The brain has an impressively high metabolic demand, receiving 15% of blood flow from cardiac output and consuming 20% of total oxygen used by the body under normal conditions (Figure 1) ^{1, 2}. To meet this metabolic demand, the brain closely autoregulates its blood supply and the amount of oxygen extracted by cerebral tissues from the blood.

The brain’s oxygen consumption can be characterized by several physiological parameters (Table 1). The total rate of oxygen consumption is proportion to cerebral blood flow (CBF) and the percent of oxygenation that has been extracted by cerebral tissues. In typical conditions, arterial blood is fully oxygenated (SaO₂ = 98%). As the blood travels into the microvasculature, oxygen molecules diffuse into brain tissue surrounding the arterioles and capillary beds, and oxygen saturation decreases. By the time blood reaches veins of size that are detectable on MRI, nearly all of the oxygen extraction has occurred. For this reason, measurements of blood oxygen saturation in veins (SvO₂) gives an indication of total oxygen extracted by upstream brain tissues that drain into the vessel (Figure 1).

Because the brain is particularly sensitive to increased or decreased blood flow, measurements of cerebral physiology are key biomarkers of brain tissue health. Impaired physiology occurs in many brain disorders, such as stroke ³, traumatic brain injury ⁴, and tumors ⁵ and may even represent early signs of neurodegeneration ⁶. The ability to noninvasively image regional oxygenation levels in the brain could provide valuable information to choose the right therapy for patients. For instance, in ischemic stroke, absolute measurements of regional oxygenation can identify the presence of viable tissue to determine whether the patient is good candidate for reperfusion therapy ⁷.

The measurement of oxygen saturation in the brain is technically challenging. To date, established methods to measure brain oxygenation have relied on positron emission tomography (PET) with [¹⁵O] radiotracers. However, [¹⁵O] PET is not clinically used because it requires injection of short-lifetime radiotracers, invasive arterial sampling, and specialized equipment that is not widely available in hospitals. As an alternative, MRI is also sensitive to oxygenation levels in the brain, and new MRI methods to image oxygenation have shown great promise.

Ideally, an MRI approach to quantify brain oxygenation has good spatiotemporal resolution, is robust to a range of oxygen saturation levels, and is easy to implement in the clinic. MRI is sensitive to different oxygenation levels because as more oxygen is extracted, the concentration of deoxyhemoglobin (dHb) in venous blood increases. The dHb molecules are paramagnetic, and thus influence magnitude signal intensity, relaxation parameters, and magnetic susceptibility detected on MRI. Current MRI approaches to image oxygenation take advantage of these different contrast mechanisms, and thus have different strengths and limitations (Table 2). These approaches have been nicely reviewed by Christen et al ⁸.

Theory: The blood oxygenation level dependent (BOLD) signal takes advantage of the extravascular effects of dHb to provide oxygenation contrast. For instance, a lower SvO₂ corresponds to a higher concentration of paramagnetic dHb molecules in the brain. These dHb molecules create local field inhomogeneities in the extravascular tissue surrounding the vessels, resulting in magnitude signal loss and decreased signal relaxation times (T2*, T2, T2’). You may be familiar with dHb-induced T2* and T2 signal changes during brain activation from functional MRI studies.

While several MRI relaxation times are sensitive to oxygenation level, studies suggest that the T2’ parameter is the most directly related to oxygenation ⁹. T2’ is the reversible component of transverse relaxation, and is defined as 1/T2’ = 1/T2* - 1/T2. New hybrid sequences that combine a gradient and spin echo (multi-echo) acquisitions allow estimation of T2*, T2, and T2’ from the same scan ¹⁰. These hybrid sequences enable mapping of relaxation parameters that are sensitive to the underlying oxygenation state of the brain.

Challenge: A major challenge of extravascular BOLD methods is that relaxation parameters are not specific to brain oxygenation. Even T2’ is the product of blood volume and dHb-induced frequency shifts. As a result, complex biophysical models are often required to interpret the BOLD signal in terms of oxygenation. Otherwise, multiple acquisitions are necessary at different brain physiological states (e.g. after breathing different gases) to tease apart different contributions to the BOLD signal.

Successes: Early quantitative BOLD (qBOLD) approaches have focused on the T2’ signal from gradient- and spin-echo acquisitions. These methods model capillary vessels in brain parenchyma as a network of randomly oriented cylinders to describe MRI signal dephasing in the presence of dHb. By fitting the signal model at each voxel, qBOLD techniques create parametric maps of SvO₂ and CMRO₂. Some qBOLD implementations assume a single extravascular tissue compartment ^{11, 12}, while others also consider blood and CSF compartments in the model fit to each voxel ^13-15.

Alternatively, respiratory calibration MRI uses gas challenges to measure BOLD at different physiological states of the brain and ultimately quantify resting tissue oxygenation. Respiratory calibration MRI collects BOLD not just at baseline, but during multiple gas breathing tasks. The BOLD, perfusion, and end-tidal O₂ (the amount of expired O₂) signal for each of these gas manipulations is modeled with the generalized calibration model ¹⁶, and provides quantitative maps of SvO₂. Several variants of this respiratory calibration approach have been implemented with pure hyperoxia (increased O₂) and hypercapnia (increased CO₂) ^{17, 18}; or gases with different combinations of O₂ and CO₂ concentrations ^{19, 20}.

In the future, BOLD methods for oxygenation imaging may synergize well with novel fingerprinting approaches. Christen et al. proposed a vascular fingerprint (from gradient- and spin- echo hybrid scans) for various cerebral blood volume, mean vessel radius, and blood oxygen saturation (SvO₂) ²¹. The measured signal curve for each voxel is then matched to a dictionary curve, which reveals a specific quantitative SvO₂ (%) for tissue in the voxel. The accuracy of vascular fingerprinting depends on whether the biophysical model for these signal curves accurately represents cerebral physiology.

Theory: Instead of looking at signal in extravascular brain tissues, intravascular MRI approaches seek to quantify T2 relaxation directly in venous blood. If more oxygen is extracted, more dHb molecules are present in the venous blood, leading to lower T2 values. Once the blood T2 relaxation is measured, a biophysical model allows us to convert venous blood T2 to quantitative SvO₂ (%), if hematocrit is also known ²².

Challenge: The main challenge is to isolate pure venous blood signal for T2 measurement, because most brain voxels represent a mixture of CSF, tissue and blood signal. Many of the first intravascular oxygenation studies chose to focus on large veins with voxels that contain only pure venous blood ^{23, 24}.

Successes: The most commonly adopted intravascular approach is T2-Relaxation Under Spin Tagging (TRUST), which measures T2 in the sagittal sinus to assess global SvO₂ ²⁵. TRUST MRI applies spin labeling pulses to collect images with and without labeling of venous blood at different echo times. In this manner, signal contributions from CSF and static tissue can be subtracted out, and the T2 measurement is made only for venous blood in the sagittal sinus. TRUST is fast and gives absolute, global SvO₂ (%) values in minutes that have been calibrated in different physiological conditions ²⁶. Recent efforts have used velocity-encoding gradients to target blood from smaller veins for T2 (and oxygenation) measurements that are more representative of local brain function ²⁷.

T2 methods have also been extended to map oxygenation in brain tissues (i.e., from the microvasculature in each voxel). QUantitative Imaging of eXtraction of Oxygen and TIssue Consumption (QUIXOTIC) MRI uses velocity-selective radiofrequency pulses to select for venular blood ²⁸. These pulses use known cutoff velocities of blood as it passes through the microvasculature to create maps of only venular blood for T2 and SvO₂ measurement. The main limitation of QUIXOTIC is low signal to noise ratio (SNR), because typical tissue voxels only have 5% blood volume. Improvements to this oxygenation mapping technique have been proposed to remove contamination from diffusion and increase SNR of the oxygenation measurements ²⁹.

Theory: A third MRI contrast mechanism for oxygenation derives from dHb-induced increases in magnetic susceptibility within veins compared to the surrounding brain tissue. This susceptibility shift creates magnetic field perturbations that manifest on MRI phase images. In this way, MRI phase images provide information about susceptibility changes that enable quantification of the underlying SvO₂ in individual vessels.

Challenges: Although magnetic susceptibility is linearly related to OEF, there is no direct way to image susceptibility by MRI. The relationship between magnetic field and MRI phase with the underlying susceptibility depends on the vein orientation and geometry in a complex and nonlocal manner. For this reason, susceptibility measurement is nontrivial and requires solution of a difficult mathematical inversion problem. Furthermore, sufficient spatial resolution must be achieved to measure phase within smaller veins, which lengthens the MRI scan time.

Successes: Susceptibility-based studies of oxygenation have been reviewed by Wehrli et al ³⁰. The first phase MRI studies to image oxygenation approximated cerebral veins as long cylinders parallel to the main magnetic field. For such a parallel vein geometry, there is a simple relationship between measured phase in the vein and its susceptibility. This approach, MRI susceptometry, has been used to study oxygenation in large draining veins such as the internal jugular vein ³¹ and sagittal sinus of the brain ³². Similar to TRUST MRI, global SvO₂ measurements from MRI phase are fast and reproducible. These fast susceptometry methods can be combined with whole-brain flow in the same sequence to study functional physiological changes ^{33, 34}.

Recent studies have also sought to assess phase-based oxygenation in smaller veins ³⁵, which is expected to be more reflective of local brain physiology. For these smaller vessels, it will be particularly important to correct for partial volume effects ³⁶ and potential orientation effects if the vessel is tilted relative to the main field ³⁷.

Going forward, SvO₂ may be available in veins of arbitrary curvature and orientation if quantitative susceptibility mapping (QSM) can directly reconstruct the 3D susceptibility distribution from measured field maps. Once the QSM map is reconstructed, susceptibility differences can be converted to SvO₂ along all resolved cerebral vessels, created a brain oxygenation venogram ³⁸. To ensure accurate and robust, more work needs to be done to understand potential OEF underestimation from the QSM reconstruction process and from second order effects of flowing spins in the vessels ³⁹.

Because MRI methods to image oxygenation are fairly new, few studies have investigated the reproducibility of these measures. The MRI technique that has been most broadly tested is TRUST MRI. Global SvO₂ was compared in six different sites on 250 healthy volunteers, with low standard error of SvO₂ of only 1.3% across sites ⁴⁰.

Each of the three MRI classes for oxygenation imaging reported SvO₂ values in the range of 50 – 75%, which is consistent with the physiological range expected from PET (Figure 3). However, only one study to date has compared MRI oxygenation to [¹⁵O] PET measurements in the same volunteers ⁴¹. This study observed decent correlation in OEF ratios of symptomatic to healthy brain by each modality in patients with carotid occlusions. Future comparison studies can leverage simultaneous PET/MRI hardware to compare concurrent measurements of brain oxygenation by PET and MRI.

Direct comparisons between MRI methods have often found discrepancies between oxygenation imaged in the same scan session. For example, although they strongly correlated, baseline OEF by susceptibility mapping was lower than baseline OEF by respiratory calibrated BOLD ⁴². A separate study implemented an interleaved scan to obtain global SvO₂ values both by T2 and susceptibility from the same sequence. This work found that baseline SvO₂ was lower for susceptometry relative to T2, but increased more in response to hypercapnia, despite acquiring data from the scan ⁴³.

To improve the robustness of new MRI methods, detailed analysis of the underlying measurement is necessary. Ni et al. showed that T2’ maps are significantly influenced by the imaging and analysis method, and should be considered when interpreting T2’ studies in terms of oxygen metabolism ⁴⁴. Blockley et al. simulated the OEF error due to inter-individual variations in physiology and non-ideal gas challenges ⁴⁵. These simulations led to optimization of the model for OEF mapping by respiratory calibration ⁴⁶. Thus, error analyses can provide valuable technical information toward a consensus oxygenation imaging and analysis protocol.

For oxygenation MRI to be clinically useful, its acquisition must be easily implementable with relatively short scan time. BOLD acquisitions have gained increasing usage in the clinic, in part because of the popularity of resting-state functional scans. However, because BOLD maps are a complex combination of cerebral physiology, BOLD studies in patients with ischemia ^{47, 48 49} and tumor ⁵ have required scans in different gas states to tease out oxygenation information. Quantitative respiratory-calibration methods also require multiple gas inhalations and would benefit from shorter protocols.

Due to its short acquisition time (~30 sec), global OEF assessment by TRUST has been applied in the sagittal sinus of many patient populations. These cohorts include neonates ⁵⁰, volunteers of different ages ⁵¹, and patients with neurodegenerative diseases such as multiple sclerosis ⁵². Similar whole-brain OEF measurements by MRI susceptometry have been shown obstructive sleep apnea with dynamic OEF imaging during a breath-hold task ⁵³. Phase susceptometry also showed that neonates with congenital heart disease have impaired brain physiology similar to premature infants that can predict eventual white matter damage ⁵⁴.

While global measurements are faster and have been shown to be fairly reliable, ultimately regional OEF information is necessary to assess many brain disorders. Susceptibility contrast is easy to obtain from a gradient echo MRI scan and can provide local oxygenation information within individual veins. Susceptibility weighted imaging has gained popularity to image oxygen disturbance in the affected versus healthy hemispheres of stroke patients ^{55, 56}, in traumatic brain injury ⁵⁷, and in multiple sclerosis ⁵⁸. At the same time, obtaining local susceptibility-based OEF information may be manually tasking, or suffer in accuracy from poor image reconstructions due to nearby susceptibility artifacts (e.g. from hemorrhagic blood products).

Importantly, these early patient studies with oxygenation MRI show promising physiological findings, and point to future technical developments to bring the new methods to clinical practice.