MRI of Cerebral Oxygenation: a quest to quantify the BOLD effect
Thomas Christen1
1INSERM, France

Synopsis

Keywords: Neuro: Brain, Image acquisition: Quantification, Image acquisition: MR Fingerprinting

In this presentation, we will follow a line of research that aims to estimate brain oxygen extraction fraction by quantifying the famous Blood Oxygen Level Dependent (BOLD) effect. We will see how the models and data acquisitions patterns have been refined over the years, and how these “quantitative BOLD” methods have been eventually fused with other MR approaches such as quantitative susceptibility mapping (QSM) or MR fingerprinting (MRF). Clinical and preclinical results will be presented in healthy brains as well as in pathologies such as stroke or neurodegenerative diseases.

Introduction

The brain represents 20% of the entire human body O2 consumption while only accounting for 2% of its mass. It has also a limited oxygen storage capacity. All these characteristics make the local oxygen extraction fraction (OEF) a key physiological parameter of the brain’s energy metabolism, and a potential biomarker in several diseases. In this presentation, we will follow a line of research that aims to estimate brain oxygen extraction fraction by quantifying the famous Blood Oxygen Level Dependent (BOLD) effect. We will see how the models and data acquisitions patterns have been refined over the years, and how these “quantitative BOLD” methods have been eventually fused with other MR approaches such as quantitative susceptibility mapping (QSM) or MR fingerprinting (MRF). Clinical and preclinical results will be presented in healthy brains as well as in pathologies such as stroke or neurodegenerative diseases.

The BOLD effect

MR is well known to be sensitive to brain blood oxygenation; the Blood Oxygen Level Dependent (BOLD) contrast has been used for many years in the field of cognitive neuroscience to study brain function at rest and during mental tasks. BOLD MR imaging is noninvasive and can be performed with high temporal and spatial resolution. The BOLD effect is rather simple: oxyhemoglobin is weakly diamagnetic and its presence in the blood vessels does not affect the MR signal. On the opposite, deoxyhemoglobin is strongly paramagnetic and induces magnetic perturbation around blood vessels. The spatial scale of these perturbations can extend up to five times farther than the vessel radii. In presence of dHb, 1H nuclei inside (but also outside) the blood vessels will thus experience different magnetic fields depending on their position and precess at different frequencies, rapidly losing phase coherence and resulting in a faster-decaying signal. The BOLD effect is classically observed as T2* changes in gradient echo experiments, but the diffusion of water molecules around vessels means that they also acquire a phase history that cannot be compensated by a spin echo: T2 weighted spin echo images are also affected.

The BOLD effect in the context of baseline blood oxygenation measurements

In theory, the BOLD effect should give access to OEF or to blood oxygen saturation (SO2) estimates. However, usual fMRI experiments only look at contrast variations between different brain states and even in this case, quantification of these variations is usually not performed. The reasons come from the complex dependence of the BOLD MR signal on blood oxygenation as well as several other parameters that affect the signal evolution. For example, the magnitude of the BOLD effect is affected by the total quantity of dHb in the imaged voxel, i.e. by the SO2, but also by the Blood Volume fraction and hematocrit. Vessel orientation in the magnetic field is also of importance. Furthermore, the signal is sensitive to any magnetic field inhomogeneity that does not originate from the presence of dHb (uncorrected main magnetic field spatial variations, susceptibility interfaces, metallic implants, calcifications, microbleeding, myelin fibers), the continuous renewal of blood in vessels that induces flow effect in the magnitude images, or the presence of tissue with high transverse relaxation T2.

Quantitative BOLD approaches over the years

Extracting baseline oxygenation information from BOLD experiments and disentangling the contribution of the different parameters has been the focus of MR techniques called quantitative BOLD (or qBOLD) approaches. These techniques have been initially proposed 25 years ago and have benefited over the years of a better understanding of brain physiology, MR signal modeling, faster multi echoes acquisitions and higher magnetic fields. The qBOLD approaches rely on mathematical models that describe the MR signal evolution of gradient echo and spin echo sequences (sampled at multiple TEs) in presence of inhomogeneous tissues including multiple blood vessels. A commonly used model was described in Yablonskiy et al. 1994 and considered a two compartments model where vessels are described as infinite straight cylinders without preferential orientation. Neglecting water diffusion (static-dephasing regime) and considering enough blood vessels in the voxel to make statistical assumptions, the mathematical expression of the field perturbations was derived as well as the temporal evolution of the MR signal. The model was later refined by He et al. 2007 by including different tissue contributions (gray matter, cerebrospinal fluid, and pure blood) and other research groups have refined the model over the years. In practice, the qBOLD method works with high signal-to-noise ratios (SNR) acquisitions, and it is difficult to obtain accurate SO2 maps and blood volume maps in vivo by adjusting all the model’s parameters at once. In order to simplify the problem and leave SO2 as the sole unknown parameter in the fitting step, Christen et al. 2011 proposed to independently measure the confounding parameters of the qBOLD model with dedicated MR scans. It is also important to note that the qBOLD models generally do not account for non-blood tissue susceptibility and it has been noted in multiple qBOLD studies in humans that the SO2 estimates appear lower in the white matter regions where myelin fibers are present.

When qBOLD meets quantitative susceptibility mapping (QSM)

In order to correct for erroneous estimates in the white matter, qBOLD methods (based on magnitude MR data) have been recently fused with the quantitative susceptibility mapping (QSM) technique (based on MR phase data). The QSM technique has been proposed 13 years ago to derive magnetic susceptibility (χ) maps from MR phase maps (Rochefort et al. 2010). QSM can be used to study iron content in the brain and can be linked to blood oxygenation in large vessels without the need to know the vessel’s orientations. It has also been used by Zhang et al. 2015 to derive local microvascular SO2. Estimating the venous blood fraction BVfv is however challenging with this approach, and is currently based on an empirical linear model, which does not suit all tissue types, in particular in pathological cases. One can notice that the qBOLD and QSM approaches have complementary strengths and weaknesses. In consequence, Cho et al. 2018 recently proposed a combined method, coined QQ (QSM + qBOLD), to improve SO2 estimates. The proposed model has 5 unknown parameters: BVfv, SO2, non-blood susceptibility, transverse relaxation rate R2 and S0, the initial amplitude of the GRE signal. This model is fitted to both the magnitude and phase data of the acquired GRE data. The model has been further improved, helped with deep learning reconstruction and results have been recently compared to PET imaging in human brains.

qBOLD in the MR fingerprinting (MRF) framework

One current limitation of the qBOLD approaches is the use of analytical models that make strong assumptions on physical properties of the vascular network or physiological processes. In particular, it is assumed that the networks don’t have preferential orientations, have a perfect cylindrical shape, etc… or that the water diffusion process can be neglected. This can be problematic in pathological environments where the vascular networks are disturbed. These models also tend to limit the type of acquisitions that can be used. A powerful alternative to analytical modeling has been proposed by (Ma et al., 2013) in 2013 and called Magnetic Resonance Fingerprinting. In MRF, a fast MR sequence known for its sensitivity to relaxation times is used to acquire data in vivo. The acquisition is repeated with a pseudo random choice of parameters such as repetition time, flip angle, and inversion time, resulting in a different signal evolution in every voxel. This ‘fingerprint’ is then matched to a dictionary of curves obtained using numerical simulations of the same experiment and is translated into quantitative maps. This concept can also be used to analyze the natural temporal evolutions of MR signals and retrieve quantitative information about the microvascular network at the sub-voxel imaging scale. For example, samples of the Free Induction Decay and Spin Echo can be considered as fingerprints and simulations can be performed using virtual voxels containing geometric structures (blood vessels, cells, etc.) (Christen et al., 2014). Here magnetic field computation at different orientations as well as water diffusion effects can be easily incorporated. With pattern matching algorithms, it is then possible to derived quantitative maps of blood volume, vessel radius and blood oxygenation with high spatial resolution and values consistent with other MR techniques and literature reports. Current studies are investigating the use of realistic vascular networks from high resolution microscopy or faster acquisition schemes.

Acknowledgements

No acknowledgement found.

References

Yablonskiy, D. A. and E. M. Haacke, Theory of NMR signal behavior in magnetically inhomogeneous tissues: The static dephasing regime, Magnetic Resonance in Medicine 32, 749 (1994).

He, X. and D. A. Yablonskiy, Quantitative BOLD: Mapping of human cerebral deoxygenated blood volume and oxygen extraction fraction: Default state, Magnetic Resonance in Medicine 57, 115 (2007).

Christen, T., B. Lemasson, N. Pannetier, R. Farion, C. Segebarth, C. Rémy and E. L. Barbier, Evaluation of a quantitative blood oxygenation level-dependent (qBOLD) approach to map local blood oxygen saturation, NMR in Biomedicine 24, 393 (2011).

Rochefort, L. de, T. Liu, B. Kressler, J. Liu, P. Spincemaille, V. Lebon, J. Wu and Y. Wang, Quantitative susceptibility map reconstruction from MR phase data using Bayesian regularization: Validation and application to brain imaging, Magnetic Resonance in Medicine 63, 194 (2010).

Zhang, J., T. Liu, A. Gupta, P. Spincemaille, T. D. Nguyen and Y. Wang, Quantitative mapping of cerebral metabolic rate of oxygen (CMRO2) using quantitative susceptibility mapping (QSM), Magnetic Resonance in Medicine 74, 945 (2015).

Cho, J., Y. Kee, P. Spincemaille, T. D. Nguyen, J. Zhang, A. Gupta, S. Zhang and Y. Wang, Cerebral metabolic rate of oxygen (CMRO2 ) mapping by combining quantitative susceptibility mapping (QSM) and quantitative blood oxygenation level-dependent imaging (qBOLD), Magnetic Resonance in Medicine 80, 1595 (2018).

Ma, D., V. Gulani, N. Seiberlich, K. Liu, J. L. Sunshine, J. L. Duerk and M. A. Griswold, Magnetic resonance fingerprinting, Nature 495, 187 (2013).

Christen, T., N. A. Pannetier, W. W. Ni, D. Qiu, M. E. Moseley, N. Schuff and G. Zaharchuk, MR vascular fingerprinting: A new approach to compute cerebral blood volume, mean vessel radius, and oxygenation maps in the human brain, NeuroImage 89, 262 (2014).

Proc. Intl. Soc. Mag. Reson. Med. 31 (2023)