High temporal resolution MRI quantification of global cerebral metabolic rate of oxygen consumption in response to apneic challenge

Abstract

We present a technique for quantifying global cerebral metabolic rate of oxygen consumption (CMRO₂) in absolute physiologic units at 3-second temporal resolution and apply the technique to quantify the dynamic CMRO₂ response to volitional apnea. Temporal resolution of 3 seconds was achieved via a combination of view sharing and superior sagittal sinus-based estimation of total cerebral blood flow (tCBF) rather than tCBF measurement in the neck arteries. These modifications were first validated in three healthy adults and demonstrated to produce minimal errors in image-derived blood flow and venous oxygen saturation (S_vO₂) values. The technique was then applied in 10 healthy adults during an apnea paradigm of three repeated 30-second breath-holds. Subject-averaged baseline tCBF, arteriovenous oxygen difference (AVO₂D), and CMRO₂ were 48.6±7.0 mL/100 g per minute, 29.4±3.4 %HbO₂, and 125.1±11.4 μmol/100 g per minute, respectively. Subject-averaged maximum changes in tCBF and AVO₂D were 43.5±9.4% and −32.1±5.7%, respectively, resulting in a small (6.0±3.5%) but statistically significant (P=0.00044, two-tailed t-test) increase in average end-apneic CMRO₂. This method could be used to investigate neurometabolic–hemodynamic relationships in normal physiology, to better define the biophysical origins of the BOLD signal, and to quantify neurometabolic responsiveness in diseases of altered neurovascular reactivity.

Introduction

Because cerebral metabolism is almost entirely oxidative, continuous O₂ delivery to the brain is critical and tightly regulated. The cerebral metabolic rate of oxygen consumption (CMRO₂), defined as the brain oxygen consumption rate per unit tissue mass, is a direct measure of oxidative metabolism, in contrast to indirect measures such as perfusion or blood-oxygen-level-dependent (BOLD) magnetic resonance (MR) imaging signal. Therefore, CMRO₂ is an ideal parameter for investigating relationships between neuronal activity, blood flow, and cerebral metabolism in normal physiology and diseases of cerebrometabolic dysfunction. In fact, alterations in cerebral oxygen metabolism are associated with many of the most common neurologic disorders, including Alzheimer's, Parkinson's, Huntington's, and multiple sclerosis.^{1, 2, 3, 4}

In recent years, significant progress has been made toward non-invasive MR-based methods for absolute CMRO₂ quantification. Much attention has focused on developing methods to quantify CMRO₂ absolutely (in physiologic units) and on a voxel-wise basis. Such voxel-wise methods model the effect of deoxygenated hemoglobin on either brain tissue T₂′ (ref. 5), T₂ (ref. 6), or BOLD signal⁷ to quantify the voxel-wise venous oxygen saturation (S_vO₂), which can be combined with arterial spin labeling cerebral blood flow (CBF) measurement to yield CMRO₂. In contrast, methods that quantify oxygen extraction globally model the effect of deoxygenated hemoglobin on the intravascular T₂ (refs 8, 9) or MR signal phase^{10, 11} of large veins to quantify intravascular S_vO₂, which combined with phase-contrast MR-based quantification of CBF yields CMRO₂. Although these intravascular methods lack the ability to measure local changes in oxygen metabolism, many physiologic states and neurologic disorders are global in nature, and therefore assessable via measurement of global CMRO₂. Furthermore, voxel-wise CMRO₂ techniques require many minutes for each CMRO₂ measurement and therefore cannot quantify changes in response to dynamic physiologic challenges or neurologic stimuli. By sacrificing spatial specificity, intravascular methods enable CMRO₂ quantification in clinically feasible scan times and at much higher temporal resolutions, seconds rather than minutes, in comparison to voxel-wise approaches.

Based on the long paramagnetic cylinder model,^{12, 13} MR susceptometry-based oximetry is a simple and robust method for intravascular CMRO₂ quantification. Unlike T₂ relaxation-based methods for quantifying intravascular S_vO₂, the paramagnetic cylinder model approach does not require prior calibration to specific scanners, sequences, or blood hemoglobin (Hb) or hematocrit (Hct) values (Hb and Hct are input parameters theoretically included in the model). It also has equal accuracy and precision across all S_vO₂ values and is scalable with field strength. These features make the model suitable for application to a variety of clinical populations and experimental conditions, including longitudinal and multi-center studies. The simplicity of this approach also enables rapid CMRO₂ quantification in response to stimuli. For example, in a recent work by some of the authors,¹⁴ CMRO₂ was measured at 25-second temporal resolution in response to hypercapnia by application of a susceptometry-based oximetry method and found to be constant during hypercapnic steady state. Though 25-second temporal resolution represents a drastic improvement over previous approaches, changes in cerebral oxygen supply and demand take place on the order of seconds, and thus require yet improved temporal resolution to be fully resolved. Calibrated BOLD-based methods can assess relative CMRO₂ changes in seconds,¹⁵ however, such methods cannot quantify CMRO₂ in absolute physiologic units. Furthermore, these BOLD-based methods require calibration via gas mixture breathing, complicating application to human subjects, and are based on the assumption that the response to such gases is isometabolic, itself a topic of debate.¹⁶

Higher temporal resolution CMRO₂ quantification would provide valuable insights into global neuronal activity during various dynamic stimuli. For instance, it could be applied to validate whether aforementioned gas mixture breathing stimuli used in calibrating the BOLD functional magnetic resonance imaging (fMRI) signal, including hypercapnic¹⁷ and hyperoxic¹⁸ gas mixture breathing as well as breath-hold,¹⁹ are in fact isometabolic and over what time frame (i.e., whether a delay exists in reaching an isometabolic steady state). Validating these assumptions is critical given the extensive use of fMRI in biomedical research and the growing interest in making BOLD fMRI more quantitative. Further, applying the dynamic CMRO₂ method to neural activation tasks could help elucidate the biophysical mechanisms underlying the BOLD response, including the relative CMRO₂ contribution to the BOLD poststimulus undershoot, a topic of significant contention.²⁰

Breath-hold apnea is another important area of investigation where high temporal resolution CMRO₂ measurement is essential. Apnea is involved in a number of important diseases, such as asthma, chronic obstructive pulmonary diseases, and obstructive sleep apnea (OSA). The normal physiologic response to apnea maintains cerebral oxygen delivery via reduced cardiac output, peripheral vasoconstriction, and cerebral vasodilation.²¹ However, it has been suggested that in OSA, the repeated nocturnal apneic events caused by upper airway mechanical failure may result in blunting of this normal response,^{22, 23, 24} potentially explaining the extensive neurologic pathology associated with the disease. Exploration of this hypothesis requires better methods for quantifying the cerebrometabolic apneic response. While non-MR methods such as Doppler ultrasound²³ or near infrared spectroscopy²⁴ have been applied to study the neurometabolic response to apnea in subjects with OSA, these techniques measure changes in either CBF or tissue O₂ saturation, but not the CMRO₂, which requires simultaneous quantification of CBF and tissue oxygen extraction. CMRO₂ is maintained across healthy subjects both at baseline¹⁰ and in response to certain physiologic stimuli such as hypercapnia,¹⁴ suggesting that it is a more significant index for assessing neurovascular dysfunction than either blood flow or oxygenation alone. Developing methods to better assess the normal CMRO₂ response to apnea and its potential alteration in OSA could improve understanding of OSA-associated neuropathology and provide insight into OSA treatment.

In this paper, we present and validate a method for dynamic CMRO₂ quantification with 3-second temporal resolution, which extends the susceptometry-based oximetry approach previously described.¹⁰ This temporal resolution is achieved by using view sharing to reduce the number of phase-encode lines by four-fold and by combining the S_vO₂ and flow quantification portions of the pulse sequence. After validating the assumptions inherent in these temporal resolution improving measures, the technique was applied to a cohort of ten healthy individuals during a volitional apnea paradigm, both to demonstrate the method's sensitivity and characterize the normal apneic CMRO₂ response.

Materials and Methods

Principles of Susceptometry-Based Global Cerebral Metabolic Rate of Oxygen Consumption Quantification

The cerebral metabolic rate of oxygen is estimated by combining venous and arterial oxygen saturation and total cerebral blood flow (tCBF) measurements using Fick's principle,²⁵

where CMRO₂ is the cerebral metabolic rate of oxygen consumption in μmol per minute per 100 g brain tissue, tCBF the total cerebral blood flow in mL per 100 g brain tissue per minute, S_aO₂ and S_vO₂ the percent hemoglobin saturation of the intravascular arterial and venous blood, respectively, and C_aO₂ the arterial oxygen content in μmol of O₂ per 100 mL blood, a product of the measured Hb concentration and the theoretical O₂ carrying capacity of Hb. The C_aO₂ is calculated from the ideal gas law using a venipuncture-derived Hb value and assuming an O₂ carrying capacity of 1.39 mL of O₂ per gram of Hb.²⁶ S_aO₂ is measured continuously with a digital pulse oximeter. This leaves S_vO₂ and tCBF to be quantified from the MR imaging experiment.

Magnetic resonance-based blood oximetry is used to quantify S_vO₂. By modeling a large vessel of interest as a long paramagnetic cylinder and accounting for field cancelation due to the Lorentz sphere phenomenon,^{12, 13} the average field difference between blood and a surrounding reference tissue can be computed as

where Δχ is the volume magnetic susceptibility difference between blood and surrounding reference tissue and θ the tilt angle of the vessel relative to the main magnetic field (B₀). For whole blood,

where Hct is venipuncture-derived hematocrit Δχ_do the volume susceptibility difference between fully deoxygenated and oxygenated erythrocytes, and Δχ_oxy the volume susceptibility difference between fully oxygenated erythrocytes and water. Values of 4π(0.273) and 4π(−0.008) p.p.m. (SI units) are used for Δχ_do and Δχ_oxy, respectively, based on ex vivo calibration experiments.^{27, 28} Equations (2) and (3) suggest that deoxygenated blood will have an increased local magnetic field, causing an average phase difference of the MR signal relative to the surrounding tissue over a time ΔTE given by

In the magnetic resonance imaging experiment, a multi-echo gradient-recalled echo is used to obtain a phase-difference map of the vessel and surrounding reference tissue with echo difference time ΔTE. Quadratic fitting is used to remove static field inhomogeneities that contribute to the phase accrual, Δφ, between intravascular blood and surrounding reference tissue.²⁹ Combining equations (2), (3), (4) enables determination of S_vO₂ as a function of Δφ,

Because the superior sagittal sinus (SSS) is long, relatively straight, and virtually parallel to the B₀ field with the subject lying supine, it is an excellent candidate for application of the long paramagnetic cylinder model. Application of the model to the SSS has been validated both theoretically³⁰ and with anatomic phantom models.¹⁰ Furthermore, the SSS is the largest cerebral vein, and it has been shown that oxygen saturation levels in the SSS are comparable to global cerebral S_vO₂ levels measured in the internal jugular vein,⁸ making the SSS an appropriate surrogate for global cerebral S_vO₂. Direct susceptometry-based measurement of S_vO₂ in the internal jugular vein is difficult because of the often severe susceptibility artifacts caused by the proximity of air spaces such as the oral cavity and trachea.

Non-gated phase-contrast magnetic resonance imaging is used to quantify tCBF. The method utilizes motion-sensitizing gradient waveforms to encode information about velocity into the phase of the MR signal. Specifically, the pulse sequence involves two interleaves; both having null zeroth-gradient moment along the direction of blood flow but non-zero first-gradient moment. The latter determines the sensitivity of the accrued phase difference between the two interleaves, Δϕ, to the velocity of the flowing spins as

where ΔM₁ is the difference in the first moment between the two interleaves and is dictated by a user-defined parameter VENC, defined as

VENC represents the velocity that causes a net phase accrual of π radians and therefore the maximum velocity that can be resolved without phase aliasing, and is typically chosen to be approximately 30% higher than the maximum velocity expected. Flow is quantified from velocity maps via multiplication of average vessel blood flow velocity by vessel cross-sectional area. To quantify CMRO₂ per unit brain mass, flow must be normalized to total brain volume, which is quantified with a T₁-weighted 3D magnetization-prepared rapid gradient-recalled-echo pulse sequence.³¹

In the previous work, these phase-based techniques for quantifying S_vO₂ and tCBF have been combined to quantify global CMRO₂ at rest¹⁰ and during hypercapnic gas breathing.¹⁴ In this approach, S_vO₂ is measured in the SSS and tCBF measured simultaneously in the internal carotid and vertebral arteries of the neck using a two-slice-interleaved multi-echo gradient-recalled-echo sequence. Four interleaves are required for each phase encoding, two to generate susceptometry weighted phase-difference maps and two to generate velocity-encoded phase maps, resulting in a temporal resolution of 25 seconds.

Pulse Sequence Modifications for Improved Temporal Resolution

Modification of the susceptometry-based CMRO₂ technique to achieve 3-second temporal resolution (Figure 1) was accomplished via two changes to the original approach:

1. Combining of sequence interleaves: rather than using two interleaves with different echo times to generate the S_vO₂ weighted phase-difference map, a multi-echo readout enables generation of a phase-difference map from data acquired in a single interleave. Any phase accrued owing to velocity encoding will equally affect both echoes of the multi-echo readout, as both echoes have the same polarity. Thus, the velocity and susceptometry interleaves can be combined. As a consequence of this modification, SSS blood flow (SSSBF) is quantified rather than tCBF from the neck arteries. However, tCBF can be accurately estimated by calibrating SSSBF based on the SSSBF:tCBF ratio measured at baseline with a two-slice-interleaved version of the sequence. This reference sequence (Figure 1A) is run immediately before starting the main (SSS-only) high temporal resolution CMRO₂ sequence (Figure 1B), which is continued for the remainder of the scan. This modification yields a two-fold temporal resolution increase.

2. Keyhole reconstruction with reduced phase-encoding lines:³² The number of phase-encode lines in the main (SSS-only) CMRO₂ sequence is reduced by a factor of 4 from 208 to 52, and the outer k-space is filled with data acquired from the same fully phase-encoded reference sequence used for calibrating SSSBF to tCBF (in Figures 1A and 1B, outer k-space data from echoes 1 and 2 is added to continuously update central k-space data from echoes 4 and 5). Unlike the main sequence, the reference sequence retains full-phase encoding to facilitate keyhole image reconstruction and allow higher fidelity quantification of the SSSBF:tCBF ratio. This modification yields a four-fold temporal resolution increase.

Figure 1.

High temporal resolution cerebral metabolic rate of oxygen consumption (CMRO₂) quantification pulse sequence: (A) a two-slice-interleaved fully phase-encoded reference sequence is run immediately before (B) the single-slice main sequence with 4x phase-encode reduction; (C) experimental schematic illustrating relative timing of the reference and main sequence; (D) sagittal maximum intensity projection indicating slice locations corresponding to the pulse sequence diagram with magnitude images highlighting (E) the superior sagittal sinus (SSS) in the head slice and (F) the internal carotid arteries (ICAs) and vertebral arteries (VAs) in the neck slice.

The resultant pulse sequence (Figures 1A and 1B) has the following parameters: FOV=208 × 208 mm² (head slice), 176 × 176 mm² (neck slice); voxel size=1 × 1 × 5 mm³ (head slice), 0.85 × 0.85 × 5 mm³ (neck slice); TR/TE1/ΔTE=28.85/5.5/7.04 milliseconds; bandwidth=521 Hz/pixel; flip angle=15 degrees; VENC=60 cm/second (head), 80 cm/second (neck); temporal resolution=12 seconds (reference sequence), 3 seconds (main sequence). The modifications described combine to provide an eightfold improvement in temporal resolution without reducing theoretical signal-to-noise ratio. In fact, because two phase-difference maps are simultaneously generated at every time point (one for each flow encoding) and subsequently averaged, signal-to-noise ratio should theoretically improve by approximately . The modifications described depend on several crucial assumptions that must be validated:

1. To determine tCBF from the SSSBF:tCBF ratio at baseline, the SSSBF:tCBF ratio must remain constant throughout the experiment. Because the SSS receives venous blood from most of the cortex, this assumption should be valid, especially during global physiologic challenges such as apnea or gas mixture breathing.

2. Keyhole reconstruction assumes that dynamic information is band-limited in k-space (i.e., image changes are low spatial-frequency processes). To satisfy this assumption, the diameter of any features of interest must be approximately greater than the keyhole reduction factor times the static resolution, or 4 mm for the 1-mm resolution and 4 × keyhole reduction factor used in the sequence described. The SSS is approximately 10 mm in diameter, and thus should fulfill this requirement.

3. Keyhole reconstruction assumes anatomic correspondence between the reference images and the main sequence images, and therefore requires that there be no movement over the course of the experiment. This is achievable at the level of the SSS, because it is easy to keep the head stationary in the MR scanner, even during a challenging paradigm such as volitional apnea.

In addition to improving temporal resolution, another motivation for velocity measurement in the SSS-only is that the neck vessels are more prone to movement, especially during physiologic paradigms such as apnea, violating assumption 3, and are also relatively smaller, violating assumption 2.

In Vivo Magnetic Resonance Imaging Studies

Human subject studies were approved by the Institutional Review Board of the University of Pennsylvania. Ten healthy volunteers (6 males, 4 females, ages 29±4 years) were recruited and participated after giving written informed consent. The subjects were judged to be healthy on the basis of their medical history. The particular population demographic was chosen to ensure maximal subject compliance to the physiologic paradigms. In all studies, images were acquired on a 3T Siemens Tim Trio system (Siemens Medical Solutions, Erlangen, Germany) using a 12-channel head coil and 2-channel neck coil. A vendor-provided gradient-recalled-echo axial localizer scan was used to select the location of the vessels of interest (SSS, internal carotid arteries, and vertebral arteries) and estimate θ, the tilt angle of the SSS with respect to B₀, from the coordinates of the centroid of the vessel for quantification of S_vO₂ as in equation (5).

Validation of Critical Methodological Assumptions

Three of the volunteers (2 males, 1 female, ages 25±1 years) completed a tube-breathing paradigm involving three minutes of normal breathing baseline, three minutes of breathing through 10 feet of plastic tubing with an attached mouthpiece to induce changes in flow and S_vO₂, and three minutes of normal breathing recovery. Tube breathing was chosen in this validation study because it induces a mixed hypercapnic/hypoxic state, similar in nature to breath-hold but sustainable over a long enough duration to acquire multiple data points at both slice locations with full-phase encoding.³³ The fully phase-encoded, two-slice-interleaved reference sequence was run during the entire paradigm, allowing quantification of the SSSBF:tCBF ratio over the course of the paradigm to test whether it remains constant during an apnea-like physiologic paradigm (assumption 1). Using full-phase encoding also allows comparison of SSSBF and S_vO₂ values obtained from retrospectively keyhole reconstructed data, where various amounts of outer k-space are replaced at each time point with the corresponding data from the first time point, as if only the central k-space had originally been acquired, as is the case when running the main sequence. This tests whether changes in parameter values are sufficiently band-limited in k-space to be accurately determined when using keyhole sampling and image reconstruction (assumption 2). Finally, because the paradigm requires both manipulation of the tube mouthpiece and significantly increased respiration, it challenges the subject's ability to remain static (assumption 3).

Quantification of Cerebral Metabolic Rate of Oxygen Consumption in Response to Apneic Challenge

Volunteers completed an apnea paradigm involving three repeated blocks of a 30-second normal breathing baseline period, a 30-second breath-hold apnea period, and a 90-second normal breathing recovery period. Before being scanned, subjects were instructed that all breath-holds should be completed at functional residual capacity, in other words, at normal end expiration. After running the reference sequence, the main sequence was run for the length of the 7.5-minute paradigm as in Figure 1C, allowing quantitation of S_vO₂ and tCBF at 3-second temporal resolution. S_aO₂ was measured continuously during the paradigm with a digital pulse oximeter placed on the right middle finger. Except in cases of abnormal cardiac anatomy, blood pumped to the brain and periphery originates from the same mixed pool in the left ventricle and therefore has the same S_aO₂. Thus, digital pulse oximetry will reflect cerebral S_aO₂. To correct for the known temporal delay in the measured S_aO₂ when using digital pulse oximetry, the S_aO₂ curve was shifted forward in time for each subject so that arterial resaturation occurs 7.5 seconds after cessation of breath-hold, matching the known circulatory transport delay between the lungs and brain³⁴ to within the temporal resolution of the MR pulse sequence (3 seconds). Breath-hold at normal end expiration was chosen to keep breath-holds as consistent as possible across repeats and subjects, ensuring that inspiration would occur immediately at the end of the breath-hold period. Subjects were verbally coached during the imaging experiment to ‘breath in', ‘breath out', and ‘stop breathing' 6, 3, and 0 seconds before the start of each apnea period, respectively, to ensure exact timing of the breath-holds. All subjects were able to successfully complete each of the breath-holds as confirmed by direct observation and pulse oximetry data. After the breath-hold paradigm, a T₁-weighted MP-RAGE image data set (voxel size=1 × 1 × 1 mm³) was acquired for normalization of tCBF to brain volume. Total brain volume was obtained using a semi-automated region-growing algorithm in ITK-SNAP.³⁵ After completion of the MR imaging experiment, each subject gave a venous blood sample, which was sent for complete blood count laboratory analysis to obtain a blood Hb and Hct value.

Data Processing

In all experiments, binary masks were generated for the carotid and vertebral arteries and SSS by thresholding of complex difference images, which robustly isolates the signal from flowing blood. S_vO₂ was quantified in the SSS from equation (5) with Δφ equal to the average phase difference between the reference tissue and the SSS regions of interest. Flow was quantified in the neck arteries and SSS by multiplying average velocity by cross-sectional area for each corresponding vessel regions of interest, summing over the four neck arteries to get tCBF.

In the tube-breathing experiments, images were retrospectively keyhole reconstructed at keyhole reduction factors of 2, 4, 8, and 16 by discarding all but the central 104, 52, 26, or 13 lines of k-space, respectively, and replacing outer k-space with corresponding data from the first image of the data set. In the apnea paradigm experiments, corresponding data acquired from the reference sequence was used to fill missing outer k-space data from the main sequence run for the duration of the paradigm.

All time-course data from the apnea experiments was averaged over the three repeated blocks of the paradigm to remove physiologic noise not related to the paradigm and improve signal-to-noise ratio. Average baseline parameter values were quantified from the first 24 seconds (eight data points) of the baseline period to exclude breathing effects from the coached inspiration and expiration during the final 6 seconds (two data points) of the baseline period. Data from only the final 15 seconds (five data points) of the apnea period were used to generate end-apnea parameter values to eliminate residual breathing effects and because physiologic changes due to apnea are not expected to occur instantaneously.

Results

Tube breathing produced a similar response across the three subjects. Time-course plots of S_vO₂, tCBF, and SSSBF in a representative subject (Figure 2A) demonstrate the expected increase in flow and S_vO₂ caused by hypercapnia that develops during the tube-breathing portion of the paradigm. Coefficients of variation of the SSSBF:tCBF ratio (assumed to remain constant for a given subject to allow tCBF estimation from SSSBF) across all time points (N=45) were 0.094, 0.075, and 0.084 for the three subjects. The SSSBF:tCBF ratio for each subject averaged across the normal breathing baseline and recovery (30 data points) and tube-breathing (15 data points) portions of the paradigm is shown in Figure 2B. Welch's t-tests for equal means between these two groups of SSSBF:tCBF values yields non-significant P-values (>0.40) for all subjects.

Figure 2.

Superior sagittal sinus blood flow (SSSBF):total cerebral blood flow (tCBF) ratio: (A) time-course plot of venous oxygen saturation (S_vO₂), tCBF, and SSSBF in response to 3 minutes of tube breathing (gray bar); (B) histograms grouped by subject showing the SSSBF:tCBF ratio averaged over the baseline/recovery (N=30) and tube-breathing (N=15) portions of the paradigm with error bars indicating±1 s.d.

Figure 3 demonstrates the effects of keyhole reconstruction on the accuracy of derived parameters as observed from the tube-breathing experiment. In Figure 3A, time-course plots of S_vO₂ and SSSBF derived from images with full-phase encoding are compared with the same plots generated from images retrospectively keyhole reconstructed with a range of keyhole reduction factors. Note the greater errors when larger keyhole reduction factors are used. In Figure 3B, these errors are plotted versus number of phase-encode lines used in the keyhole reconstruction. Values are averaged over the tube-breathing portion of the paradigm only, where errors should be greatest as observed in Figure 3A. For the keyhole reduction factor of 4 used in the CMRO₂ quantification sequence, mean error and root-mean-square error had magnitudes less than 0.04 for both S_vO₂ and SSSBF for all subjects.

Figure 3.

Error due to keyhole reconstruction: (A) Time-course plot of venous oxygen saturation (S_vO₂) and superior sagittal sinus blood flow (SSSBF) in response to tube breathing (gray bar) derived from images reconstructed with full-phase encoding (208 phase-encode lines) and with retrospective keyhole reconstruction using 104, 52, 26, or 13 phase-encode lines. (B) Mean error and root-mean-square (RMS) error in S_vO₂ and SSSBF during tube breathing using different numbers of phase-encode lines for retrospective keyhole reconstruction.

Figure 4 displays data from the apnea paradigm experiment in a typical subject. Changes in flow and S_vO₂ in response to apnea are visualized in the corresponding velocity and phase-difference maps (Figure 4A). Time-course plots of the measured parameters (Figures 4B and C) demonstrate an increase in S_vO₂ and tCBF and a decrease in S_aO₂ in response to apnea (gray bar). From these data, the arteriovenous oxygen saturation difference (AVO₂D, equal to S_aO₂–S_vO₂) is quantified and plotted alongside the tCBF, the product of which yields CMRO₂ (Figures 4D and E).

Figure 4.

Apnea paradigm representative subject data: (A) magnitude image with the superior sagittal sinus (SSS) outlined and corresponding velocity and phase-difference (Δφ) maps from specific time points (denoted by black symbols in B); (B) time-course plot of pulse oximetry measured arterial oxygen saturation (S_aO₂) and image-derived venous oxygen saturation (S_vO₂) and total cerebral blood flow (tCBF) absolute parameter values with black symbols corresponding to images in A; (C) percent changes in S_aO₂, S_vO₂, and tCBF parameter values normalized to average baseline value; (D) tCBF, arteriovenous oxygen saturation difference (AVO₂D), and cerebral metabolic rate of oxygen consumption (CMRO₂) absolute parameter values and (E) percent changes in parameter values normalized to baseline. Gray bars indicate the apnea period. All values in time-course plots are averaged across the three repeats of the paradigm. The bracketed sections ‘Base' and ‘EA' indicate the data used for computing average baseline values and end-apnea values for each subject.

Table 1 lists the parameters extracted from the time-course data for each subject, both at baseline and in response to apnea. The average baseline S_vO₂, tCBF, and CMRO₂ values were 68.6±3.0%HbO₂, 48.6±7.0 mL/100 g/minute, and 125.1±11.4 μmol/100 g per minute, respectively, consistent with previous findings.¹⁴ As previously observed,¹⁴ oxygen delivery (the product of C_aO₂ and tCBF) was negatively correlated with oxygen extraction (AVO₂D) at baseline (r=−0.76, P=0.011, two-tailed t-test). Maximum percent changes in tCBF and AVO₂D were 43.5±9.4% and −32.1±5.7%, respectively, resulting in a small (6.0±3.5%) but significant (P=0.00044, two-tailed t-test) increase in CMRO₂ between baseline and end apnea (final 15 seconds of apnea period).

Table 1. Summary of extracted physiologic parameters at rest and in response to volitional apnea.

	Subject
	1	2	3	4	5	6	7	8	9	10	Mean	s.d.
SvO2 (%HbO2)
Average baseline	66.0	69.0	68.2	72.3	62.7	67.3	72.5	71.0	67.2	69.5	68.6	3.0
Maximum	73.6	72.4	77.0	76.1	67.3	73.0	76.0	79.8	72.3	74.0	74.2	3.4
Change (%)	11.6	5.0	12.9	5.2	7.3	8.5	4.9	12.4	7.6	6.4	8.2	3.1
tCBF (mL/100 g per minute)
Average baseline	52.3	51.7	43.8	54.8	45.4	37.5	55.8	58.0	39.9	46.4	48.6	7.0
Maximum	83.6	70.1	62.0	82.9	62.8	52.1	73.9	91.4	55.8	64.8	69.9	12.9
Change (%)	60.0	35.6	41.3	51.1	38.2	38.9	32.5	57.6	40.0	39.7	43.5	9.4
SaO2 (%HbO2)
Average baseline	99.4	99.3	98.0	97.0	98.9	97.2	98.2	98.2	96.1	97.3	98.0	1.1
Minimum	92.7	94.7	96.3	90.7	92.3	94.0	93.3	96.3	91.7	93.2	93.5	1.9
Change (%)	−6.8	−4.6	−1.7	−6.5	−6.7	−3.3	−5.0	−1.9	−4.6	−4.2	−4.5	1.8
AVO2D (%HbO2)
Average baseline	33.4	30.2	29.8	24.7	36.2	29.9	25.7	27.1	28.9	27.8	29.4	3.4
Minimum	20.0	22.3	19.4	15.7	25.7	21.5	17.3	16.5	22.3	19.2	20.0	3.1
Change (%)	−40.0	−26.2	−35.0	−36.4	−29.1	−28.2	−32.8	−39.2	−22.8	−30.9	−32.1	5.7
CMRO2 (μmol/100 g/minute)
Average baseline	142.2	135.8	118.8	122.0	125.5	104.9	140.0	123.5	117.4	120.6	125.1	11.4
End apnea	147.1	155.8	122.6	130.3	132.5	107.2	147.4	130.1	126.9	126.9	132.7	14.1
Change (%)	3.5	14.7	3.3	6.9	5.6	2.2	5.3	5.3	8.1	5.2	6.0	3.5

Abbreviations: AVO₂D, arteriovenous oxygen saturation difference; CMRO₂, cerebral metabolic rate of oxygen consumption; %HbO₂, percent hemoglobin saturation; S_aO₂, arterial oxygen saturation; S_vO₂, venous oxygen saturation, tCBF, total cerebral blood flow.

Individual subject and group (mean and s.d.) values of various extracted parameters: average baseline (average value over the first 24 seconds of the baseline period), maximum/minimum (highest/lowest value reached over the entire paradigm), end apnea (CMRO₂ only: average value over the final 15 seconds of the apnea period); change (percent change between average baseline and corresponding maximum (S_vO₂, tCBF), minimum (S_aO₂, AVO₂D), or end-apnea (CMRO₂) parameter value.

The apneic response averaged across the 10 subjects is displayed in the time-course plots of Figure 5. It is evident that flow increases during apnea and then undershoots before returning to baseline while oxygen extraction decreases during apnea and then overshoots before returning to baseline. The slightly larger magnitude of the flow increase compared with the AVO₂D decrease causes a small increase in CMRO₂ during apnea followed by a transient undershoot before return to baseline.

Figure 5.

Apnea paradigm cohort data: (A) time-course plot of cohort-averaged arterial oxygen saturation (S_aO₂), venous oxygen saturation (S_vO₂), and total cerebral blood flow (tCBF) absolute parameter values and (B) percent changes in parameter values normalized to average baseline; (C) tCBF, arteriovenous oxygen saturation difference (AVO₂D), and cerebral metabolic rate of oxygen consumption (CMRO₂) absolute parameter values and (D) percent changes in parameter values normalized to average baseline. Error bars indicate±1 s.d. Gray bars indicate the apnea period. All values in time-course plots are averaged across the three repeated blocks of the paradigm. The bracketed sections ‘Base' and ‘EA' indicate the data used for computing average baseline values and end-apnea values.

Discussion

We have introduced a magnetic resonance imaging-based method for absolute quantification of CMRO₂ in humans with 3-second temporal resolution. Key methodological assumptions were validated in a tube-breathing paradigm. The sensitivity of the technique to detect dynamic changes in CBF, oxygen extraction, and CMRO₂ was assessed in response to a dynamic volitional apnea paradigm in a cohort of young healthy adults.

Results from the tube-breathing experiment (Figures 2 and 3) suggest application of the technique in the SSS produces only small systematic errors. Coefficients of variation of the SSSBF:tCBF ratio across all time points of the paradigm were small (<0.10) for each subject. Furthermore, Welch's t-tests for equal means comparing the SSSBF:tCBF ratio for tube-breathing and non-tube-breathing portions of the experiment were non-significant for all subjects (P>0.40). Therefore, it appears that SSSBF closely parallels tCBF in response to tube breathing. This result is expected considering that the SSS accounts for nearly half of tCBF and because a global physiologic paradigm such as tube breathing would not be expected to have a regional bias. Because of the similarity between tube breathing and apnea, the results support estimation of tCBF based on SSSBF for quantifying CMRO₂ in response to apnea or other global physiologic challenges.

Systematic errors due to keyhole reconstruction were observed to decrease expectedly as the keyhole reduction factor was decreased, with mean errors and root-mean-square errors in flow and S_vO₂ having magnitude less than 0.04 for all subjects and less than 0.02 averaged across subjects when using a reduction factor of 4. Importantly, because flow and S_vO₂ are not independent but both increase in response to apnea, absolute errors in S_vO₂ and flow are both negative, which should cause the resultant error in CMRO₂ to be lower than would be predicted if the two errors were not correlated.

In response to apnea, we observed a small (6.0±3.5%) but significant (P=0.00044, two-tailed t-test) increase in CMRO₂. Apnea has been used in the past as an assumed isometabolic stimulus in BOLD fMRI studies, both for calibrating the BOLD signal to quantify relative CMRO₂ changes¹⁹ and as an isometabolic standard in studies investigating the BOLD poststimulus undershoot.³⁶ Our results suggest that apnea may be slightly pro-metabolic. Increased CMRO₂ in response to apnea could represent a physiologic mechanism for buffering the brain energy supply in anticipation of prolonged apnea, which eventually would lead to exhaustion of energy stores and neuronal cell death. This would be consistent with other observations of the normal apneic response, such as reduced cardiac output, peripheral vasoconstriction, and cerebral vasodilation, which serve to maintain oxygen stores in the brain at the expense of the periphery.

To the authors' knowledge, no previous studies have directly quantified CMRO₂ during apnea, though studies of the CMRO₂ response to various gas mixtures provides insight into the present work. In recent work from the authors' laboratory, a similar CMRO₂ technique employing susceptometry-based oximetry has shown that CMRO₂ does not change during hypercapnic steady state.¹⁴ However, apnea represents a mix of both hypercapnia and hypoxia, and never reaches a steady state. Data from the periods between steady states were not recorded in the prior study, and the temporal resolution used (25 seconds) would be unable to distinguish the transient changes in CMRO₂ detected in the present study. Studies of the CMRO₂ response to hypercapnia using T₂-based methods for S_vO₂ quantification have yielded mixed results, with T₂-based intravascular approaches reporting both no change³⁷ and a 13.4±2.3% decrease³⁸ in CMRO₂ in response to moderate hypercapnia. The latter T₂-based approach was recently applied to detect a 5.0±2.0% average increase in CMRO₂ in response to mild (14% inspired FiO₂) steady-state hypoxia,³⁹ a difference of 18.4% compared with the CMRO₂ response to hypercapnia using the same methodology. Given that apnea is a mixed hypercapnic/hypoxic stimulus, that similar susceptometry-based oximetry techniques as the one used in the present study have found hypercapnia to be isometabolic,¹⁴ and that T₂-based approaches support a large CMRO₂ difference between hypoxia and hypercapnia, the small apneic CMRO₂ increase observed in this study is not unexpected. Nevertheless, extrapolations based on steady-state gas mixture breathing are of limited relevance to apnea, which is inherently non-steady state, involving continuously increasing levels of both hypercapnia and hypoxia.

Application of the proposed technique during administration of breathing gases (CO₂ and O₂) would better establish the relative contributions of hypercapnia and hypoxia to the observed apneic CMRO₂ response. Such studies would also suggest the extent to which hypercapnia and hypoxia are isometabolic, not only at steady state, but in the transition to steady state. Such information is critical given the use of CO₂ and O₂ in calibrating BOLD fMRI, as there is growing interest in making BOLD fMRI more quantitative through respiratory calibration.

The ability of the method to capture details of the temporal dynamics of the apneic response is especially well illustrated by the group-averaged time-course plots (Figure 5), which illustrate not only the neurovascular effects of apnea, but also the more subtle effects of respiration. Coached inspiration from 6 to 3 seconds before apnea causes reduced intrathoracic pressure and increased right atrial venous return, changes reflected by the small observed increase in CBF just before apnea initiation. In contrast, coached expiration during the final 3 seconds before apnea has the opposite effect, and the end-expiratory breath-hold state reached upon apnea initiation is also known to result in decreased cerebral venous blood flow.⁴⁰ Indeed, CBF was observed to transiently decrease during the beginning of the apnea period, taking nearly half the apnea period for the flow increase due to apnea-induced hypoxia and hypercapnia to overcome the small flow decrease caused by the end expiration-induced reduction in cerebral venous blood flow. Finally, immediately after apnea cessation, another transient, sharp increase in flow is observed, likely arising from the large initial inspiration at the end of the apnea period. While opposite magnitude changes in the AVO₂D difference were observed during the same aforementioned inspiratory and expiratory periods, they are of lesser magnitude, and, therefore, flow driven CMRO₂ changes are recorded. It is difficult to ascertain whether these transient CMRO₂ changes are real or arise from a temporary mismatch between the true (arterial) tCBF, and the SSSBF used to quantify tCBF in the described technique. Inspiration and expiration during other free-breathing portions of the paradigm is not temporally matched across subjects and paradigm repeats, and is therefore averaged out of the time-course data.

One possible limitation of the proposed method is the necessity of measuring S_aO₂ with pulse oximetry. No arterial vessels of suitable geometry for application of the long paramagnetic cylinder model exist in the head or neck region. Furthermore, the signal-to-noise ratio of susceptometry-based oximetry phase-difference maps is proportional to the accrued phase, which is small in highly oxygenated arterial blood. Accurate S_aO₂ quantification is critical as even a small underestimation in the S_aO₂ drop would cause an overestimation of the observed increase in CMRO₂ in response to apnea. However, if underestimation of the change in S_aO₂ was the cause of the observed apneic CMRO₂ increase, one would expect the percent changes in S_aO₂ and CMRO₂ in response to apnea to be positively correlated across subjects, however, this correlation was small, negative, and insignificant (r=−0.18, P=0.62, two-tailed t-test).

An alternative approach to fast CMRO₂ quantification is the use of projection-based T₂ measurement, which achieves S_vO₂ quantification in 15 seconds, is independent of vessel orientation, and is not sensitive to field inhomogeneities.⁸ However, the method assumes that flow remains constant over the course of each 15-second measurement. It is therefore not suitable for a paradigm, such as apnea, in which significant flow changes occur over seconds. Furthermore, the model used for determining percent hemoglobin saturation values from T₂ measurements must be empirically calibrated to specific Hct values, with errors due to deviations in Hct becoming especially large for higher blood oxygen saturation levels.

In conclusion, we have introduced and validated an approach for rapid quantification of CMRO₂ with 3-second temporal resolution, and applied it to characterize the CMRO₂ response to apnea. Potential clinical applications include investigation of diseases of altered neurometabolic response, for instance, OSA. More broadly, by providing a simple, robust, and quantitative method for assessing CMRO₂ in response to physiologic stimuli, the technique can be used to investigate neurometabolic–hemodynamic relationships in a variety of normal physiologic and pathologic conditions.

The authors declare no conflict of interest.