- Interpretation of diffusion MRI with respect to the true underlying tissue microstructure remains a challenge.

- Validating diffusion MRI models and findings with quantitative metrics is paramount to drawing strong conclusions.

- Histological staining and other optical microscopy methods are emerging as powerful tools for validation.

- Validation in injury and disease is a significant challenge due to the co-existence of multiple pathologies.

Scientists and clinicians interested in understanding the link between diffusion MRI findings and true tissue microstructures in the normal and pathological states.The audience will develop a better understanding of the relationship between diffusion MRI and tissue microstructure. This will enable the development of better model or acquisition methods to refine diffusion MRI, as well as more accurate insight into injury and disease states.

Diffusion MRI captures the molecular motion of water molecules within their biological environment. It is unique in its ability to derive noninvasively microstructural tissue information at a macroscopic scale. However, since structure is inferred from measurements of diffusion, validating DWI findings with other modalities is important for a complete understanding of diffusion MRI and its relationship to the true underlying tissue microstructure.

Whereas diffusion tractography typically relies on injectable tracers as the “gold-standard” for measures of brain connectivity, no single standard has emerged with which to validate local tissue microstructure. However, many validation approaches have been developed and used, and the approach generally is determined by the desired goal, which can be loosely classified into measures to quantify local fiber orientation, physical tissues metrics, and pathological alterations. An important feature of any validation technique includes derivation of quantitative and objective measures, but automation and ease-of-use are also desirable.

Local Fiber Orientation Distributions

For diffusion tractrography or connectivity applications, models of the diffusion MRI signal attempt to resolve the underlying fiber distributions based on the measured diffusion MRI signal behavior. Consequently, measures of the true fiber orientations can be derived from other modalities and compared to their diffusion MRI counterparts. Typically, these include histological sectioning and subsequent analysis of digital microscopy images using many different methods to quantify the orientations, including manual tracing (Leergaard et al., 2010), automated analysis of stained cells (Bock et al., 2010 and Jespersen et al., 2012), spatially registering histological and DTI images (Flint et al., 2010 and Hansen et al., 2011), and texture-based methods (Budde et al., 2011 and Choe et al., 2012) and others. Thick sections stained with a lipophilic dye and imaged with 3D microscopy have recently been shown to be useful in relating directly to diffusion MRI fiber distributions (Khan et al. 2015, Schilling et al. 2016). Other microscopy techniques rely on the inherent birefringence of tissues, particularly myelin, that enable 3D imaging of large tissue specimens (Axer et al. 2011; Wang et al. 2016). Collectively across many studies, these validation efforts have demonstrated a relatively high correspondence between diffusion MRI estimates and the true fiber structure, particularly in coherent white matter fiber tracts. However, in regions containing multiple fibers intersecting one another, or so-called “crossing-fibers”, it is evident that simple models such as diffusion tensor imaging (DTI) are incomplete. More advanced models show considerable improvement in their accuracy to resolve multiple fibers. Relatively few studies have performed large-scale comparisons between histological specimens and diffusion MRI advanced models. This remains an active area of interest, and recent efforts to map the micro-connectome are likely to yield important insights as high resolution datasets and technologies become more widely available.

Tissue Microstructural Features

Another goal of certain diffusion MRI models is to capture the microscopic physical parameters of tissues, such as cell densities, sizes, and other features. Again, while the specific approach is dictated by the intended goal, these approaches are relatively straightforward if the derived diffusion MRI parameters reflect a physical property that can also be directly measured. For instance, quantitative measures of cell counts or density (Wang et al., 2011), cell sizes or diameters (Barazany et al., 2009), and other various features of the tissue derived from histological measurements have been directly compared to the same parameters derived from diffusion MRI. In more complex tissues however, such as the cerebral gray matter, the diffusion signal behavior become more ambiguous with respect to the underlying tissues. The presence of cell bodies, dendritic fibers, and subcellular structures such as mitochondria give rise to complex diffusion features that are only beginning to be realized.

Pathology in Disease or Injury

The diffusion MRI signal behavior is most ambiguous following injury or disease, since the underlying pathology can affect the measures in unique and sometime conflicting ways. Often, the goal is to determine which pathology is being captured by a specific diffusion metric. For instance, fractional anisotropy, a parameter derived from DTI, is altered from a wide variety of different causes, including neurodegeneration, demyelination, inflammation, and edema, all of which cause FA to decrease. In animal models, induction of a primary pathology such as a demyelinating disease can lead to predictable changes in FA and other parameters. In coherent white matter tracts, a number of studies demonstrated that demyelination is most associated with and increase in diffusion perpendicular to fibers (radial or transverse diffusivity) (Song et al. 2003). On the other hand, axonal degeneration is most associated with a decrease in diffusion parallel to fibers (axial or longitudinal diffusivity) (Budde et al. 2009; Zhang et al. 2009). However, there are many caveats to these relationships that are critical to their interpretation and application outside of these models. First, the relationships are evident in coherent fiber tracts may not hold in complex fiber configurations, since crossing-fibers introduce ambiguity in the underlying tissue organization and model. Second, it should be noted that any neurological injury or disease involves many co-existing pathologies, and a single pathological feature is unlikely to exist in isolation. For instance, whereas axonal injury decreases axial diffusivity, edema can increase and either confound or mask the expected results. Third, while a specific pathological feature can cause a predictable change in diffusion MRI measures, the inverse relationships are not necessarily true. For example, while demyelination causes an increase in radial diffusivity, an observed increase in radial diffusivity, without corroborating evidence, does not necessarily imply that demyelination is the underlying pathology. Consequently, relating a specific pathology to a specific diffusion MRI alteration is unlikely to hold true across all situations, and even moreso with simple diffusion MRI signal models. Again, advanced models have provided greater detail and better interpretability. However, it should be remembered that all models have their own limitations and assumptions, direct corroborating evidence is needed to make strong conclusions.