Perfusion measurements can be obtained by modeling the dynamic passage of a bolus through the cerebral vasculature. Traditionally, this is done by using an injected contrast agent and rapid imaging, then modeling the time courses of the concentration functions.¹ A system response is determined by deconvolution, which is performed between the time courses in the tissue and a time course selected as an input (arterial input function, AIF). In this work, instead of an injected contrast agent, we use a block hyperoxia stimulus. The extra oxygen arrives in the arteries dissolved in the plasma, which is used first for metabolic demand, leaving more oxygen bound to the hemoglobin, thus reducing the concentration of deoxygenated hemoglobin on the venous side and increasing T₂*-weighted signals. We hypothesized that modeling the dynamic concentration of bound venous oxygen through the brain using blood oxygenation level dependent (BOLD) imaging could give us physiologically accurate perfusion values of cerebral blood flow (CBF), the venous component of the cerebral blood volume (CBV_V) and mean transit time through the venous component (MTT_V).

Imaging was performed on a 3T MR scanner (Discovery 750, GE Healthcare) equipped with a 12-channel neurovascular coil. Five healthy participants were recruited for this IRB approved study. A respiratory circuit was used with a non-rebreathing facemask to manipulate the oxygen content of air inhaled by the subjects during imaging. The end-tidal partial pressures of oxygen (P_ETO₂) and carbon dioxide were measured during imaging (BIOPAC Inc.). BOLD imaging was performed to rapidly image the relative increase in bound venous oxygen. The sequence had parameters of TR/TE/α of 2000 ms/30 ms/80° and an acquisition matrix of 64×64×43 over a field of view of 224 mm x 224 mm x 150.5 mm. This provided whole brain coverage with an isotropic spatial resolution of 3.5 mm. Hyperoxia was achieved with 70% oxygen, balance medical air. Several block designs were tested: the first subject was stimulated with 3-2-1 minute blocks of normoxia-hyperoxia-normoxia, the next three subjects had two stimulus blocks of two minutes of hyperoxia (3-2-3-2-3 minute sequence), and the final subject had a single longer stimulus block (3-5-5 minutes). The P_ETO₂ values were used as a surrogate measurement of the concentration of oxygen in the arterial blood ($$$PaO_2$$$), $$CaO_2(t)=\phi[Hb]SaO_2+\epsilon PaO_2$$where $$$\phi$$$, $$$[Hb]$$$, and $$$\epsilon$$$ are 1.34 mlO₂g_Hb^-1, 15 g_HbdL_Blood^-1, and 0.0031 mlO₂dL_Blood^-1mmHg^-1, respectively. The Severinghaus equation was used to find saturated oxygen ($$$SaO_2$$$) from $$$PaO_2$$$.2 BOLD image preprocessing steps included motion correction, spatial smoothing with a 5-mm kernel, and drift correction. P_ETO₂ values and BOLD signals were independently fit with the model shown in Figure 1, where the increased oxygen concentration is modeled with a single-exponential response. The concentration of bound oxygen in the venous blood was found using the Davis equation.³ The CBV_V measurement was found using a method described by Blockley et al.,⁴ and the CBF was determined by deconvolving the fitted tissue concentration function from the AIF. MTT_V was found as MTT_V=CBV_V/CBF. Gray matter (GM) and white matter (WM) ROIs were selected by registering the ICBM atlas to the subjects and excluding voxels with poor quality of fit (R²<0.4).

Figure 2 shows example fits to the end-tidal oxygen and a representative BOLD signal from GM. Figure 3 contains several maps obtained from the subject with the 3-5-5 block design. R-squared maps provide the quality of the fit of the BOLD signals to the model in Figure 1. Good contrast between GM and WM was observed. Measurements of CBV_V, CBF, and MTT_V in gray matter were found to be 1.83 ± 0.51 ml 100 g^-1, 58.92 ± 4.16 ml 100 g^-1min^-1, and 5.17 ± 1.04 s averaged over the five subjects, and in white matter they were found to be 1.31 ± 0.34 ml 100 g^-1, 23.62 ± 2.97 ml 100 g^-1min^-1, and 9.22 ± 2.00 s.The venous component of CBV is typically found to be between 40% and 50% of the total CBV, which is in the range of 2-6 ml 100 g^-1, so the values estimated here are within the range expected.^1,5,6 The effect of hyperoxia on BOLD signal is much smaller than that created by a contrast agent; hyperoxia causes signal changes of 1-3% while contrast agents can create a signal attenuation of 70% or more in dynamic susceptibility contrast MRI. Nonetheless, estimating CBV from hyperoxia has been presented previously.⁴ The novelty of this work is the calculation of CBF from the dynamic oxygen passage. To conclude, we were able to obtain physiologically reasonable perfusion measurements using a dynamic oxygen passage experiment.