 respectively. We also evaluated the impact of a maximum of 10% decrease in blood flow during HO periods, as this parameter is often assumed (normalized CBFHO from 0.90 to 1.0).

Precision of QUO2 model-derived estimates—sensitivity to random errors

Within-subject precision of the QUO2 model-based M, OEF and CMRO2 estimates can be affected by a certain real physiological within-subject variability as well as random errors in the measurement of brain’s responses to hypercapnia and hyperoxia. To evaluate the effect, on OEF and M precision, of such errors in measurement, we performed an error propagation analysis of the QUO2 model employing Test A data. In addition to the analysis of errors in measured CBF during HC, also discussed in Gauthier and Hoge [5], we examined the impact of errors in measured R2* changes during both respiratory challenges. For each observed source of error, individual OEF and M values were computed based on ‘real’ GM and respiratory measures, with simulated error ranging from -33% to +33% added to the examined source. This simulated error can be translated into coefficient of variation of the observed input: the CV being the percent of variability with respect to the mean value, it is lower when an error is added to the measurement (CV of 20% for an error of +33%) rather than subtracted (CV of 28% for an error of -33%). Simulated group-average M and OEF were computed from the individual simulated values, while the latter were compared with the ‘real’ values to calculate the simulated CVs with respect to the added error.

Results

One participant was excluded from the analysis because their CBF response to CO2 during Test A was found to be an outlier (value beyond twice the standard deviation). This participant reported a high level of anxiety during Test A due to a first MRI scan and hypercapnia experience.

Gas manipulations

The average and standard deviation of ETO2 levels in Test A and B are shown in Fig 2A. Breathing medical air (~160 mmHg of O2) yielded average ETO2 levels of 111 ± 6 mmHg and 110 ± 4 for the two scan time points. During hyperoxia induction, when participants received a gas mixture with 380 mmHg of O2, ETO2 levels increased to 370 ± 9 mmHg during Test A and 374 ± 12 mmHg during Test B. During hypercapnia periods, participants demonstrated a slight increase in the minute volume ventilation, which increased ETO2 levels in comparison to air-breathing (118 ± 4 and 117 ± 4 mmHg) despite a slightly lower inhaled O2 concentration (~152 mmHg). Fig 2B shows the average and standard deviation of ETCO2 levels in Test A and B. ETCO2 levels at baseline and during hyperoxia were found to be similar, with 40 ± 2 mmHg and 39 ± 2 mmHg (in both tests) respectively. During HC, the ETCO2 levels increased to 49 ± 1 mmHg and 50 ± 2 mmHg in Test A and B respectively. Table 1 shows, for both ETO2 and ETCO2 measurements, P-values of Shapiro-Wilk W-test, Kendall’s rank coefficient and t-Student paired test, as well as wsCV and CR. All distributions of differences in tests were found to be normal (Shapiro-Wilk W-test, P > 0.05) and no dependence was observed between the differences in measurements and mean of measurements (Kendall’ τ, P > 0.2). Finally, no significant difference (t-Student paired test, P > 0.1) was found between Test A and Test B end-tidal levels, which were also reproducible (ETO2: CR<20 mmHG, wsCV<5%; ETCO2: CR<6 mmHG, wsCV<6%).

thumbnail
Download:
Fig 2. Gas manipulation.

Test A and B end-tidal O2 and CO2 values at baseline (0), during hyperoxia (HO) and during hypercapnia (HC). Errors bars indicate standard deviation.

https://doi.org/10.1371/journal.pone.0163071.g002

thumbnail
Download:
Table 1. End-tidal measurements.

https://doi.org/10.1371/journal.pone.0163071.t001

Exclusion of non-parenchymal or artifact voxels

Fig 3 shows sagittal group-averaged maps of voxels excluded from the ROI analysis given that they were considered as non-parenchymal or dominated by susceptibility artifacts, based on individual ΔR2*HC, ΔR2*HO and T2*0. Both tests presented similar patterns of excluded voxels, and, as expected, the latter were mainly situated in regions adjacent to the nasal cavity due to the paramagnetic effect of molecular O2. In Table 2, we present, for each ROI, the group average and standard deviation of the number of voxels with GM probability higher than 0.5, preceding and following the exclusion process. No significant difference was found between Test A and Test B number of pre- and post-exclusion voxels (all P > 0.23). In whole GM, 24 ± 4% were excluded from the analysis. The hippocampus and the anterior cingulate were the most affected by the procedure, with a percentage of exclusion of 31 ± 10% and 30 ± 9% respectively, whereas less than 2% of voxels were excluded in the other regions.

thumbnail
Download:
Fig 3. Exclusion of non-parenchymal or artifact voxels.

For both tests, voxels considered as non-parenchymal or affected by susceptibility artifacts based on individual ΔR2*HC, ΔR2*HO and T2*0 are shown overlapping a sagittal slice of the ICBM template. The latest column presents the overall excluded voxels in each test.

https://doi.org/10.1371/journal.pone.0163071.g003

thumbnail
Download:
Table 2. Exclusion of non-parenchymal or artifact voxels.

https://doi.org/10.1371/journal.pone.0163071.t002

T1 shortening

Estimates of arterial blood T1 during HO were found to be 1.558 ± 0.004 sec for Test A and 1.556 ± 0.006 sec for Test B, with no significant difference between them (paired t-test, P = 0.39). Computation of GM-averaged post-T1-correction Δ%CBFHO resulted in 0.4 ± 4.5% for Test A and -2.8 ± 3.3% for Test B. No significant difference (paired t-test, P = 0.25) was found between them and both were not significantly different from zero (P > 0.07).

Detection rate of solutions in ROI

The percentage of voxels where a solution was found for M and OEF were computed for each individual and test. The results, presented in the Table 3, show no significant difference between Test A and Test B (P > 0.21) and all detection rates were above 87% besides in HIP where the detection rare were 78%. These percentages are based on the number of voxels after exclusion of non-parenchymal and artifact voxels found in the Table 2.

thumbnail
Download:
Table 3. Detection rate of solutions in ROI (%)

https://doi.org/10.1371/journal.pone.0163071.t003

Test and retest group-averaged metrics across ROIs

A summary of the dual-echo pCASL and QUO2 ROI-averaged metrics is presented in Fig 4. For each combination of metric and ROI, we present the group average of Test A, Test B and of both tests. Resting CBF, O2 delivery and resting R2* displayed the highest group-averaged test-retest reproducibility. Resting R2* presented the lowest population variance in both tests. Resting CBF and O2 delivery, due to their direct relation, showed a similar pattern of values across ROIs. Similarly for M and ΔR2*HC, given that QUO2 M values are very sensitive to change in R2* and CBF during HC while nearly insensitive to change during HO (as later demonstrated in the “Precision of group GM-averaged values–sensitivity to random errors” section). A larger decrease in R2* (higher BOLD increase) was observed during HC than during HO in every ROIs. Although every metrics, but OEF, showed a certain tendency of heterogeneous values across ROIs, all differences were not found statistically significant, as presented in the following section.

thumbnail
Download:
Fig 4. Test and retest group-averaged metrics across ROIs.

ROI-averaged of Test A, Test B and all tests averaged are presented with standard deviation.

https://doi.org/10.1371/journal.pone.0163071.g004

Comparison of all tests averaged between ROIs

Data from all subjects and tests were averaged and compared across ROIs. Fig 5 shows where a significant difference (P < 0.05) was found between two ROIs, after correction for multiple comparisons. No statistical significant difference was found in fractional CBF change to HC across ROIs, except in the HYP, were the smallest change was found. R2* change to HO was found to be the smallest in AC among all ROIs except compared to HYP values, whereas the smallest R2* change to HC was found in HYP. Due to its low population variance within ROI, resting R2* presented values generally significantly different across ROIs. On the contrary, OEF were found to have consistent values across ROIs (P > 0.2).

thumbnail
Download:
Fig 5. Comparison of all tests averaged between ROIs.

For each metric, group average of all subjects and tests was compared across ROIs. Colors code for limits in P values after correction for multiple comparisons: dark blue indicates absence of significant difference (P > 0.05), while light blue (P < 0.05), green (P < 0.005) and orange (P < 0.0005) illustrate a significant difference between two ROIs (designed in the X and Y axis).

https://doi.org/10.1371/journal.pone.0163071.g005

ROI reproducibility analysis

In Table 4, we report the reproducibility analysis of the dual-echo pCASL and QUO2 measurements in each ROI, more precisely, the P-value of Kendall’s τ, Shapiro-Wilk and paired t-test, the CR and wsCV. Shapiro-Wilk and Kendall’s τ test detected, for a certain combination of metric-ROI, if the distribution deviated from normality (condition a in Table 4) and if there was a dependency between the differences in measurements and the mean of measurements (condition b in Table 4) respectively. In all cases, a log10 transformation of the data resulted in the satisfaction of both conditions (reproducibility of transformed data is presented in boldface) except for Δ%CBFHC in AC which distribution deviated from normality after the transformation (reproducibility of original data is presented in italic boldface). Paired t-tests detected no significant difference between Test A and Test B (P > 0.07), except for %M in AC (P < 0.001), OEF in HIP (P < 0.03) and CMRO2 in HIP (P < 0.05).

thumbnail
Download:
Table 4. Reproducibility of QUO2 measurements in different ROI.

https://doi.org/10.1371/journal.pone.0163071.t004

Fig 6 compares individual-subject reproducibility (wsCV) with the population variance (bsCV) of QUO2 measurements in each ROI. Among all ROIs, the measurement that showed lower inter- and intra-subject variability was resting R2* (wsCV < 2%; bsCV < 5.5%). CBF0 and O2 delivery come next, with a low individual-subject reproducibility (ROI-averaged wsCV = 5.5%) compared to the population variance (ROI-averaged bsCV = 14.6%), meaning the latter cannot be explained by poor reproducibility of the technique. It is also an indication of reproducible estimate of arterial O2 content. ΔR2HC was also found to have a considerably higher intra-subject consistency than across subjects (ROI-averaged wsCV = 9.3%; ROI-averaged bsCV = 22.3%), as opposition to ΔR2HO (ROI-averaged wsCV = 17.9%; ROI-averaged bsCV = 17.5%). The highest inter- and intra-subject variability was found in Δ%CBFHC (wsCVGM = 22.4%; wsCVallROI = [22.4%,46.9%]; bsCVGM = 21.6%; bsCVallROI = [21.6%,37.9%]). As for the QUO2 estimated measurements M, OEF and CMRO2, their wsCV in GM were found to be lower than 17% and they ranged from 13% to 27% across all ROIs. Their variability across subjects were higher by a factor of 1.3 in average across ROIs. Values averaged over GM were found to have the lowest inter- and intra-subject variability among regional averages (mean wsCV of 11.0 ± 7.4%; mean bsCV of 16.15 ± 6.55%) whereas variability in smaller regions, i.e. HIP and PC, was generally found higher. A significant correlation between the variability of a certain metric and the volume of observed region was detected only for resting R2* measurement (Pearson correlation coefficient R2 = -0.76, P = 0.48).

thumbnail
Download:
Fig 6. ROI analysis of within- and between-subject reproducibility.

Within each ROI, are shown the within- and between-subject coefficient of variation (respectively wsCV (blue) and bsCV (red)) for each metric.

https://doi.org/10.1371/journal.pone.0163071.g006

Test-retest correlation plots for each metric across the different ROIs can be found in Fig 7. For each individual, Test A against Test B measurements is shown, with colors coding for the different ROIs. For each metric, a linear regression model with zero intercept was applied on the data. The computed coefficient of determination (R2) is displayed in each graph. In each measurement, the regression line that fit the data is shown in solid gray. Values lying in the vicinity of the identity line (solid black line) indicate a good reproducibility of the measurements. Among all quantities, resting R2*, resting CBF, O2 delivery and ΔR2*HC present the best reproducibility, whereas Δ%CBFHC, ΔR2*HO, %M, OEF and CMRO2 show a larger scattering of the data around the identity line.

thumbnail
Download:
Fig 7. ROI analysis of individual reproducibility—Correlation plots.

For each measurement, individual Test A against Test B ROI-averaged is plotted. Values close to the identity line (solid black line) indicate a good reproducibility. For each metric, a linear regression with zero intercept was performed on the data. The resulted coefficient of detection (R2) and the fit to the data (gray line) are presented.

https://doi.org/10.1371/journal.pone.0163071.g007

An additional illustration of reproducibility is offered by Bland-Altman plots for each measurement, as shown in Fig 8. The individual mean of measurements against the difference in measurements is plotted, with colors coding for the different ROIs. For each ROI, a solid line represents mean of difference, whereas dashed lines depict the confidence intervals (i.e. mean difference ± CR). We can observe a trend toward higher variability in smaller regions (HIP, AC, and PC) compared to larger regions.

thumbnail
Download:
Fig 8. ROI analysis of individual reproducibility–Bland-Altman graphs.

For each measurement, the difference in sessions against the mean is displayed. Within each ROI, a solid line represents the mean difference between measurements. The dashed lines represent the ROI-averaged limits of agreement (mean ± 1.96 * dSD i.e. mean ± coefficient of repeatability (CR)) indicating that 95% of repeated measures will fall in between them.

https://doi.org/10.1371/journal.pone.0163071.g008

Parametric maps

In Fig 9, we show group-average Test A and B maps of the different measurements. One axial and one sagittal slice of the ICBM152 template are shown. The group-average maps generally demonstrate good qualitative agreement between test and retest, with the main exception being Δ%CBFHC. Fig 10 shows OEF and M maps in individual subjects, for Tests A and B, as well as group average and standard deviation maps. One axial slice of the ICBM152 template is shown. The subject count maps indicate, at each voxel, the number of subject included in the group average, given that voxels with no solution found for OEF and M were excluded from the average. Individual OEF and M maps generated using the optimized QUO2 present little random fluctuation in values and very few voxels with no solution. Group average maps for both parameters were qualitatively very similar in appearance, although differences between Tests A and B are noted in individual subjects.

thumbnail
Download:
Fig 9. Group maps.

For each metric, Test A and B group maps are shown in one axial and one sagittal slice. Maps were non-linearly registered to ICBM152 before being averaged. M, OEF and CMRO2 maps were averaged using an approach where non-solution voxels were excluded from the average.

https://doi.org/10.1371/journal.pone.0163071.g009

thumbnail
Download:
Fig 10. Individual OEF and M maps.

Test A and B individual OEF and M maps are shown in one axial slice of ICBM152 template. Followed maps are the group average, the standard deviation, and the subject count maps. The latter maps indicate at each voxel the number of subjects where a solution was found for OEF and M, meaning they were part of the group average.

https://doi.org/10.1371/journal.pone.0163071.g010

Accuracy of group GM-averaged values—sensitivity to systematic errors

Sensitivity of the model to systematic errors resulting from the assumed parameters is summarized in Fig 11. Based on our Test A group-averaged changes in ET during each respiratory manipulation, our evaluated ranges of percent change in metabolism were ±10% and ±7% in HC and HO respectively. Effects of the assumed α, β, [Hb], CMRO2HC, CMRO2HO and CBFHO on the HC and HO curves are presented (Fig 11A1 to 11A6 respectively), followed by resultant estimates of M, OEF and CMRO2 (Fig 11B1, 11B2 and 11B3 respectively). The β and CMRO2HC assumed values are the principal sources of variation in M, while variations in [Hb], CMRO2HO and CBFHO yield almost no change to the estimate. Fig 11A5 and 11A6 show that, by not affecting the HC curve, variations in metabolism and blood flow during HO shift the HO curve on the nearly horizontal section of the HC curve, resulting in almost no alteration in M solution. The assumption of change in CMRO2 during HO is the principal sources of variation in OEF, especially if a decrease in metabolism is considered (0.93 < CMRO2HO < 1.0). Furthermore, OEF is similarly sensible to variation in [Hb], CMRO2HC and CBFHO, while presenting almost no change as a variation in β. Both M and OEF are similarly influenced by a variation in α. The used value of 0.18 for α results in an M and OEF of 5.0% and 0.39 respectively, while if α is changed to the commonly adopted values of 0.38, the calculated M and OEF values become 5.6% and 0.44. Estimate of metabolism being the product of resting CBF by OEF and arterial O2 content, a certain percent increase/decrease in OEF leads to the same percent increase/decrease in CMRO2 with the exception where [Hb] is varied, given that it leads to a change in the arterial O2 content in the opposite direction as in OEF.

thumbnail
Download:
Fig 11. Accuracy of group GM-averaged values—sensitivity to systematic errors.

Observed effects, on M, OEF and CMRO2 estimates, of the assumed QUO2 parameters, are summarized. Estimates were based on group-averaged Test A de-PCASL measurements in GM and ETO2 (‘true’ values), while the assumed parameters α, β, [Hb], normalized change in CMRO2 due to HC (CMRO2HC) and HO (CMRO2HO), as well as normalized change in CBF during HO (CBFHO) were varied independently. The HC and HO curves resulting from the use of six different values of α, β, [Hb], CMRO2HC, CMRO2HO and CBFHO are presented (A1 to A6 respectively). Each red dot represents the HC and HO curves intersection (hence one M and OEF solution) when either one of the extremity in the observed range is in use. The remaining M and OEF solutions lie on a line connecting both red dots and passing by the subsequent intersections. M, OEF and CMRO2 estimates are presented in function of each varied parameter (B1, B2 and B3 respectively, ranges mentioned below each plot). Below the CBFHO range is shown the corresponding range of blood T1 during HO. The original values employed in this study were α = 0.18, β = 1.5, [Hb] = 15 g Hb/dl blood, CMRO2HC = 1 (isometabolic hypercapnia), CMRO2HO = 1 (isometabolic hyperoxia), while our post-T1-correction group-averaged normalized CBFHO in Test A was estimated to be 1 (i.e. no CBF change during HO). Resulted ‘true’ group-averaged Test A M, OEF and CMRO2 estimates in GM were 4.96%, 0.39 and 152 μmol/100 g/min respectively (shown by the blue stars).

https://doi.org/10.1371/journal.pone.0163071.g011

Precision of group GM-averaged values—sensitivity to random errors

Sensitivity of M and OEF to random errors in BOLD and CBF measurements is summarized in Fig 12. Given that neither resting CBF nor arterial O2 content are varied here, the relative impacts of errors in BOLD and CBF responses measurements on CMRO2 and OEF are equivalent. Therefore, estimates and variation of CMRO2 are not shown. Effect of errors in CBFHC, R2*HC and R2*HO on HC and HO curves are presented in Fig 12A1, 12A2 and 12A3 respectively. Errors in CBFHC or R2*HC affect both M and OEF estimates, whereas errors in R2*HO, by shifting the HO curve on the nearly horizontal section of the HC curve, result in almost no alteration in M solution. OEF and M estimates as a function of percent errors or variability of measurement are shown in Fig 12B1 and 12B2 respectively. Fig 12B1 exhibits a nearly linear increase in OEF errors as each of the input’s error increases. Fig 12B2 presents the same observation for M estimates, with the exception mentioned above where errors in R2*HO have almost no noticeable effect. The direction of the effect on OEF and M estimates depend of the source of error: an underestimated R2*HC will lead to an underestimated OEF or M, whereas if overestimated, OEF and M will also be overestimated. On the other hand, CBFHC have the opposite effect on OEF and M as well as R2*HO on OEF. Fig 12C1 and 12C2 show resulted variability of OEF and M estimates respectively. We observe that for the same percent error, CBFHC will induce a lower variability in OEF than R2*, and a lower variability in M than R2*HC. However, it is known that one of the limiting factors of such calibrated MRI methods is the low signal-to-noise ratio of ASL measurements, which is more susceptible to induce higher errors than in BOLD measurements. Fig 12C bring our attention to the fact that the test-retest variability found in our experimental CBFHC (wsCV = 22%) and R2*HO (wsCV = 16%) data, likely contributed to a minimum of 10% variability in our M and OEF estimates.

thumbnail
Download:
Fig 12. Precision of group GM-averaged values—sensitivity to random errors.

Observed effects, on M and OEF estimates, of error in BOLD and CBF measurements, are summarized. Estimates were based on group-averaged Test A de-PCASL measurements in GM and ETO2 (‘true’ values), while a simulated error was added to each observed input independently. Evaluated error ranged from -33% to 33%, which is equivalent to a maximum measurement CV of 28% (when underestimated) and 20% (when overestimated). The HC and HO curves resulting from six different errors in CBFHC, BOLDHC and BOLDHO are shown (A1, A2 and A3 respectively). Each red dot represents the HC and HO curves intersection (hence one M and OEF solution) when either one of the extremity in the observed range is in use. The remaining M and OEF solutions lie on a line connecting both red dots and passing by the subsequent intersections. OEF and M estimates (B1 and B2 respectively) in addition to CV between simulated and ‘true’ OEF/M values (C1 and C2) were computed for each source of errors (colored lines). Without addition of error in measurements, ‘true’ group-averaged Test A M and OEF estimates were 4.96% and 0.39 respectively (shown by the blue stars and the perforated grey lines).

https://doi.org/10.1371/journal.pone.0163071.g012

Discussion

We have characterized the inter- and intra-subject variability of the estimated M, OEF and absolute CMRO2 derived from the optimized QUO2 approach, which is a crucial step in the development and clinical application of this MRI technique for the quantitative measurement of oxygen delivery and consumption in the human brain. A visual inspection of the individual maps of M and OEF revealed a larger stability of values and fewer regions with no solution than the individual maps presented in the initial proof-of-concept of the QUO2 model [6].

We obtained physiologically plausible tests-averaged M, OEF and CMRO2 estimates in GM that were compared to those in literature summarized in Table 5. Variability among different studies in estimated values can arise from: the imaging technique employed, the type of breathing manipulations, the values of the assumed parameters, and the strategy employed to define the grey-matter region. Our M average of 4.69 ± 0.91% falls in the low end of reported values, although this may be mainly due to a more aggressive exclusion of large venous voxels. Our OEF and CMRO2 of 0.37 ± 0.06 and 144 ± 34 μmol/100 g/min respectively fall within the reported range of values.

thumbnail
Download:
Table 5. Literature values of M, OEF and resting-state CMRO2.

https://doi.org/10.1371/journal.pone.0163071.t005

Regional inter- and intra-subject reproducibility

Our individual-subject reproducibility in every parameters was found to be lower than our population variance, as also reported in previous study [38,47,60], except for our CBF response to HC and BOLD response to HO. Inter- and intra-subject variability of QUO2 measurements was found to be the lowest when averaged throughout all GM, with general trends toward higher CVs when averaged over smaller regions, a finding that is to be expected based on statistical considerations and consistent with previous studies [60,61]. The most intra-subject reproducible measurements was CBF0, along with O2 delivery and R2*0. Our regional CBF0 wsCV were comparable to that reported by Wang et al [61].

To our knowledge, reproducibility of OEF and absolute CMRO2 estimated from combined HC and HO calibrated BOLD technique has not been characterized in studies so far published. On the other hand, variability of M calibration parameter and relative change in CMRO2 were studied. In a small area of the visual cortex, Leontiev and Buxton [48] evaluated the intra-subject variability in M, CBF and BOLD responses to HC employing a different CV computation. Adjusting our CV computation to theirs, we obtain a similar wsCV for M of 13% in GM (compared to 12.3%) and a lower variability in BOLDHC and CBFHC with 7% vs. 26.1 and 20% vs. 40.6% respectively. They reported the presence of an important leak in the CO2 delivery during day 2 for two of their participants, thus increasing their intra-subject variability in BOLD and CBF responses to HC. In another study at 7 Tesla, using a simultaneous hypercapnia-hyperoxia (carbogen) gas-breathing challenge, Krieger et al [60] detected more intra-subject variability in their M value and BOLD changes to carbogen than what we found (wsCV≃34% vs. wsCV<17%) while they reported less variable CBF changes to carbogen than our CBF changes during HC (wsCV≃9% vs. wsCV = 22%). The difference among variability in M, CBF and BOLD changes to gas may arise from the different choice of MRI acquisition and gas challenge.

Studies employing dynamic 15O PET as a direct measure of OEF and CMRO2 have reported lower intra-subject variability in OEF and CMRO2. Coles et al [55] reported a whole brain test-retest coefficient of variation of 4.6% and 3.7% for OEF and CMRO2 respectively, while Bremmer et al [57] obtained a variability of 8.8%, 9.3% and 5.3% for CBF0, OEF and CMRO2 in GM respectively. In our study, resting CBF in GM varied less (wsCV = 3.9%) while OEF and CMRO2 had a higher degree of variability (13.6% and 15.2% respectively). Higher intra-subject variability in QUO2 estimates than PET estimates may arise partly from a larger noise accumulation in QUO2 techniques as an indirect measure of OEF and CMRO2.

Accuracy of group GM-averaged values—sensitivity to systematic errors

The QUO2 sensibility to assumed physiological parameters α, β, [Hb], CMRO2HC, CMRO2HO and CBFHO was examined. Recently, Wise et al. employed a Bayesian estimation framework in order to estimate α and β [4]. Merola et al [41] presented a simplified calibration model that substitute the standard α and β parameters with a single one and yielded improved estimates of OEF. Accuracy of estimates is likely to benefit from those calibration techniques where α and β are automatically estimated. Individual OEF estimates can also benefit from a simple blood test to determine the hemoglobin concentration. Isometabolism during hypercapnia and hyperoxia remains of a debate to date. Some studies suggest that CMRO2 does not change with HC [62–64] while others report an increase [65], or a decrease [66,67]. Similarly, heterogeneous results are found with regard to metabolism alteration during hyperoxia [68]. Without a clear consensus on the matter, we opted to preserve the assumption of no change in CMRO2 during both respiratory manipulations.

We have reported a certain variation in OEF (and CMRO2) as a function of reduction in blood flow caused by the hyperoxia manipulation. It is common to assume a fixed T1-corrected CBFHO for every subjects, for example Bulte et al. assumed a blood flow decrease of 4% during periods of hyperoxia at 50% O2 [2,69], or, as in the initial proof-of-concept of the QUO2 model, to compute the group-averaged measured CBFHO after T1-correction and use it for every subject. In the present study, we obtained physiologically plausible arterial blood T1 values (1.56 sec at 60% O2, consistent with 1.49 sec at 95% O2 [70]) and post-T1-correction CBFHO in GM. The latter was used as an individual basis, hence capturing any intra-subject variation between test and re-test blood flow decrease during HO. When correcting for T1 shortening during HO, rat blood was used as a surrogate for human blood which is justified since they both possess similar blood constitution, are likely to have very similar longitudinal relaxivity to molecular oxygen relationship and to experience similar physiological responses to hyperoxia.

Precision of group GM-averaged values—sensitivity to random errors

The error propagation analysis highlighted how the precision in QUO2 estimates was affected by random errors in the measurements of CBF and BOLD responses to the respiratory challenges. Especially, considering the same range of error, BOLD had a larger impact on QUO2 precision than CBF. However in practice, CBF response to HC is more challenging and carries a bigger uncertainty due to the low contrast-to-noise ratio. The range of error observed (-33% to +33%) is likely to approach the order of error in measured CBFHC while a lower uncertainty is expected in BOLD measurement, yielding smaller effects on QUO2 precision errors. It is therefore believed that the principal limitation in the QUO2 M, OEF and CMRO2 intra-subject reproducibility remains the error in CBF responses. Although the exact contribution of the measurements errors compared to the intrinsic day-to-day physiological variability could be further studied with a within-session test-retest reproducibility design, the present study offers clinically relevant assessment of variability in QUO2 measurements for follow-up studies.

QUO2 qualitative individual and group maps reproducibility

As expected, the reproducibility of QUO2 parameter estimates was generally improved with more extensive averaging of voxels and subjects. While group average maps demonstrated qualitatively good reproducibility, it was found that physiological and measurement noise still limits reproducibility of voxel values at the individual subject level.

Potential QUO2 accuracy and precision improvement

There are other potential avenues to improve the accuracy and precision of functional ASL. One strategy may include the use of a higher magnetic field, which would increase the SNR by increasing spin polarization and the T1 relaxation time of arterial blood. The gain in SNR might be partly lost due to an increase in physiological noise, however, this effect could be diminished by applying a denoising technique like RETROICOR [71,72]. Improved control for magnetization transfer effects is another mechanism for improving accuracy [73]. Additional approaches may also enhance accuracy and reproducibility of QUO2 measurements by correcting for region-specific variations in tag arrival times [74–76], for drops in label efficiency [77,78], for partial volume [79] and hemodynamic response delays [80]. Benefits may also be offered by improved imaging readout methods such as the use of a 3D imaging readout [81], multiband excitation [82] and background suppression [83–85].

Additional factors that can influence accuracy and precision of measurements in such calibrated approaches are the inhaled concentrations of oxygen and carbon dioxide. A higher O2 and CO2 concentration would yield BOLD changes closer to the M value to be extrapolated, reducing measurement errors. Furthermore, higher CO2 concentration would have the advantage of increasing the contrast-to-noise ratio due to higher CBF changes. However, administration of high levels of O2 and CO2 introduce potential problems: higher O2 concentration can complicate the quantification of CBF due to blood T1 reduction and susceptibility artifacts in areas close to sinuses and airways, while higher levels of CO2 can lead to anxiety and potentially alter brain physiology in ways other than the intended vasodilatory effect [86,87]. Future studies should assess the effect of oxygen and carbon dioxide concentration in reproducibility of quantitative calibrated methods such as QUO2. An investigation on the effect of O2 concentration in QUO2 estimates is presently in process, comparing 60% vs. 100% O2 during hyperoxia periods (manuscript in preparation).

The imaging parameters of the de-pCASL used in the present study were adjusted to optimize the detection rate of CBF responses in GM while acquiring whole-brain image with a minimal gap between slices. In order to allow the labeled blood bolus to arrive in the tissue of the imaged region, a post labeling delay (PLD) is employed. In our 2D acquisition, the first and last slices are acquired after a delay of 900 msec and 1986 msec respectively, resulting in an brain-averaged PLD of 1443 msec which is shorter than that recommended in the ASL white paper [16]. The latter recommends a 3D readout PLD of 1800 msec for healthy subjects of age below 70 years old. MacIntosh et al [74] evaluated the regional arterial time transit (ATT) in a cohort of healthy participants with range of age equivalent to that of our group. Among the temporal, parietal, frontal and occipital lobes, only the latter was reported to have a mean ATT (935 ± 0.108ms) slightly higher than our nominal PLD (900 msec), but lower than the acquisition time of our second slice (953 msec). Although our choice of PLD might not be optimal in lower occipital region of certain healthy participants, we believe that in the large majority of cases, the acquired ASL signal was accurately reflecting CBF and that an increase in our PLD would have resulted in a loss in SNR, especially in hypercapnic situation where the ATT is known to diminish.

Small cohort sizes like 8 have been common in recent years, particularly for complex fMRI protocols with greater physiological specificity than the classic BOLD contrast. Although our conclusion are limited by the relatively small sample size, we felt that this regional analysis of inter- and intra-subject variability of the QUO2 estimates in such a cohort remains of interest.

In conclusion, the variability of the optimized QUO2 estimates across subjects and the intra-subject reproducibility of estimates in different brain regions were characterized while the impact of errors on the accuracy and precision of such estimates was determined. These results will help guide power analyses for research applications as well as future developments aimed at further improving the reproducibility of the QUO2 method.

Acknowledgments

The authors would like to thank Scott Nugent, Marius Tuznik, Bahare Sabouri, Carollyn Hurst and André Cyr for their excellent technical assistance and insightful discussions. Jiongjiong Wang at UCLA is acknowledged for providing the pseudo-continuous arterial spin-labeling sequence.

Author Contributions

  1. Conceptualization: IL FBT RDH.
  2. Data curation: IL FBT RDH.
  3. Formal analysis: IL.
  4. Funding acquisition: RDH.
  5. Investigation: IL FBT RDH.
  6. Methodology: IL FBT RDH.
  7. Project administration: RDH.
  8. Resources: RDH.
  9. Software: IL FBT RDH.
  10. Supervision: RDH.
  11. Validation: IL FBT RDH.
  12. Visualization: IL.
  13. Writing – original draft: IL.
  14. Writing – review & editing: IL FBT RDH.

References

  1. 1. Frackowiak RS, Lenzi GL, Jones T, Heather JD. Quantitative measurement of regional cerebral blood flow and oxygen metabolism in man using 15O and positron emission tomography: theory, procedure, and normal values. J Comput Assist Tomogr. 1980;4:727–736. pmid:6971299
  2. 2. Bulte DP, Kelly M, Germuska M, Xie J, Chappell MA, Okell TW, et al. Quantitative measurement of cerebral physiology using respiratory-calibrated MRI. NeuroImage. 2012;60:582–591. pmid:22209811
  3. 3. Gauthier CJ, Desjardins-Crépeau L, Madjar C, Bherer L, Hoge RD. Absolute quantification of resting oxygen metabolism and metabolic reactivity during functional activation using QUO2 MRI. NeuroImage. 2012;63:1353–1363. pmid:22986357
  4. 4. Wise RG, Harris AD, Stone AJ, Murphy K. Measurement of OEF and absolute CMRO2: MRI-based methods using interleaved and combined hypercapnia and hyperoxia. NeuroImage. 2013;83:135–147. pmid:23769703
  5. 5. Gauthier CJ, Hoge RD. A generalized procedure for calibrated MRI incorporating hyperoxia and hypercapnia. Hum. Brain Mapp. 2012;34:1053–1069. pmid:23015481
  6. 6. Gauthier CJ, Hoge RD. Magnetic resonance imaging of resting OEF and CMRO2 using a generalized calibration model for hypercapnia and hyperoxia. NeuroImage. 2011;60:1212–1225. pmid:22227047
  7. 7. Tancredi FB, Lajoie I, Hoge RD. A simple breathing circuit allowing precise control of inspiratory gases for experimental respiratory manipulations. BMC Res Notes. 2014;7:235. pmid:24725848
  8. 8. Tancredi FB, Lajoie I, Hoge RD. Test-retest reliability of cerebral blood flow and blood oxygenation level-dependent responses to hypercapnia and hyperoxia using dual-echo pseudo-continuous arterial spin labeling and step changes in the fractional composition of inspired gases. J. Magn. Reson. Imaging. 2015;42:1144–1157. pmid:25752936
  9. 9. Parkes LM, Rashid W, Chard DT, Tofts PS. Normal cerebral perfusion measurements using arterial spin labeling: Reproducibility, stability, and age and gender effects. Magn. Reson. Med. 2004;51:736–743. pmid:15065246
  10. 10. Dai W, Garcia D, de Bazelaire C, Alsop DC. Continuous flow-driven inversion for arterial spin labeling using pulsed radio frequency and gradient fields. Magn. Reson. Med. 2008;60:1488–1497. pmid:19025913
  11. 11. Severinghaus JW. Water vapor calibration errors in some capnometers: respiratory conventions misunderstood by manufacturers? Anesthesiology. 1989;70:996–998. pmid:2499226
  12. 12. Tancredi FB, Hoge RD. Comparison of cerebral vascular reactivity measures obtained using breath-holding and CO2 inhalation. J Cereb Blood Flow Metab. 2013:1–9.
  13. 13. Chiarelli PA, Bulte DP, Wise R, Gallichan D, Jezzard P. A calibration method for quantitative BOLD fMRI based on hyperoxia. NeuroImage. 2007;37:808–820. pmid:17632016
  14. 14. Poulin MJ, Liang PJ, Robbins PA. Dynamics of the cerebral blood flow response to step changes in end-tidal PCO~and PO, in humans. J Appl Physiol. 1996;81:1084–1095. pmid:8889738
  15. 15. Wang J, Alsop DC, Song HK, Maldjian JA, Tang K, Salvucci AE, et al. Arterial transit time imaging with flow encoding arterial spin tagging (FEAST). Magn. Reson. Med. 2003;50:599–607. pmid:12939768
  16. 16. Alsop DC, Detre JA, Golay X, Günther M, Hendrikse J, Hernandez-Garcia L, et al. Recommended implementation of arterial spin-labeled perfusion MRI for clinical applications: A consensus of the ISMRM perfusion study group and the European consortium for ASL in dementia. Magn. Reson. Med. 2015;73:102–116. pmid:24715426
  17. 17. Pilkinton DT, Hiraki T, Detre JA, Greenberg JH, Reddy R. Absolute cerebral blood flow quantification with pulsed arterial spin labeling during hyperoxia corrected with the simultaneous measurement of the longitudinal relaxation time of arterial blood. Magn. Reson. Med. 2011;67:1556–1565. pmid:22135087
  18. 18. Chalela JA, Alsop DC, Gonzalez-Atavales JB, Maldjian JA, Kasner SE, Detre JA. Magnetic Resonance Perfusion Imaging in Acute Ischemic Stroke Using Continuous Arterial Spin Labeling. Stroke. 2000;31:680–687. pmid:10700504
  19. 19. Zaharchuk G, Martin AJ, Dillon WP. Noninvasive Imaging of Quantitative Cerebral Blood Flow Changes during 100% Oxygen Inhalation Using Arterial Spin-Labeling MR Imaging. American Journal of Neuroradiology. 2008;29:663–667. pmid:18397966
  20. 20. Chen JJ, Pike GB. MRI measurement of the BOLD-specific flow–volume relationship during hypercapnia and hypocapnia in humans. NeuroImage. 2010;53:383–391. pmid:20624474
  21. 21. Boxerman JL, Hamberg LM, Rosen BR, Weisskoff RM. Contrast due to Intravascular Magnetic Susceptibility Perturbations. Magn. Reson. Med. 1995;34:555–566. pmid:8524024
  22. 22. Jenkinson M, Smith S. A global optimisation method for robust affine registration of brain images. Med Image Anal. 2001;5:143–156. pmid:11516708
  23. 23. Klein A, Tourville J. 101 labeled brain images and a consistent human cortical labeling protocol. Front. Neurosci. 2012;6:171. pmid:23227001
  24. 24. Alsop DC, Detre JA, Grossman M. Assessment of cerebral blood flow in Alzheimer's disease by spin-labeled magnetic resonance imaging. Ann Neurol. 2000;47:93–100. pmid:10632106
  25. 25. Johnson NA, Jahng G-H, Weiner MW, Miller BL, Chui HC, Jagust WJ, et al. Pattern of cerebral hypoperfusion in Alzheimer disease and mild cognitive impairment measured with arterial spin-labeling MR imaging: initial experience. Radiology. 2005;234:851–859. pmid:15734937
  26. 26. Du AT, Jahng GH, Hayasaka S, Kramer JH, Rosen HJ, Gorno-Tempini ML, et al. Hypoperfusion in frontotemporal dementia and Alzheimer disease by arterial spin labeling MRI. Neurology. 2006;67:1215–1220. pmid:17030755
  27. 27. Asllani I, Habeck C, Scarmeas N, Borogovac A, Brown TR, Stern Y. Multivariate and univariate analysis of continuous arterial spin labeling perfusion MRI in Alzheimer's disease. J Cereb Blood Flow Metab. 2007;28:725–736. pmid:17960142
  28. 28. Tohgi H, Yonezawa H, Takahashi S, Sato N, Kato E, Kudo M, et al. Cerebral blood flow and oxygen metabolism in senile dementia of Alzheimer's type and vascular dementia with deep white matter changes. Neuroradiology. 1998;40:131–137. pmid:9561514
  29. 29. Yao J, Rettberg JR, Klosinski LP, Cadenas E, Brinton RD. Shift in brain metabolism in late onset Alzheimer’s disease: Implications for biomarkers and therapeutic interventions. Mol Aspects Med. 2011;32:247–257. pmid:22024249
  30. 30. Collins DL, Holmes CJ, Peters TM, Evans AC. Automatic 3-D Model-Based Neuroanatomical Segmentation. Hum. Brain Mapp. 1995;3:190–208.
  31. 31. Sherif T, Rioux P, Rousseau M-E, Kassis N, Beck N, Adalat R, et al. CBRAIN: a web-based, distributed computing platform for collaborative neuroimaging research. Front. Integr. Neurosci. 2014;8:54. pmid:24904400
  32. 32. Padhani AR, Hayes C, Landau S, Leach MO. Reproducibility of quantitative dynamic MRI of normal human tissues. NMR Biomed. 2002;15:143–153. pmid:11870910
  33. 33. Floyd TF, Ratcliffe SJ, Wang J, Resch B, Detre JA. Precision of the CASL-perfusion MRI technique for the measurement of cerebral blood flow in whole brain and vascular territories. J. Magn. Reson. Imaging. 2003;18:649–655. pmid:14635149
  34. 34. Jain V, Duda J, Avants B, Giannetta M, Xie SX, Roberts T, et al. Longitudinal reproducibility and accuracy of pseudo-continuous arterial spin-labeled perfusion MR imaging in typically developing children. Radiology. 2012;263:527–536. pmid:22517961
  35. 35. Chen Y, Wang DJJ, Detre JA. Test-retest reliability of arterial spin labeling with common labeling strategies. J. Magn. Reson. Imaging. 2011;33:940–949. pmid:21448961
  36. 36. Bland MJ, Altman DG. Measurement error proportional to the mean. BMJ. 1996;313:106. pmid:8688716
  37. 37. Bland JM, Altman DG. Statistical methods for assessing agreement between two methods of clinical measurement. Lancet. 1986;i:307–310.
  38. 38. Tjandra T, Brooks JCW, Figueiredo P, Wise R, Matthews PM, Tracey I. Quantitative assessment of the reproducibility of functional activation measured with BOLD and MR perfusion imaging: Implications for clinical trial design. NeuroImage. 2005;27:393–401. pmid:15921936
  39. 39. Mark CI, Fisher JA, Pike GB. Improved fMRI calibration: Precisely controlled hyperoxic versus hypercapnic stimuli. NeuroImage. 2011;54:1102–1111. pmid:20828623
  40. 40. Gustard S, Williams EJ, Hall LD, Pickard JD, Carpenter TA. Influence of baseline hematocrit on between-subject BOLD signal change using gradient echo and asymmetric spin echo EPI. J. Magn. Reson. Imaging. 2003;21:599–607.
  41. 41. Merola A, Murphy K, Stone AJ, Germuska MA, Griffeth VEM, Blockley NP, et al. Measurement of oxygen extraction fraction (OEF): An optimized BOLD signal model for use with hypercapnic and hyperoxic calibration. NeuroImage. The Authors; 2016;129:159–174. pmid:26801605
  42. 42. Bulte DP, Drescher K, Jezzard P. Comparison of hypercapnia-based calibration techniques for measurement of cerebral oxygen metabolism with MRI. Magn. Reson. Med. 2009;61:391–398. pmid:19165902
  43. 43. Ances BM, Leontiev O, Perthen JE, Liang C, Lansing AE, Buxton RB. Regional differences in the coupling of cerebral blood flow and oxygen metabolism changes in response to activation: Implications for BOLD-fMRI. NeuroImage. 2008;39:1510–1521. pmid:18164629
  44. 44. Ances BM, Liang CL, Leontiev O, Perthen JE, Fleisher AS, Lansing AE, et al. Effects of aging on cerebral blood flow, oxygen metabolism, and blood oxygenation level dependent responses to visual stimulation. Hum. Brain Mapp. 2009;30:1120–1132. pmid:18465743
  45. 45. Chen Y, Parrish TB. Caffeine's effects on cerebrovascular reactivity and coupling between cerebral blood flow and oxygen metabolism. NeuroImage. 2009;44:647–652. pmid:19000770
  46. 46. Gauthier CJ, Madjar C, Tancredi FB, Stefanovic B, Hoge RD. Elimination of visually evoked BOLD responses during carbogen inhalation: Implications for calibrated MRI. NeuroImage. 2011;54:1001–1011. pmid:20887792
  47. 47. Lin A-L, Fox PT, Yang Y, Lu H, Tan L-H, Gao J-H. Evaluation of MRI models in the measurement of CMRO 2and its relationship with CBF. Magn. Reson. Med. 2008;60:380–389. pmid:18666102
  48. 48. Leontiev O, Buxton RB. Reproducibility of BOLD, perfusion, and CMRO2 measurements with calibrated-BOLD fMRI. NeuroImage. 2007;35:175–184. pmid:17208013
  49. 49. Perthen JE, Lansing AE, Liau J, Liu TT, Buxton RB. Caffeine-induced uncoupling of cerebral blood flow and oxygen metabolism: a calibrated BOLD fMRI study. NeuroImage. 2008;40:237–247. pmid:18191583
  50. 50. Xu F, Ge Y, Lu H. Noninvasive quantification of whole-brain cerebral metabolic rate of oxygen (CMRO2) by MRI. Magn. Reson. Med. 2009;62:141–148. pmid:19353674
  51. 51. Bolar DS, Rosen BR, Sorensen AG, Adalsteinsson E. QUantitative Imaging of eXtraction of oxygen and TIssue consumption (QUIXOTIC) using venular-targeted velocity-selective spin labeling. Magn. Reson. Med. 2011;66:1550–1562. pmid:21674615
  52. 52. Lu H, Ge Y. Quantitative evaluation of oxygenation in venous vessels using T2-Relaxation-Under-Spin-Tagging MRI. Magn. Reson. Med. 2008;60:357–363. pmid:18666116
  53. 53. Jain V, Langham MC, Wehrli FW. MRI estimation of global brain oxygen consumption rate. J Cereb Blood Flow Metab. Nature Publishing Group; 2010;30:1598–1607. pmid:20407465
  54. 54. van Zijl PCM, Eleff SM, Ulatowski JA, Oja JME, Ulug AM, Traystman RJ, et al. Quantitative assessment of blood flow, blood volume and blood oxygenation effects in functional magnetic resonance imaging. Nat. Med. 1998;4:159–167. pmid:9461188
  55. 55. Coles JP, Fryer TD, Bradley PG, Nortje J, Smielewski P, Rice K, et al. Intersubject variability and reproducibility of 15O PET studies. J Cereb Blood Flow Metab. 2006;26:48–57. pmid:15988475
  56. 56. Leenders KL, Perani D, Lammertsma AA, Heather JD. Cerebral blood flow, blood volume and oxygen utilization. Brain. 1990;113:27–47. pmid:2302536
  57. 57. Bremmer JP, Berckel BNM, Persoon S, Kappelle LJ, Lammertsma AA, Kloet R, et al. Day-to-Day Test–Retest Variability of CBF, CMRO2, and OEF Measurements Using Dynamic 15O PET Studies. Mol Imaging Biol. 2010;13:759–768.
  58. 58. Ibaraki M, Shinohara Y, Nakamura K, Miura S, Kinoshita F, Kinoshita T. Interindividual variations of cerebral blood flow, oxygen delivery, and metabolism in relation to hemoglobin concentration measured by positron emission tomography in humans. J Cereb Blood Flow Metab. 2010;30:1296–1305. pmid:20160738
  59. 59. Ito H, Kanno I, Kato C, Sasaki T, Ishii K, Ouchi Y, et al. Database of normal human cerebral blood flow, cerebral blood volume, cerebral oxygen extraction fraction and cerebral metabolic rate of oxygen measured by positron emission tomography with 15 O-labelled carbon dioxide or water, carbon monoxide and oxygen: a multicentre study in Japan. Eur J Nucl Med Mol Imaging. 2004;31:635–643. pmid:14730405
  60. 60. Krieger SN, Gauthier CJ, Ivanov D, Huber L, Roggenhofer E, Sehm B, et al. Regional reproducibility of calibrated BOLD functional MRI: Implications for the study of cognition and plasticity. NeuroImage. 2014;101:8–20. pmid:25008001
  61. 61. Wang Y, Saykin AJ, Pfeuffer J, Lin C, Mosier KM, Shen L, et al. Regional reproducibility of pulsed arterial spin labeling perfusion imaging at 3T. NeuroImage. 2011;54:1188–1195. pmid:20800097
  62. 62. Chen JJ, Pike GB. BOLD-specific cerebral blood volume and blood flow changes during neuronal activation in humans. NMR Biomed. 2009;22:1054–1062. pmid:19598180
  63. 63. Hino JK, Short BL, Rais-Bahrami K, Seale WR. Cerebral blood flow and metabolism during and after prolonged hypercapnia in newborn lambs. Crit Care Med. 2000;28:3505–3510. pmid:11057808
  64. 64. McPherson RW, Derrer SA, Traystman RJ. Changes in cerebral CO2 responsivity over time during isoflurane anesthesia in the dog. J Neurosurg Anesthesiol. 1991;3:12–19. pmid:15815378
  65. 65. Horvath I, Sandor NT, Ruttner Z, McLaughlin AC. Role of nitric oxide in regulating cerebrocortical oxygen consumption and blood flow during hypercapnia. J Cereb Blood Flow Metab. 1994;14:503–509. pmid:8163593
  66. 66. Zappe AC, Uludağ K, Logothetis NK. Direct measurement of oxygen extraction with fMRI using 6% CO2 inhalation. J. Magn. Reson. Imaging. 2008;26:961–967.
  67. 67. Xu F, Uh J, Brier MR, Hart J, Yezhuvath US, Gu H, et al. The influence of carbon dioxide on brain activity and metabolism in conscious humans. J Cereb Blood Flow Metab. 2010;31:58–67. pmid:20842164
  68. 68. Xu F, Liu P, Pascual JM, Xiao G, Lu H. Effect of hypoxia and hyperoxia on cerebral blood flow, blood oxygenation, and oxidative metabolism. J Cereb Blood Flow Metab. 2012;32:1909–1918. pmid:22739621
  69. 69. Bulte DP, Chiarelli PA, Wise RG, Jezzard P. Cerebral perfusion response to hyperoxia. J Cereb Blood Flow Metab. 2006;27:69–75. pmid:16670698
  70. 70. Siero JCW, Strother MK, Faraco CC, Hoogduin H, Hendrikse J, Donahue MJ. In vivoquantification of hyperoxic arterial blood water T1. NMR Biomed. 2015;28:1518–1525. pmid:26419505
  71. 71. Glover GH, Li TQ, Ress D. Image-based method for retrospective correction of physiological motion effects in fMRI: RETROICOR. Man. Reason. Med. 2000;44:162–167.
  72. 72. Wu Wen-Chau, Edlow BL Elliot MA, Wang Jiongjiong, Detre JA. Physiological Modulations in Arterial Spin Labeling Perfusion Magnetic Resonance Imaging. IEEE Trans. Med. Imaging. 2009;28:703–709. pmid:19150788
  73. 73. Zhang W, Silva AC, Williams DS, Koretsky AP. Measurement of Perfusion Using Arterial Spin Labeling Without Saturation. Magn. Reson. Med. 1995;33:370–376. pmid:7760704
  74. 74. MacIntosh BJ, Filippini N, Chappell MA, Woolrich MW, Mackay CE, Jezzard P. Assessment of arterial arrival times derived from multiple inversion time pulsed arterial spin labeling MRI. Magn. Reson. Med. 2010;63:641–647. pmid:20146233
  75. 75. Gallichan D, Jezzard P. Variation in the shape of pulsed arterial spin labeling kinetic curves across the healthy human brain and its implications for CBF quantification. Magn. Reson. Med. 2009;61:686–695. pmid:19132757
  76. 76. Donahue MJ, Faraco CC, Strother MK, Chappell MA, Rane S, Dethrage LM, et al. Bolus arrival time and cerebral blood flow responses to hypercarbia. J Cereb Blood Flow Metab. 2014;34:1243–1252. pmid:24780904
  77. 77. Aslan S, Xu F, Wang PL, Uh J, Yezhuvath US, van Osch M, et al. Estimation of labeling efficiency in pseudocontinuous arterial spin labeling. Magn. Reson. Med. 2010;63:765–771. pmid:20187183
  78. 78. Jahanian H, Noll DC, Hernandez-Garcia L. B0 field inhomogeneity considerations in pseudo-continuous arterial spin labeling (pCASL): effects on tagging efficiency and correction strategy. NMR Biomed. 2011;24:1202–1209. pmid:21387447
  79. 79. Asllani I, Borogovac A, Brown TR. Regression algorithm correcting for partial volume effects in arterial spin labeling MRI. Magn. Reson. Med. 2008;60:1362–1371. pmid:18828149
  80. 80. Blockley NP, Driver ID, Francis ST, Fisher JA, Gowland PA. An improved method for acquiring cerebrovascular reactivity maps. Magn. Reson. Med. 2010;65:1278–1286. pmid:21500256
  81. 81. Günther M, Oshio K, Feinberg DA. Single-shot 3D imaging techniques improve arterial spin labeling perfusion measurements. Magn. Reson. Med. 2005;54:491–498. pmid:16032686
  82. 82. Feinberg DA, Setsompop K. Ultra-fast MRI of the human brain with simultaneous multi-slice imaging. Journal of Magnetic Resonance. 2013;229:90–100. pmid:23473893
  83. 83. Ye FQ, Frank JA, Weinberger DR, McLaughlin AC. Noise Reduction in 3D Perfusion Imaging by Attenuating the Static Signal in Arterial Spin Tagging (ASSIST). Magn. Reson. Med. 2000;44:92–100. pmid:10893526
  84. 84. Garcia DM, Duhamel G, Alsop DC. Efficiency of inversion pulses for background suppressed arterial spin labeling. Magn. Reson. Med. 2005;54:366–372. pmid:16032674
  85. 85. Ghariq E, Chappell MA, Schmid S, Teeuwisse WM, van Osch MJP. Effects of background suppression on the sensitivity of dual-echo arterial spin labeling MRI for BOLD and CBF signal changes. NeuroImage. 2014;103:316–322. pmid:25280450
  86. 86. Brannan S, Liotti M, Egan G, Shade R, Madden L, Robillard R, et al. Neuroimaging of cerebral activations and deactivations associated with hypercapnia and hunger for air. Physiology. 2001;98:2029–2034.
  87. 87. Liotti M, Brannan S, Egan G, Shade R, Madden L, Abplanalp B, et al. Brain responses associated with consciousness of breathlessness (air hunger). Physiology. 2001;98:2035–2040.