A calibration method for quantitative BOLD fMRI based on hyperoxia

www.elsevier.com/locate/ynimg NeuroImage 37 (2007) 808 – 820

A calibration method for quantitative BOLD fMRI based on hyperoxia Peter A. Chiarelli,1 Daniel P. Bulte,⁎,1 Richard Wise, Daniel Gallichan, and Peter Jezzard FMRIB Centre, Department of Clinical Neurology, University of Oxford, John Radcliffe Hospital, Oxford, OX3 9DU, UK Received 27 October 2006; revised 1 May 2007; accepted 10 May 2007 Available online 31 May 2007 The estimation of changes in CMRO2 using functional MRI involves an essential calibration step using a vasoactive agent to induce an isometabolic change in CBF. This calibration procedure is performed most commonly using hypercapnia as the isometabolic stimulus. However, hypercapnia possesses a number of detrimental side effects. Here, a new method is presented using hyperoxia to perform the same calibration step. This procedure requires independent measurement of PaO2, the BOLD signal, and CBF. We demonstrate that this method yields results that are comparable to those derived using other methods. Further, the hyperoxia technique is able to provide an estimate of the calibration constant that has lower overall intersubject and intersession variability compared to the hypercapnia approach. © 2007 Elsevier Inc. All rights reserved.

Introduction Changes in the cerebral metabolic rate of oxygen (CMRO2) induced by neural activity can be estimated non-invasively using blood oxygenation level dependent (BOLD) fMRI and arterial spin-labelling measurements. The deoxyhaemoglobin dilution model outlined by Davis et al. (1998) utilizes CO2 breathing to calibrate the resting-state BOLD signal, by inducing an isometabolic global increase in cerebral blood flow (CBF). Hypercapnia was then employed to estimate relative changes in CMRO2 (Kim et al., 1999) and was extended through the use of graded hypercapnia to accurately estimate the resting-CMRO2 BOLD−CBF relationship, along with graded functional stimuli that established a linear coupling relationship between changes in CMRO2 and changes in CBF (Hoge et al., 1999a,b,c).

⁎ Corresponding author. Fax: +44 1865 222 717. E-mail address: [email protected] (D.P. Bulte). 1 These authors contributed equally to the work. Available online on ScienceDirect (www.sciencedirect.com). 1053-8119/$ - see front matter © 2007 Elsevier Inc. All rights reserved. doi:10.1016/j.neuroimage.2007.05.033

The hypercapnia-calibrated BOLD technique has since been widely adopted for investigating activation- and deactivationrelated flow-metabolism coupling in the human cortex, for examining the interaction between baseline blood flow and activation amplitude, and for measuring the effects of pathology on cerebral haemodynamics (Chiarelli et al., 2007a; Kastrup et al., 2002; Stefanovic et al., 2004, 2005, 2006; Uludag et al., 2004). Although the use of CO2 has multiple drawbacks – including a potential influence on CMRO2 (Kliefoth et al., 1979), a definite correlation with the onset of breathlessness (Rostrup et al., 2000), and potential intolerability in distressed or infirm subjects – CO2 remains the dominant option for fMRI calibration studies, and the assumption is commonly made that CO2 has a negligible impact on the baseline neural activity, despite the fact that it has a regionally different effect on CBF (Rostrup et al., 2000). The hypercapnia method also yields large potential variability in the calculated calibration parameter (M—the maximum theoretical BOLD signal change) (Hoge et al., 1999a), due to the use of perfusion imaging for CBF measurement, which is a low signal-to-noise ratio (SNR) technique (Chiarelli et al., 2007a,b). Even calibration techniques which do not employ CBF modulation suffer from the low SNR inherent with MRI perfusion measurements (Fujita et al., 2006). Acetazolamide injection has also been used to elevate global CBF, although this method is invasive, and acts through a similar chemical pathway to CO2 (Bickler et al., 1988). Breath holding has been employed as a simple technique for causing short epochs of mild hypercapnia (Kastrup et al., 1998, 1999b; Vazquez et al., 2006) and demonstrated the potential to reduce variability in fMRI studies (Handwerker et al., 2007; Kastrup et al., 1999a; Thomason et al., 2007). Breath holding has many advantages, the greatest being that no specialized equipment is needed; however, it suffers from a number of complications limiting its practical usage. These issues include the requirement of subject compliance, the stress it can induce in subjects, the reproducibility, and most of all the fact that metabolism and arterial oxygen levels are dynamically changing during the breath hold. All of these factors combine to limit its application. Although a very useful research tool it is yet to be in widespread use.

P.A. Chiarelli et al. / NeuroImage 37 (2007) 808–820

809

Our group has recently demonstrated that variability in M has a large impact on the measured CMRO2−CBF coupling constant from a neural activation task, as well as an impact on the linearity of the relationship between these two variables (Chiarelli et al., 2007b). It is therefore valuable to consider alternative techniques for the calibration of BOLD−CBF coupling studies. In this work, we introduce hyperoxia (breathing of O2enriched air) as a new technique for calibration of the baseline BOLD−CBF relationship. Whereas hypercapnia uses the CBF change as a means to increase oxygenation in the venous vasculature, O2 breathing directly alters the oxygen saturation of arterial plasma and arterial Hb, inducing measurable changes in the concentration of oxygenated capillary and venous Hb (Berkowitz, 1997; Bulte et al., 2006, 2007; Kwong et al., 1995; Rostrup et al., 1995). Hyperoxia calibration requires measurement of the BOLD signal and end-tidal O2 values, both of which are relatively low-variability measurements. Acquisition of CBF data can also be performed to correct for a small reduction in CBF which occurs during hyperoxia. Hyperoxia is very well tolerated for long periods of time, allowing the acquisition of long temporal epochs.

contributions to the CBV/CBV0 and [dHb]/[dHb]0 terms of Eqs. (3a) (3b).

Background theory and methods

D½dHbv ½dHbv ¼ 1 ½dHbv0 ½dHbv0

ð6Þ

½dHbv CBF0 ¼ ½dHbv0 CBF

ð7Þ

The hyperoxia-calibrated model Analogous to the derivation of the hypercapnia-calibrated model (Hoge et al., 1999a), hyperoxia calibration makes use of the expression for BOLD signal change, DBOLD ¼ TEDR2*jdHb ; BOLD0

ð1Þ

as well as the expression for R2⁎|dHb derived by Boxerman et al. (1995). R* 2 jdHb ¼

AðCBV½dHbbv Þ

10from O 1 1 0from O from CBF from CBF b z}|{ 2 z}|{ z}|{ 2 z}|{ B BCBV þ DCBV CB½dHbv þ D½dHbvC C DBOLD CB C C B ¼ MB A@ A A @1 @ BOLD0 CBV0 ½dHbv0 0

ð4Þ 0 



B DBOLD CBV B ¼ M B1  @ BOLD0 CBV0 |fflfflfflfflfflffl{zfflfflfflfflfflffl}

0

1b

B ½dHb D½dHbv B v þ B @½dHbv0 ½dHbv |{z}0

C C C A

CBF correction term

!

CBF correction term

ð5Þ Since an isolated change in Hb oxygenation will not impact CBV, the CBV term in Eq. (4) arising from an oxygenation change is equal to CBV0. The CBV correction term in Eq. (5) can be replaced with CBF by assuming the Grubb relationship (Eq. (2)) (Grubb et al., 1974). Furthermore, the Δ[dHb]v / [dHb]v0 contribution can be replaced with a term related to CBF.

0

0

!! b

B   B B DBOLD CBF a B B ½dHbv þ CBF0  1 B ¼ M B1  B ½dHb BOLD0 CBF CBF 0 @ v0 |fflfflfflfflffl ffl{zfflfflfflfflfflffl} |fflfflfflfflfflffl{zfflfflfflfflfflffl} @|{z}

ð8Þ

C

A

B |fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl ffl}

D

ð2Þ

[dHb]v represents the concentration of deoxygenated Hb in the venous vasculature, CBV is the cerebral blood volume, and A is a proportionality constant related to the magnetic field strength and the volume susceptibility difference between blood and tissue. Combining Eqs. (1) and (2) to describe changes in R2⁎|dHb, and factoring out the resting condition,  h ib h ib  DBOLD ð3aÞ ¼ TEd A CBV dHb  CBV0 dHb v v0 BOLD0 0 !b 1   DBOLD CBV ½dHbv A b @ : ¼ TEd Ad CBV0 d ½dHbv0 1− ½dHbv0 BOLD0 CBV0 |fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl} M

ð3bÞ As we have shown in previous work, hyperoxia yields a small relative decrease in grey matter CBF, on the order of ∼7% at 100% inspired O2 (Bulte et al., 2007). It is necessary to account for this CBF decrease, which will influence the BOLD signal through a direct effect on [dHb] concentration, and through a secondary effect on CBV. These two effects are described as additional

Eq. (7) is derived from Fick's principle (Hoge et al., 1999a) and is used similarly by Hoge et al. under the assumption that CMRO2 does not change during the calibration procedure. Notably, metabolic rate does not appear to change in adults in response to hyperoxia, although changes have been observed in newborns (Mortola et al., 1992). Eq. (8) is the final form of the hyperoxiacalibrated model. Constants α and β are equivalent to those used in the hypercapnia-calibrated model (Davis et al., 1998; Uludag et al., 2004). Using experimentally acquired BOLD, CBF, and PaO2 (partial pressure of oxygen in arterial blood) data, Eq. (8) can be used to obtain the calibration value M. As already discussed, the CBF term labelled A in Eq. (8) is expected to decrease from its baseline value during the hyperoxic stimulus and will be a fraction b1. Smaller values of term A will yield an increased magnitude of the term labelled D, contributing to a decreased estimate of M. Term B in Eq. (8) is obtained via a measurement of end-tidal O2, as described below. This term is b 1 and decreases in magnitude with increasing oxygenation. Term C is a fraction slightly N 0. Since this term represents the impact of the CBF reduction on [dHb]v (which opposes the effect of term B), it makes sense that decreased CBF will increase the value of this term, thus reducing the overall magnitude of term D, and

810

P.A. Chiarelli et al. / NeuroImage 37 (2007) 808–820

increasing the calculated value of M. Note that terms A and C have opposite effects on the estimate of M, due to the inversely related influences of CBF and CBV changes on the BOLD signal. It is instructive to observe that Eq. (8) reduces to an expression analogous to the hypercapnia-calibrated model in the case where no CBF correction is applied (CBF/CBF0 = 1):   ! ½dHbv0 b DBOLD ¼M 1 ð9Þ ½dHbv BOLD0 |fflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflffl} E

Changes in [dHb]v can be estimated by reformulating standard physiological relationships of oxygen transport in blood, and by assuming a value for the baseline oxygen extraction fraction (OEF) (Eidelman and Sprung, 1992; Nelson, 1992; Varon, 1992). Although OEF is not measured here, it has been found to be remarkably consistent throughout the brain, in contrast to CMRO2 and CBF (Raichle and Gusnard, 2002). In this work, we assume the value OEF = 0.30, based on previous PET imaging data (Perlmutter et al., 1985). Arterial oxygen tension (PaO2) can be inferred via the sampling of end-tidal O2 partial pressure (PETO2) throughout the MRI experiment. The fractional oxygen saturation of arterial Hb (SaO2) may be calculated from PaO2, although this value is ≈ 1.0 at fractions of inspired oxygen (FiO2) at or above atmospheric concentration (FiO2 ≥0.21). In the general case the Severinghaus equation (Severinghaus, 1979a,b) may be used to relate measured PaO2 to SaO2, as follows: 1

!

23400 ðPaO2 Þ3 þ 150ðPaO2 Þ

ð10Þ

þ1

Once PaO2 and SaO2 are determined, the arterial oxygen content (CaO2) may be calculated, with contributions from O2 bound to Hb, and O2 dissolved in the arterial plasma. CaO2 ¼ ðud ½Hbd SaO2 Þ þ ðPaO2 d eÞ |fflfflfflfflfflfflffl{zfflfflfflfflfflfflffl} |fflfflfflffl{zfflfflfflffl}

ð12Þ

Eq. (12) indicates that the total amount of extracted O2 is assumed to remain constant during hyperoxia. Given a value of CvO2, the venous oxygen saturation of Hb (SvO2) may be calculated as follows:

SvO2 ¼

CvO2  ðPvO2 d eÞ ud ½Hb

ð13Þ ð14Þ

PvO2 represents an estimated measure of oxygen dissolved in venous plasma and is expected to be a minor fraction (b1%) of the total O2 content in venous blood, due to the high affinity of Hb for O2 and the significant deoxygenation of venous blood, even at high FiO2. The calculation of the deoxygenated fraction of Hb from SvO2 is straightforward (FdHb = 1 − SvO2). The change in the deoxygenated fraction represents the experimental x-axis variable in the hyperoxia-calibrated model ([dHb]v/[dHb]v0). Experimental design

Estimation of venous oxygen saturation

O2 bound to Hb

CvO2 ¼ CaO2  ðCaO2 j0 d OEFÞ

CvO2 ¼ ðud ½Hbd SvO2 Þ þ ðPvO2 d eÞ

Eq. (9) does, however, differ from the solution provided by Hoge et al. (1999a) in two ways: Firstly, [dHb]v0/[dHb]v is substituted for CBF/CBF0 as per Eq. (7), and secondly the Grubb exponent α is not present because CBV is assumed not to be influenced by a pure oxygenation change. Importantly, M retains the same meaning in this approach as for the hypercapnia-calibrated approach. Therefore, calibration can be performed using hyperoxia to obtain M, and subsequent analysis of functional data can be performed by substitution of the hyperoxia-derived M value into the conventional deoxyhaemoglobin dilution model described by Hoge et al. (1999a) and Davis et al. (1998).

SaO2 ¼

subjects. The resting OEF is used to calculate the amount of O2 removed from arterial blood, yielding the venous oxygen content (CvO2).

ð11Þ

dissolved O2

φ represents the species-dependent O2-carrying capacity of haemoglobin (1.34 ml O2/gHb for humans), and ε is the solubility coefficient of oxygen in blood (0.0031 mlO2/(dlblood * mm Hg)). We assume here a normal value for the concentration of haemoglobin ([Hb] = 15 g Hb * dl− 1 blood). Alternatively, this value may be measured by sampling blood from individual

Six healthy volunteers were scanned (age range 24–32, 2 female), each giving informed consent. A schematic of the experimental paradigm is provided in Fig. 1. Prior to the main experiment, an initial 165 s functional localizer experiment was performed, in which subjects were presented with a black and white checkerboard (8 Hz oscillation frequency) in a 20-s “on”/ 20-s “off” block-design with 4 repeats, and were instructed to perform a bilateral motor task coincident with the visual stimulus. This motor task consisted of voluntary cyclic opposition between the thumb and each of the fingers, according to the following pattern: (Tap 1) index, (Tap 2) middle, (Tap 3) ring, (Tap 4) little, (Tap 5) ring, and (Tap 6) middle. Whole-brain BOLD fMRI data were acquired during this time and were analyzed using the in-line data analysis package provided with the scanner software to obtain functional activation maps from which oblique slices could be prescribed that passed through the visual and motor cortices. Following the functional localizer, 5 oblique slices were prescribed to cover the primary visual and motor areas, and a second region of interest (ROI) localizer experiment was performed, in which alternating BOLD and CBF data were acquired over a period of 414 s, 4 on/off blocks plus one dummy segment, where each segment is 9 s. Simultaneous visual and motor stimulation were implemented using a 45-s “on”/45-s “off” block design. ROIs were defined for each individual subject by the overlap of statistically thresholded BOLD (z-stat N 4, cluster pthreshold N0.05) and CBF (z-stat N 2.3, cluster p-threshold N 0.05) images. This method is similar to those used in the experiments of others (Stefanovic et al., 2006; Uludag et al., 2004). The hyperoxia experiment lasted for 43 min and involved alternating epochs of hyperoxia and normoxia in a 5-min “on”/5-min “off” block design. Filtered air (21% O2) and 100% O2 were mixed to deliver fixed-inspired O2 concentrations of 21%, 40%, 60%, 80%, and 100% in a pseudo-randomized order. By investigating the full range of hyperoxic levels of FiO2, the relationship between

P.A. Chiarelli et al. / NeuroImage 37 (2007) 808–820

811

Fig. 1. Schematic of experimental design, with a functional localizer, an ROI localizer, and the main experiment. Forty-three minters was devoted to the alternating presentation of hyperoxic and normoxic stimuli.

BOLD and FiO2 can be shown to be accurately described by the proposed model. A tight-fitting facemask (Hans Rudolph Inc., Kansas City MO USA, 8920 Series) was worn throughout the experiment and ensured delivery of appropriate O2 concentrations, and respiratory composition within the mask was continuously sampled using equipment from Applied Electrochemistry Inc., Pittsburgh, USA (CD-3A CO2 sensor, S-3A/I O2 sensor, and Flowcontrol R-2 vacuum pump). Respiratory data logging was performed at intervals of 10 ms, using Powerlab software (ADInstruments, Colorado Springs, USA). MRI parameters Images were acquired on a Siemens Trio 3T MRI scanner, using an 8-channel head radiofrequency receive coil. An interlaced BOLD/pulsed arterial spin-labelling (ASL) sequence was used to collect T2⁎-weighted conventional EPI images and macrovascularcrushed Q2TIPS (Luh et al., 1999; Wong et al., 1998) cerebral perfusion images in the following scheme: [ASL tag image/BOLD image/ASL control image/BOLD image]n. BOLD measurements had TR/TE = 4.5 s/32 ms, and ASL experiments had TR/TE/ TI = 4.5 s/23 ms/1.4 s. In both BOLD and ASL, 5 slices were acquired per TR, with 4 × 4 × 6 mm3 voxel dimensions and a 64 × 64 imaging matrix. MP-RAGE T1-weighted images were acquired for registration of the EPI information to anatomical data, with 1 × 1 × 1 mm3 voxel dimensions and a 192 × 256 imaging matrix.

Data analysis Analysis of the ROI localizer and the main experiment was performed using the FMRIB Software Library (FSL) package (Smith et al., 2004). Each fMRI acquisition yielded a sequence of interleaved EPI BOLD and ASL images, which were split apart and analyzed separately. Processing steps for BOLD data included (i) 2D-FT and EPI ghost removal; (ii) motion correction (Jenkinson et al., 2002); (iii) volume-by-volume extraction of brain matter from surrounding tissue and skull (Smith, 2002); (iv) splitting images from the sensorimotor stimulation portion of the experiment and images from the hyperoxia portion into two separate data sets; (v) spatial smoothing with a 5-mm FWHM Gaussian kernel on both data sets; (vi) high-pass temporal filtering, with a cut-off of 120 s applied to the sensorimotor data and a cut-off of 540 s applied to the hyperoxia data; (vii) autocorrelation correction, using a general linear model (GLM)-based nonparametric estimation method (FMRIB's Improved Linear Model) (Woolrich et al., 2001); and (viii) voxel-wise GLM-based correlation of the signal time course with an appropriate reference model, constructed by convolution of a block design boxcar function with a gammavariate haemodynamic response model with 60-s mean lag and standard deviation to account for the very slow response to inspired gas stimuli. Analysis of Q2TIPS perfusion data included steps (i)–(iii) as described above. Temporal sinc-interpolation was performed on the set of tag images and the set of control images, shifting the signal

812

P.A. Chiarelli et al. / NeuroImage 37 (2007) 808–820

intensity of tag images to appear as if acquired 0.5 TRs later and control images to appear as if acquired 0.5 TRs earlier, followed by pair-wise subtraction of image intensity values. Steps (iv)–(vii) as listed above were not applied to the perfusion data. GLM estimation of activation (step viii) was performed in the same manner as for BOLD data. After subtraction of control and tag images, the perfusion data had a temporal resolution half that of the BOLD data set. BOLD and CBF estimates were calculated within an all-grey matter ROI, as well as within the functionally defined visual and motor cortex ROIs. The all-grey matter ROI was obtained by automated segmentation of the T1-weighted images using a Markov random field model-based segmentation tool (FAST) (Zhang et al., 2001). Functional ROIs were defined for each subject by the overlap of statistically thresholded BOLD, as described above, using data from the coregistered ROI localizer. Calculation of CBF The arterial spin-labelled (ASL) imaging signal during a hyperoxia challenge is influenced both by changes in CBF itself, and also by oxygenation-related changes in arterial T1 (Bulte et al., 2007). During a 100% inspired O2 challenge, the T1 of arterial blood decreases from its normal resting value of 1660 ms (at 3 T), by ∼17% to ∼1380 ms (Bulte et al., 2007; D'Othée et al., 2003; Lu et al., 2004). This drop in T1 is the net result of two competing effects—a lengthening of T1 due to the increased oxygenation of arterial blood (from ∼ 98% to ∼100%), and a shortening of T1 due

to the increase in dissolved plasma O2. The procedure for correcting hyperoxic CBF measurements for the confounding T1 effects has been described in a previous publication (Bulte et al., 2007). When substituted into the kinetic model (Buxton et al., 1998), the Q2TIPS experiment is expected to depend strongly on the T1 of arterial blood, but is not as affected by the estimated changes in tissue T1 or blood T2. For a pulsed ASL experiment the model is given by: 8 <¼ 0 DM ðtÞ ¼ 2M0B f ðt  DtÞaet=T1b qp ðtÞ : ¼ 2M0B f saet=T1b qp ðtÞ

0 b t b Dt Dt b t b s þ Dt s þ Dt b t

where f is the blood flow, and T1b is the T1 of arterial blood, other factors are as described in Wong et al. (1998). The change in CBF ( f ) can thus be calculated from the changes in magnetization between tag and control images (ΔM(t)) independently from changes in relaxation times as FiO2 is altered. Hyperoxic stimuli were delivered in a pseudo-random order (rather than a stepped order of increasing O2 concentration) to minimize potential distortion of the data by the high-pass temporal filtering procedure, and to minimize the influence of long-term artefacts (due to scanner heating, or subject acclimatization to the scanner environment) on the observed trend in the BOLD signal. Previous work (Bulte et al., 2007) has illustrated that the use of a mixed-order paradigm presents a potential confound for accurate CBF measurement during hyperoxia, inducing an artifactually noisy trend between CBF and FiO2 due to the long time constant

Table 1 Respiratory values calculated from each subject at normoxia (FiO2 = 0.21) and four grades of hyperoxia (FiO2 = 0.4, 0.6, 0.8, and 1.0) Subject

FiO2 (%)

PaO2 (%)

PaO2 (mm Hg)

SaO2

CaO2 (%)

CvO2 (%)

SvO2

1 − SvO2

[dhb]v0 / [dhb]v

[dhb]v / [dhb]v0

1

21 40 60 80 100 21 40 60 80 100 21 40 60 80 100 21 40 60 80 100 21 40 60 80 100 21 40 60 80 100

14.94 36.97 60.00 79.73 94.72 12.15 35.78 58.26 78.17 92.32 14.59 34.57 56.59 77.24 93.20 14.52 29.45 53.99 73.06 93.13 14.78 29.95 53.29 68.99 79.61 14.24 30.99 53.18 70.65 81.42

113.5 281.0 456.0 606.0 719.9 92.3 271.9 442.8 594.1 701.7 110.9 262.7 430.1 587.0 708.3 110.3 223.8 410.3 555.2 707.8 112.3 227.6 405.0 524.3 605.0 108.2 235.5 404.1 536.9 618.8

0.9844 0.9989 0.9998 0.9999 0.9999 0.9716 0.9988 0.9997 0.9999 0.9999 0.9833 0.9987 0.9997 0.9999 0.9999 0.9831 0.9979 0.9997 0.9999 0.9999 0.9839 0.9980 0.9996 0.9998 0.9999 0.9821 0.9982 0.9996 0.9998 0.9999

20.14 20.95 21.51 21.98 22.33 19.82 20.92 21.47 21.94 22.27 20.11 20.89 21.43 21.92 22.29 20.10 20.75 21.37 21.82 22.29 20.13 20.77 21.35 21.72 21.97 20.08 20.79 21.35 21.76 22.02

14.14 14.95 15.51 15.98 16.33 13.82 14.92 15.47 15.94 16.27 14.11 14.89 15.43 15.92 16.29 14.10 14.75 15.37 15.82 16.29 14.13 14.77 15.35 15.72 15.97 14.08 14.79 15.35 15.76 16.02

0.7034 0.7438 0.7716 0.7948 0.8125 0.6873 0.7423 0.7695 0.7930 0.8096 0.7019 0.7407 0.7675 0.7919 0.8107 0.7016 0.7339 0.7644 0.7870 0.8106 0.7027 0.7346 0.7636 0.7822 0.7947 0.7003 0.7360 0.7635 0.7842 0.7968

0.2966 0.2562 0.2284 0.2052 0.1875 0.3127 0.2577 0.2305 0.2070 0.1904 0.2981 0.2593 0.2325 0.2081 0.1893 0.2984 0.2661 0.2356 0.2130 0.1894 0.2973 0.2654 0.2364 0.2178 0.2053 0.2997 0.2640 0.2365 0.2158 0.2032

1.000 1.157 1.298 1.446 1.581 1.000 1.213 1.357 1.511 1.643 1.000 1.150 1.282 1.432 1.574 1.000 1.122 1.267 1.401 1.575 1.000 1.120 1.257 1.365 1.448 1.000 1.135 1.267 1.389 1.475

1.000 0.864 0.770 0.692 0.632 1.000 0.824 0.737 0.662 0.609 1.000 0.870 0.780 0.698 0.635 1.000 0.892 0.789 0.714 0.635 1.000 0.893 0.795 0.733 0.691 1.000 0.881 0.789 0.720 0.678

2

3

4

5

6

P.A. Chiarelli et al. / NeuroImage 37 (2007) 808–820

813

Fig. 2. Reduction in venous deoxygenated Hb, due to multiple levels of FiO2. Values are expressed as a change in the percentage saturation of Hb. Red squares represent the group average.

for PCO2 acclimatization in response to O2 changes (Haldane effect). Given that the trend between CBF and FiO2 has been established in healthy volunteers for the levels employed in this study, we measure CBF at FiO2 = 1.0 and FiO2 = 0.21, and interpolate CBF values between FiO2 = 1.0 and 0.21 using the previously observed exponential trend. Results Respiratory monitoring Table 1 provides a summary of the pertinent respiratory data from each subject. Hyperoxia yields an increase in PETO2 as well as an assumed increase in the concentration of oxygenated venous haemoglobin. Note that the values listed for FiO2 represent target levels. In practice, the inspired gasses are manually controlled, leading to unavoidable slight deviations from the target FiO2 value. Such deviation does not impact the experimental outcome, as changes in venous haemoglobin deoxygenation are calculated directly from PETO2, which is known with high precision. Breath rate was obtained from the respiratory traces and showed no tendency for subjects to increase their rate during the hyperoxic blocks in accordance with previous experience (Bulte et al., 2007); however, tidal volume could not be measured and so it is not known if minute volume was consistent throughout the experi-

Fig. 4. BOLD signal changes as a function of (a) FiO2 and (b) PaO2 within grey matter. FiO2 and PaO2 are expressed as a percentage of atmospheric pressure. Data from individual subjects (black) and averaged over the group (red) is shown, with individual values connected by lines, and a linear fit through the average values.

ment. Regardless, changes in end-tidal oxygen and carbon dioxide will dominate any CBF effects and these were continuously monitored. Fig. 2 displays the calculated amount by which [dHb]v is reduced at elevated levels of inspired O2. The data from each

Fig. 3. BOLD activation at four grades of hyperoxia (FiO2 = 0.40, 0.60, 0.80, and 1.00), within a 5-slice region prescribed to intersect the visual and motor cortices.

814

P.A. Chiarelli et al. / NeuroImage 37 (2007) 808–820

Table 2 M values obtained within grey matter, visual, and motor regions for all subjects Subject

1 2 3 4 5 6 Average

Grey matter

Visual

Motor

Mcorrected ± CIm

Muncorrected ± CIm

Mcorrected ± CIm

Muncorrected ± CIm

Mcorrected ± CIm

Muncorrected ± CIm

7.4 ± 1.5 6.3 ± 1.4 9.5 ± 1.7 9.1 ± 2.3 7.4 ± 2.0 6.1 ± 0.39 7.5 ± 0.66

6.2 ± 1.3 5.0 ± 1.2 6.9 ± 1.3 7.4 ± 1.6 6.5 ± 1.8 5.7 ± 0.34 6.3 ± 0.47

– 6.3 ± 1.4 10.8 ± 1.0 7.6 ± 1.4 6.8 ± 1.9 7.6 ± 2.2 7.5 ± 0.76

– 5.6 ± 1.2 7.1 ± 0.71 7.1 ± 1.2 6.3 ± 1.8 7.1 ± 2.0 6.6 ± 0.49

– 5.7 ± 1.3 5.7 ± 0.46 8.2 ± 1.5 7.5 ± 1.2 5.5 ± 0.27 6.3 ± 0.57

– 4.1 ± 0.95 4.1 ± 0.35 5.5 ± 0.65 5.1 ± 0.90 5.1 ± 0.25 4.7 ± 0.33

Subject 1 was scanned without performing the ROI localizer.

subject are labelled uniquely, with adjacent points connected by lines and the group average at each FiO2 shown in grey. We predict an ∼ 10% reduction in the venous deoxyhaemoglobin concentration at FiO2 = 1.0. Given that the average calculated SvO2 at normoxia is ∼ 70%, the ∼ 10% change corresponds to a Hb saturation of ∼80% in the venous vasculature during 100% inspired O2. Slight sublinear behavior between 0.8 ≤ FiO2 ≤ 1.0 may be due to changes in respiration, or due to a FiO2dependent reduction in CBF, which will act to increase the level of [dHb]v.

therefore data from this subject are only shown for the all-grey matter ROI. Data are shown both after the CBF correction and before the CBF correction, corresponding to fitting by Eqs. (8) and (9), respectively. The result of CBF correction is a consistent increase in the estimated value of M, on the order of an ∼ 5–20% change from its uncorrected value. Values for

fMRI response to hyperoxia Statistically thresholded BOLD activation maps are displayed in Fig. 3, for a single individual at FiO2 = 0.4, 0.6, 0.8, and 1.0. At 100% inspired O2, a significant BOLD signal increase occurs both in grey and white matter, although the absolute signal change is largest in grey matter. The image corresponding to FiO2 = 0.40 illustrates that the signal is enhanced near macroscale venous structures such as the sagittal sinus, where there is a large volume of dHb at rest. Note that the calibration constant, M (Eqs. (3a) (3b)), is proportional both to resting CBV and resting [dHb]β. The average signal change in grey matter at FiO2 = 1.0 is ∼ 3.3% for the subject shown in Fig. 3. Fig. 4 shows the grey matter BOLD signal change for each subject plotted over a range of FiO2 values. Group average data (grey squares) reveal linear relationships between average BOLD signal change and FiO2 (Fig. 4a), as well as PaO2 (Fig. 4b). Data from all subjects are consistently labelled, for comparison with Fig. 2 and all subsequent figures. Note that trends from individual subjects are well clustered about the mean. The average BOLD response to 100% inspired O2 is ∼ 3.1% across all grey matter, and the largest observed response in the group of 6 subjects was ∼ 3.6%. Within the visual and motor cortex ROIs specifically, group average BOLD signal changes at FiO2 = 1.0 were ∼3.3% and ∼ 2.3%, respectively. Hyperoxia calibration A summary of the M values determined using the hyperoxia calibration method is given in Table 2. Data were obtained within the all-grey matter, visual cortex, and motor cortex ROIs and are supplied as average values within the ROI ± CIm (95% confidence interval of the mean). Grey matter ROIs were included in the analysis to show the deviation from the mean grey matter values in different regions of the brain. The ROI localizer was not performed on Subject 1, and

Fig. 5. Grey matter ROI hyperoxia calibration from all subjects, used to determine M. BOLD signal change is plotted against (a) term D of Eq. (8), representing the CBF-corrected model, and (b) term E of Eq. (9), representing the uncorrected model. In both cases, M is equal to the slope of a linear fit, with error margins provided as ±CIm.

P.A. Chiarelli et al. / NeuroImage 37 (2007) 808–820

the group average (bottom row) were obtained by linear fitting of the model to the entire data set, rather than simple averaging of the M values obtained for individual subjects. The group error estimates represent CIm of the slope. Notably, estimated values of M appear in considerable agreement with those described in a previous publication, calculated on the same MRI system within a subject group of similar size, age group and gender distribution (and for which there is a 33% overlap in group members between the two studies) using the hypercapnia approach where values of Mvisual = 6.6 ± 3.4 and Mmotor = 4.3 ± 3.5 were obtained (Chiarelli et al., 2007a). The average CBF-corrected hyperoxia-based M values are larger than both of these estimates (7.5 ± 0.76 and 6.3 ± 0.57, respectively), although the values agree within the bounds of error, and both approaches suggest that a larger M value exists within the visual cortex compared to the motor cortex. A detailed analysis of M values in different brain regions as found by our own group and others using the hypercapnia method is presented in Chiarelli et al. (2007a). Fig. 5 displays the result of hyperoxia calibration in the all-grey matter ROI for all 6 subjects. A linear fit is performed between the BOLD signal change and the terms labelled D and E in Eqs. (8) and (9), respectively. Values of M are obtained from the slope, with the fit appropriately constrained to pass through the origin. Error margins (grey) in the group average data represent ±CIm of the 6 samples at each FiO2. A comparison between Figs. 5a and b demonstrates the importance of correction for the change in CBF which occurs during hyperoxia. The CBF correction represents a leftward shift along the x-axis, yielding a larger slope. Based on visual inspection of these plots, the overall distribution of values between individuals appears to change little. The high degree of

815

linearity R2 = 0.97 in these plots represents a confirmation of the suitability of this approach. Using the hyperoxia calibration, the change in [dHb]v replaces the change in CBF as the independent variable of interest. Grey matter data from each subject are displayed in Fig. 6 to illustrate the close correspondence between hyperoxia calibration data and the predictions of the deoxyhaemoglobin dilution model. In this case, both the data and the model are plotted prior to CBF correction for simplicity. Figs. 7 and 8 displays the hyperoxia calibration data in the visual and motor ROIs, respectively. The well-behaved nature of these plots, and the linearity within per-subject data across the four levels of FiO2 suggest that hyperoxia calibration is reliable and represents a valuable alternative to hypercapnia calibration. The method presented here is obviously too cumbersome to use in addition to a functional experiment, and it is far too long. Thus a simplified, shorter variant of the method was examined to test the accuracy of practical techniques which are more directly comparable to hypercapnia methods. The results of this study are presented in Appendix A. Model sensitivity analysis Using experimental data from the all-grey matter ROI, the sensitivity of the calculated M value to model parameters was tested by independent variation in α, β, and OEF0 (Fig. 9). The values for each of these parameters assumed in our previous analysis were α = 0.38, β = 1.5, and OEF = 0.3. The range of variation for OEF (0.2 to 0.4) was chosen based on physiologically plausible resting values in the brain, and the ranges of variation for α and β (0.15 to 0.45 and 1.0 to 1.6, respectively) were chosen to match sensitivity ranges employed in analyses by others (Hoge et al., 1999a; Uludag et al., 2004).

Fig. 6. Hyperoxia data plotted on BOLD−[dHb]v axes for subjects 1–6 (a–f), analogous to fits shown using the hypercapnia calibration. Using the hypercapnia approach, BOLD and CBF data are fit to an exponential relationship based on the dHb dilution model. [dHb] data as plotted on the x-axis are not corrected for CBF but show similar trends to CBF-corrected data.

816

P.A. Chiarelli et al. / NeuroImage 37 (2007) 808–820

groups. Oxygen is very economical and is readily available in most clinical settings. Furthermore, the results obtained using hyperoxia suggest a higher precision for measurement of the M calibration value compared to hypercapnia and the possibility of highprecision per-subject estimation of M. Both the hyperoxia and inhaled hypercapnia techniques require the subject to wear a tightfitting facemask throughout the experiment, although this is tolerated well by the subjects this could potentially be replaced by a 2-tube nasal cannula with the hyperoxia method. In the future, a rigorous comparison between the two techniques may be obtained by employing both techniques within a single session. Such an experiment could benefit also from the acquisition of functional data during the session, to allow for comparison of flowmetabolism coupling trends as well as M values. The post-processing stage of hyperoxia calibration is more involved compared to hypercapnia. Whereas hypercapnia calibration requires (i) the substitution of experimental CBF and BOLD signal changes into the dHb dilution model to calculate M, and (ii) use of M for estimation of fractional changes in functional CMRO2 values, the hyperoxia calibration requires (i) calculation of changes

Fig. 7. Visual ROI hyperoxia calibration from all subjects, used to determine M. BOLD signal change is plotted against (a) term D of Eq. (8), representing the CBF-corrected model, and (b) term E of Eq. (9), representing the uncorrected model. In both cases, M is equal to the slope of a linear fit, with error margins provided as ±CIm.

We observe a very low sensitivity of the M value estimate to α, and a sensitivity to β comparable to that observed in the hypercapnia approach. The sensitivity to the resting OEF value is the principle source of model-based variation in the correct estimate of M. Our choice in OEF was based on observed values (Perlmutter et al., 1985), and OEF0 has been found to vary little across regions of the cortex. Given that the determined value of M is positively correlated with the value of the CMRO2−CBF coupling constant (Chiarelli et al., 2007b), a larger resting OEF (higher M) would imply larger proportional changes in CMRO2 per unit CBF during functional activation. Discussion Hyperoxia calibration in context: potential limitations and future directions Hyperoxia calibration is valuable because it eliminates the need for CO2, thus providing the possibility of longer tolerated paradigms and application to an increased number of patient

Fig. 8. Motor ROI hyperoxia calibration from all subjects, used to determine M. BOLD signal change is plotted against (a) term D of Eq. (8), representing the CBF-corrected model, and (b) term E of Eq. (9), representing the uncorrected model. In both cases, M is equal to the slope of a linear fit, with error margins provided as ±CIm.

P.A. Chiarelli et al. / NeuroImage 37 (2007) 808–820

Fig. 9. Sensitivity of the hyperoxia-calibrated model to variations in α, β, and resting-state OEF. The following parameter variations were used 0.15 b α b 0.45, 1 b β b 1.6, and 0.2 b OEF0 b 0.4. The initial default values employed in the calibration used in this study were α = 0.38, β =1.5, and OEF0 =0.4.

in [dHb]v from experimental PaO2 values, (ii) correction of experimental perfusion data for changes in arterial blood T1, (iii) use of BOLD, CBF, and [dHb]v changes to calculate M, and (iv) use of M for estimation of fractional changes in functional CMRO2 values, through the dHb dilution equation. The analysis of hyperoxia data in comparison to hypercapnia is somewhat simplified though by the fact that the variations found in vascular responses across the population, which can be a confound for hypercapnia are not an issue for hyperoxia. The reason for this being that the change in BOLD signal is almost entirely due to changes in the SvO2, which is measured via the end-tidal values, thus there is minimal vascular element. By using the end-tidal time course as the functional analysis regressor any subject dependent influences are accounted for. Hyperoxia is known to induce mild vasoconstriction, both due to the direct influence of elevated O2 on the vasculature, and due to a drop in PaCO2 resulting from the Haldane effect (Demchenko et al., 2002; Floyd et al., 2003). At FiO2 = 1.0, a small reduction in CBF was observed within the all-grey matter, visual cortex, and motor cortex ROIs (Fig. 10). CBF values are corrected for the altered T1 of blood during hyperoxia, through a previously described fitting method based on the kinetic model for QUIPSS II arterial spin labelling MRI described by Buxton et al. (Bulte et al., 2007; Buxton et al., 1998; Wong et al., 1998). Hyperoxia has been shown to have little or no effect on CBV even at FiO2 = 1.0 (Kolbitsch et al., 2002; Reinstrup et al., 2001). Similarly changing from FiO2 = 0.21 to 1.0 has been shown to have no effect on CMRO2 (Sicard and Duong, 2005). The hyperoxia-calibrated model relies on the assumption that an increase in FiO2 does not affect resting CMRO2. This assumption may also be tested in future work, potentially using PET to measure glucose uptake. The dependence of M on the value of OEF0 poses a potential limitation, as this value must either be acquired separately (using PET or SPECT), or assumed based on previous work of others. The lack of a dependence on OEF is a key benefit of the dHb dilution model developed by Davis et al. (1998) (Hoge et al., 1999a).

817

Differences in OEF0 are small across the cerebral cortex, but may still contribute to regional variation in the calculated value of M, and will clearly be a confounding factor in certain disease populations. It is therefore of value to investigate the potential for absolute measurement of regional OEF0 estimates. At high arterial pressures (PaO2 ≥ 350 mm Hg), O2 dissolved in the plasma may also influence MR contrast, exerting a slight attenuating effect on the T2⁎-weighted signal (Berkowitz, 1997). This change is likely to affect primarily the arterial side of the vasculature, where appreciable plasma concentrations will exist during FiO2 ≥ 0.60. Note that significant signal attenuation due to plasma O2 would be expected to yield BOLD signal changes at high FiO2 consistently below the predictions of the calibrated model, which we observe not to be the case (Fig. 6). Despite the fact that T2⁎ attenuation from plasma O2 is not a primary concern, future work with the hyperoxia-calibrated model could apply a correction term to the measured BOLD signal increase, accounting for this effect. A potential limitation of the hyperoxia technique is the fact that under certain conditions the administration of increased fractions of oxygen can be contraindicated. Hyperoxia is known to cause adverse responses such as lung toxicity under hyperbaric conditions or after long periods of exposure. However, under the conditions for which oxygen would be administered for BOLD calibration, it is extremely unlikely to result in any complications. Brueckl et al. have noted that: “The length of the initiation phase preceding changes in global lung function and morphology varies inversely with the concentration of oxygen and has been reported to range between 14 and 30 h at O2 concentrations of 70% or higher in healthy human subjects. Shorter episodes of normobaric hyperoxia are therefore generally considered as clinically safe” (Brueckl et al., 2006). This is, however, not the case for patients suffering from certain conditions that would exclude them from being administered even mild levels of hyperoxia. For example patients suffering from chronic obstructive pulmonary disease (COPD), for whom even slight increases in the inspired oxygen fraction can result in adverse effects (New, 2006), should be excluded.

Fig. 10. Comparison of the estimated reduction in CBF within all-grey matter, visual cortex and motor cortex ROI's, during 100 % inspired O2.

818

P.A. Chiarelli et al. / NeuroImage 37 (2007) 808–820

technique of hypercapnia calibration through the inhalation of CO2-enriched air. Rather than using changes in CBF to induce an oxygenation increase in the venous vasculature, a direct change in Hb oxygenation is implemented to induce the same effect. An alternative derivation of the dHb dilution model has been developed that accounts for a small reduction in CBF during hyperoxia, along with a correction for arterial T1 in the ASL measurement. Results obtained using hyperoxia to estimate the calibration constant (M) are in good agreement with previous data acquired using a hypercapnia calibration. Hyperoxia calibration yields group estimates with lower apparent variability and is potentially applicable to a wider range of subjects than the hypercapnia-based approach. We describe the need for future work investigating the dependence of the hyperoxia-calibrated model on resting OEF. Acknowledgments The authors would like to thank the following funding bodies: The UK Medical Research Council, The UK Engineering and Physical Sciences Research Council, The Rhodes Trust, The Wellcome Trust, The Wingate Foundation, and The Dunhill Medical Trust. Appendix A

Fig. A1. M values in the motor and visual cortices for each subject as calculated from each of the 2 separate scans using the simplified, practical hyperoxia method.

Hyperoxia calibration produces substantially smaller errors for both individual and group data, compared to data we previously published using the hypercapnia calibration (Chiarelli et al., 2007a). We estimate an ∼ 2–4 × reduction in errors when using the practical hyperoxia method described as opposed to hypercapnia calibration methods (see Appendix A). For individual subjects, error margins using hypercapnia calibration are impracticably large (greater than the magnitude of the M measurement itself). Using hyperoxia, individual confidence intervals (CIm) range between 0.25 and 2.3. Such estimates make per-subject calibration a more feasible technique. The majority of the decrease in variance is due to the high signal-to-noise ratio associated with end-tidal respiratory measurements in comparison to the inherently low SNR found with flow measurements using ASL. To provide a full and fair comparison between hyperoxia and hypercapnia would require a series of experiments where a practical protocol was employed that administered both techniques to the same subject, using the same system, within the same session. However, as the hypercapnia method is not yet at the stage of a “gold standard” itself the ideal test of the methods would be to compare the relative changes in metabolism calculated by both the hyperoxia and hypercapnia methods to those measured using a PET method such as the triple-oxygen technique (Ito et al., 2005). Conclusion A new method has been described that uses hyperoxia to calibrate the BOLD fMRI signal for estimation of CMRO2. Hyperoxia calibration represents an alternative to the current

Having demonstrated the viability of the hyperoxia technique, a practical variant of the method would be to administer two 5-min blocks of FiO2 = 0.5 at the end of an experiment with graded neurological stimulation. This would be equivalent in length to most practical hypercapnia techniques used (Chiarelli et al., 2007a). To eliminate the need to wear a facemask throughout the experiment the oxygen could be administered easily using a 2-tube nasal cannula which delivers gas from one line while sampling from the other. The exact mixture delivered is not critical as long as the expired pO2 is accurately measured. A standard respiratory monitor would suffice to collect the data providing it could output the values to a monitoring computer. Such systems are readily available in most clinical settings. A pilot study was conducted on 5 healthy male volunteers (aged 24–34). A functional localizer scan was first performed, as before, using 24 slice BOLD to cover the whole brain and yield activation ROIs for the motor and visual cortices. The subjects were then scanned using a simple gradient echo BOLD sequence (TR/ TE = 4.5 s/32 ms) lasting 20 min and 6 s, during which they experienced two 5-min blocks of FiO2 = 0.5, (3 OFF: 5 ON: 4 OFF: 5 ON: 3 OFF). Gases were delivered and monitored using the same system as described previously. Each subject was scanned twice on the same day during the same session. For the analysis it was assumed that the decrease in perfusion caused by the oxygen was 5% (Bulte et al., 2007). FSL tools were Table A1 Mean corrected M values across subjects as calculated using the simplified, short, single level, hyperoxia method (OEF 0.4), and by a comparable hypercapnia method Method

Bo

β

T Hyperoxia short Stefanovic et al. (2006)

3 1.5

TE

n

ms 1.3 1.5

32 50

4 12

Visual

Motor

M ± sd

M ± sd

8.7 ± 0.7 7.6 ± 1.3

7.3 ± 0.4 6.1 ± 1.1

P.A. Chiarelli et al. / NeuroImage 37 (2007) 808–820

used for analysis of the BOLD data, and the BOLD signal change (%) was calculated for each subject and each session in the motor and visual cortices. The end-tidal oxygen values measured during the two oxygen blocks in a single scan were averaged to produce an average PaO2 estimate for that scan. These values were then substituted into Eq. (8) to produce an estimate of M in each region for each subject during each scan. The results of these calculations are shown in Fig. A1. As can be seen, the results are consistent between subjects and between scans 1 and 2. The data from the subject represented by the triangle were discarded as outliers as the BOLD values calculated for the second scan were nearly 5 times greater than the first scan and were substantially larger than any other hyperoxia BOLD response (even those at FiO2 = 1.0). The mean value of M was calculated for each remaining subject for the two scans and then these 4 values were averaged to produce the results shown in Table A1. For comparison the values in these regions calculated by Stefanovic et al. (2006) using a comparable hypercapnia method are included. The practical hyperoxia method is thus shown to have excellent retest results, to produce consistent results both intersession and intersubject, and to have a consistently lower variance than comparable hypercapnia methods. Further improvements could potentially be made to the technique to reduce scan time, simplify the equipment, and improve patient comfort. References Berkowitz, B.A., 1997. Role of dissolved plasma oxygen in hyperoxiainduced contrast. Magn. Reson. Imaging 15, 123–126. Bickler, P.E., Litt, L., Banville, D.L., Severinghaus, J.W., 1988. Effects of acetazolamide on cerebral acid–base balance. J. Appl. Physiol. 65, 422–427. Boxerman, J.L., Bandettini, P.A., Kwong, K.K., Baker, J.R., Davis, T.L., Rosen, B.R., Weisskoff, R.M., 1995. The intravascular contribution to fMRI signal change: Monte Carlo modeling and diffusion-weighted studies in vivo. Magn. Reson. Med. 34, 4–10. Brueckl, C., Kaestle, S., Kerem, A., Habazettl, H., Krombach, F., Kuppe, H., Kuebler, W.M., 2006. Hyperoxia-induced reactive oxygen species formation in pulmonary capillary endothelial cells in situ. Am. J. Respir. Cell Mol. Biol. 34, 453–463. Bulte, D., Alfonsi, J., Bells, S., Noseworthy, M.D., 2006. Vasomodulation of BOLD signal in skeletal muscle. J. Magn. Reson. Imaging 24, 886–890. Bulte, D.P., Chiarelli, P.A., Wise, R.G., Jezzard, P., 2007. Cerebral perfusion response to hyperoxia. J. Cereb. Blood Flow Metab. 27, 69–75. Buxton, R.B., Frank, L.R., Wong, E.C., Siewert, B., Warach, S., Edelman, R.R., 1998. A general kinetic model for quantitative perfusion imaging with arterial spin labeling. Magn. Reson. Med. 40, 383–396. Chiarelli, P.A., Bulte, D.P., Gallichan, D., Piechnik, S.K., Wise, R., Jezzard, P., 2007a. Flow-metabolism coupling in human visual, motor, and supplementary motor areas assessed by magnetic resonance imaging. Magn. Reson. Med. 57, 538–547. Chiarelli, P.A., Bulte, D.P., Piechnik, S., Jezzard, P., 2007b. Sources of systematic bias in hypercapnia-calibrated functional MRI estimation of oxygen metabolism. NeuroImage 34, 35–43. Davis, T.L., Kwong, K.K., Weisskoff, R.M., Rosen, B.R., 1998. Calibrated functional MRI: mapping the dynamics of oxidative metabolism. Proc. Natl. Acad. Sci. U. S. A. 95, 1834–1839. Demchenko, I.T., Oury, T.D., Crapo, J.D., Piantadosi, C.A., 2002. Regulation of the brain's vascular responses to oxygen. Circ. Res. 91, 1031–1037. D'Othée, B.J., Rachmuth, G., Munasinghe, J., Lang, E.V., 2003. The effect of hyperoxygenation on T1 relaxation time in vitro. Acad. Radiol. 10, 854–860.

819

Eidelman, L.A., Sprung, C.L., 1992. Direct measurements and derived calculations using the pulmonary artery catheter. In: Sprung, C.L. (Ed.), The Pulmonary Artery Catheter: Methodology and Clinical Applications. Critical Care Research Associates, Closter, NJ, pp. 101–118. Floyd, T.F., Clark, J.M., Gelfand, R., Detre, J.A., Ratcliffe, S., Guvakov, D., Lambertsen, C.J., Eckenhoff, R.G., 2003. Independent cerebral vasoconstrictive effects of hyperoxia and accompanying arterial hypocapnia at 1 ATA. J. Appl. Physiol. 95, 2453–2461. Fujita, N., Matsumoto, K., Tanaka, H., Watanabe, Y., Murase, K., 2006. Quantitative study of changes in oxidative metabolism during visual stimulation using absolute relaxation rates. NMR Biomed. 19, 60–68. Grubb Jr., R.L., Raichle, M.E., Eichling, J.O., Ter Pogossian, M.M., 1974. The effects of changes in PaCO2 on cerebral blood volume, blood flow, and vascular mean transit time. Stroke 5, 630–639. Handwerker, D.A., Gazzaley, A., Inglis, B.A., D'Esposito, M., 2007. Reducing vascular variability of fMRI data across aging populations using a breathholding task. Hum. Brain Mapp., doi:10.1002/hbm.20307. Hoge, R.D., Atkinson, J., Gill, B., Crelier, G.R., Marrett, S., Pike, G.B., 1999a. Investigation of BOLD signal dependence on cerebral blood flow and oxygen consumption: the deoxyhemoglobin dilution model. Magn. Reson. Med. 42, 849–863. Hoge, R.D., Atkinson, J., Gill, B., Crelier, G.R., Marrett, S., Pike, G.B., 1999b. Linear coupling between cerebral blood flow and oxygen consumption in activated human cortex. Proc. Natl. Acad. Sci. U. S. A. 96, 9403–9408. Hoge, R.D., Atkinson, J., Gill, B., Crelier, G.R., Marrett, S., Pike, G.B., 1999c. Stimulus-dependent BOLD and perfusion dynamics in human V1. NeuroImage 9, 573–585. Ito, H., Ibaraki, M., Kanno, I., Fukuda, H., Miura, S., 2005. Changes in cerebral blood flow and cerebral oxygen metabolism during neural activation measured by positron emission tomography: comparison with blood oxygenation level-dependent contrast measured by functional magnetic resonance imaging. J. Cereb. Blood Flow Metab. 25, 371–377. Jenkinson, M., Bannister, P., Brady, M., Smith, S., 2002. Improved optimization for the robust and accurate linear registration and motion correction of brain images. NeuroImage 17, 825–841. Kastrup, A., Li, T.Q., Takahashi, A., Glover, G.H., Moseley, M.E., 1998. Functional magnetic resonance imaging of regional cerebral blood oxygenation changes during breath holding. Stroke 29, 2641–2645. Kastrup, A., Krüger, G., Glover, G.H., Moseley, M.E., 1999a. Assessment of cerebral oxidative metabolism with breath holding and fMRI. Magn. Reson. Med. 42, 608–611. Kastrup, A., Li, T.Q., Glover, G.H., Moseley, M.E., 1999b. Cerebral blood flow-related signal changes during breath-holding. Am. J. Neuroradiol. 20, 1233–1238. Kastrup, A., Krüger, G., Neumann-Haefelin, T., Glover, G.H., Moseley, M.E., 2002. Changes of cerebral blood flow, oxygenation, and oxidative metabolism during graded motor activation. NeuroImage 15, 74–82. Kim, S.G., Rostrup, E., Larsson, H.B.W., Ogawa, S., Paulson, O.B., 1999. Determination of relative CMRO2 from CBF and BOLD changes: significant increase of oxygen consumption rate during visual stimulation. Magn. Reson. Med. 41, 1152–1161. Kliefoth, A.B., Grubb Jr, R.L., Raichle, M.E., 1979. Depression of cerebral oxygen utilization by hypercapnia in the rhesus monkey. J. Neurochem. 32, 661–663. Kolbitsch, C., Lorenz, I.H., Hörmann, C., Hinteregger, M., Löckinger, A., Moser, P.L., Kremser, C., Schocke, M., Felber, S., Pfeiffer, K.P., Benzer, A., 2002. The influence of hyperoxia on regional cerebral blood flow (rCBF), regional cerebral blood volume (rCBV) and cerebral blood flow velocity in the middle cerebral artery (CBFVMCA) in human volunteers. Magn. Reson. Imaging 20, 535–541. Kwong, K.K., Wanke, I., Donahue, K.M., Davis, T.L., Rosen, B.R., 1995. EPI imaging of global increase of brain MR signal with breath-hold preceded by breathing O2. Magn. Reson. Med. 33, 448–452. Lu, H., Clingman, C., Golay, X., van Zijl, P.C., 2004. Determining the

820

P.A. Chiarelli et al. / NeuroImage 37 (2007) 808–820

longitudinal relaxation time (T1) of blood at 3.0 Tesla. Magn. Reson. Med. 52, 679–682. Luh, W.M., Wong, E.C., Bandettini, P.A., Hyde, J.S., 1999. QUIPSS II with thin-slice TI1 periodic saturation: a method for improving accuracy of quantitative perfusion imaging using pulsed arterial spin labeling. Magn. Reson. Med. 41, 1246–1254. Mortola, J.P., Frappell, P.B., Dotta, A., Matsuoka, T., Fox, G., Weeks, S., Mayer, D., 1992. Ventilatory and metabolic responses to acute hyperoxia in newborns. Am. Rev. Respir. Dis. 146, 11–15. Nelson, L.D., 1992. Mixed venous oxygen measurements. In: NJ. C. (Ed.), The Pulmonary Artery Catheter: Methodology and Clinical Applications. Critical Care Research Associates, Closter, NJ, pp. 157–174. New, A., 2006. Oxygen: kill or cure? Prehospital hyperoxia in the COPD patient. Emerg. Med. J. EMJ 23, 144–146. Perlmutter, J.S., Herscovitch, P., Powers, W.J., 1985. Standardized mean regional method for calculating global positron emission tomographic measurements. J. Cereb. Blood Flow Metab. 5, 476–480. Raichle, M.E., Gusnard, D.A., 2002. Appraising the brain's energy budget. Proc. Natl. Acad. Sci. U. S. A. 99, 10237–10239. Reinstrup, P., Ryding, E., Ohlsson, T., Dahm, P.L., Uski, T., 2001. Cerebral blood volume (CBV) in humans during normo- and hypocapnia: influence of nitrous oxide (N2O). Anesthesiology 95, 1079–1082. Rostrup, E., Larsson, H.B., Toft, P.B., Garde, K., Henriksen, O., 1995. Signal changes in gradient echo images of human brain induced by hypo- and hyperoxia. NMR Biomed. 8, 41–47. Rostrup, E., Law, I., Blinkenberg, M., Larsson, H.B.W., Born, A.P., Holm, S., Paulson, O.B., 2000. Regional differences in the CBF and BOLD responses to hypercapnia: a combined PET and fMRI study. NeuroImage 11, 87–97. Severinghaus, J.W., 1979a. Current trends in continuous blood gas monitoring. Biotelem. Patient Monit. 6, 9–15. Severinghaus, J.W., 1979b. Simple, accurate equations for human blood O2 dissociation computations. J. Appl. Physiol. 46, 599–602. Sicard, K.M., Duong, T.Q., 2005. Effects of hypoxia, hyperoxia, and hypercapnia on baseline and stimulus-evoked BOLD, CBF, and CMRO2 in spontaneously breathing animals. NeuroImage 25, 850–858. Smith, S.M., 2002. Fast robust automated brain extraction. Hum. Brain Mapp. 17, 143–155.

Smith, S.M., Jenkinson, M., Woolrich, M.W., Beckmann, C.F., Behrens, T.E.J., Johansen-Berg, H., Bannister, P.R., De Luca, M., Drobnjak, I., Flitney, D.E., Niazy, R.K., Saunders, J., Vickers, J., Zhang, Y., De Stefano, N., Brady, J.M., Matthews, P.M., 2004. Advances in functional and structural MR image analysis and implementation as FSL. NeuroImage 23, S208–S219. Stefanovic, B., Warnking, J.M., Pike, G.B., 2004. Hemodynamic and metabolic responses to neuronal inhibition. NeuroImage 22, 771–778. Stefanovic, B., Warnking, J.M., Kobayashi, E., Bagshaw, A.P., Hawco, C., Dubeau, F., Gotman, J., Pike, G.B., 2005. Hemodynamic and metabolic responses to activation, deactivation and epileptic discharges. NeuroImage 28, 205–215. Stefanovic, B., Warnking, J.M., Rylander, K.M., Pike, G.B., 2006. The effect of global cerebral vasodilation on focal activation hemodynamics. NeuroImage 30, 726–734. Thomason, M.E., Foland, L.C., Glover, G.H., 2007. Calibration of BOLD fMRI using breath holding reduces group variance during a cognitive task. Hum. Brain Mapp. 28, 59–68. Uludag, K., Dubowitz, D.J., Yoder, E.J., Restom, K., Liu, T.T., Buxton, R.B., 2004. Coupling of cerebral blood flow and oxygen consumption during physiological activation and deactivation measured with fMRI. NeuroImage 23, 148–155. Varon, A.J., 1992. Hemodynamic monitoring: arterial and pulmonary artery catheters. In: Civetta, J.M., Taylor, R.W., Kirby, R.R. (Eds.), Critical Care. J.B. Lippincott, Philadelphia, pp. 255–270. Vazquez, A.L., Cohen, E.R., Gulani, V., Hernandez-Garcia, L., Zheng, Y., Lee, G.R., Kim, S.G., Grotberg, J.B., Noll, D.C., 2006. Vascular dynamics and BOLD fMRI: CBF level effects and analysis considerations. NeuroImage 32, 1642–1655. Wong, E.C., Buxton, R.B., Frank, L.R., 1998. Quantitative imaging of perfusion using a single subtraction (QUIPSS and QUIPSS II). Magn. Reson. Med. 39, 702–708. Woolrich, M.W., Ripley, B.D., Brady, M., Smith, S.M., 2001. Temporal autocorrelation in univariate linear modeling of fMRI data. NeuroImage 14, 1370–1386. Zhang, Y.Y., Brady, M., Smith, S., 2001. Segmentation of brain MR images through a hidden Markov random field model and the expectationmaximization algorithm. IEEE Trans. Med. Imag. 20, 45–57.