Synopsis

Keywords: Neuro: Brain, Contrast mechanisms: Diffusion, Cross-organ: Inflammation

Diffusion MRI holds great potential to capture inflammatory processes in vivo and non-invasively, as the diffusion of water molecules in the brain is highly sensitive to changes in the microstructure resulting from glial proliferation, cytotoxic or vasogenic edema, demyelination and other cellular processes. Here we describe how various neuroinflammatory processes are expected to affect metrics derived from the diffusion and kurtosis tensors (sensitivity), with examples from both preclinical animal models and clinical human studies. We briefly discuss how biophysical models could disentangle between individual pathological processes (specificity), but emphasize the need for further development and validation of these methods.

Outline

Diffusion MRI holds great potential to capture inflammatory processes in vivo and non-invasively, as the diffusion of water molecules in the brain is highly sensitive to changes in the microstructure resulting from glial proliferation, cytotoxic or vasogenic edema, demyelination and other cellular processes. There are two broad strategies for relating diffusion-weighted MRI signals to microstructure: using either signal representations (e.g. diffusion and kurtosis tensors) or biophysical models. In this lecture, we describe how various neuroinflammatory processes are expected to affect metrics derived from the diffusion and kurtosis tensors (sensitivity). We present examples from the literature in both preclinical animal models - often with histological validation - and clinical human studies. Finally, we briefly discuss how biophysical models could be exploited to disentangle between individual pathological processes (specificity), but emphasize the need for further development and validation of these methods.

Sensitivity of DTI and DKI to inflammatory processes

In signal representations, the MRI signal is fit by a mathematical model that captures its features (e.g. the decay as a function of b-value), without making any assumption about the underlying tissue or microstructure. This constitutes both their strength and weakness: they are applicable to any tissue type, but they do not provide metrics for specific features of microstructure. The most popular signal representation is Diffusion Tensor Imaging (DTI). Diffusion Kurtosis Imaging (DKI) [1] extends conventional DTI by capturing the non-Gaussian diffusion components apparent at moderate diffusion weighting (b~2000 s/mm2), a hallmark of tissue heterogeneity (e.g. multiple compartments).
DTI and DKI metrics are very sensitive to a variety of cellular processes, with the sign of their change (increased or decreased) depending on the specific cellular pathological process. Early inflammatory response associated with microglial proliferation and astrocytosis can produce an isotropic decrease in apparent diffusivity and increase in kurtosis due to the increased cellularity and tissue heterogeneity. Other concurrent neurodegenerative processes, such as axonal swelling or beading in the white matter may also decrease the apparent diffusion and increase the apparent kurtosis, but predominantly in the axial direction along the axons, while cytotoxic edema in the gray matter may isotropically decrease the diffusivity and increase the kurtosis. On the other hand, vasogenic edema, as observed on T2-weighted MRI, may overall increase the diffusivity but to a lesser extent kurtosis as the free water component is mainly observed at low b-values, as probed by DTI. In addition, several neurodegenerative processes, such as demyelination and neuronal loss, have been associated with an increase in apparent diffusivity and decrease in kurtosis.
These cellular processes often happen at the same time, or their exact timeline is unclear, both in acute neurological events, as well as in chronic neuroinflammatory and neurodegenerative diseases. Dedicated longitudinal animal studies along with histological validation may provide some answers as to which cellular processes affect the microstructure to a dMRI-measurable extent, or which process dominates the detected change in DTI and/or DKI metrics.
Important validation work has been carried out in the Low-dose lipopolysaccharide model [2-3], the cuprizone model [4-7], traumatic brain injury [8-9], experimental autoimmune encephalitis (EAE) [10-12] and stroke [13-16].
In contrast to animal studies, human studies typically do not allow for direct validation of diffusion metrics with respect to their sensitivity or specificity to different neuroinflammatory processes. However, human studies of neuroinflammatory disorders may bring some insight into DTI’s and DKI’s sensitivity to neuroinflammation, either by comparison with other more specific modalities such as PET, or with accompanying animal validation studies. Longitudinal studies are also crucial as they may reveal biphasic diffusion changes with early phases dominated by neuroinflammation and later stages by neurodegeneration. Examples include studies of multiple sclerosis [17-19], stroke [20], alcohol exposure [21], obesity [22], amyloid (A-beta) deposition [23] and COVID-19 [24].

Specificity of diffusion MRI to cellular inflammatory processes

DTI or DKI metrics have limited specificity in terms of identifying distinct pathological mechanisms. Depending on the degrees of freedom for the model parameters, biophysical models can in principle provide more specific insight into microstructural changes than signal representations, for instance as estimates of axonal water fraction, diffusivities inside distinct compartments (e.g. intra-axonal, extracellular…), soma sizes, etc. Among the main limitations of biophysical models however are the challenging parameter estimation and the need for extensive validation of the estimated parameters before a claim for specificity can be made. These efforts are still ongoing and beyond the scope of this lecture. Here we provide however a brief overview of how different biophysical models have been used so far to characterize brain inflammatory responses, whether in animal models (cuprizone [4-7], EAE [25-27], Wallerian degeneration [28]) or in patients (multiple sclerosis [29-33], Alzheimer’s disease [23, 34-36], stroke [14], schizophrenia [37-38]).

Acknowledgements

I would like to thank Els Fieremans, with whom I have co-authored a book chapter on the very topic of diffusion MRI in neuroinflammation [39].