Revealing the uncharted brain: frontiers in diffusion medical-resonance imaging

Diffusion magnetic-resonance imaging (MRI) comprises a growing number of noninvasive methods that employ diffusion measurements of water molecules in each volume element (voxel) in an image. Clinical applications include diagnosing a stroke in progress,¹ assessing white-matter changes following trauma or neurodegenerative diseases,² and even discovering differences in the ‘brain wiring’ of subjects with cognitive impairments³ or psychiatric disorders.⁴

Indeed, diffusion MRI has even been suggested as a functional imaging method, because it can map spatial and temporal changes in neuronal activity.⁵ Increasingly, diffusion-MRI data is used in therapeutic applications—such as helping to plan the delivery of chemotherapeutic agents to brain tumors⁶—or in neurosurgical planning, where it helps identify viable white-matter pathways that should be avoided during surgery.

The most widely used form of the technique is diffusion-tensor MRI,⁷ which measures a diffusion tensor of water within each voxel. This offers several useful parameters, which reveal different structural characteristics or anatomical features⁸ that help physicians diagnose and assess a large number of disorders and diseases. An example is a direction-encoded color map depicting the orientation of white-matter pathways in the brain (see Figure 1).

One desired aim of this and other medical-imaging methods is the ability to provide artifact-free imaging data or maps of useful physical parameters. This quantitative approach is necessary in both longitudinal studies—where a single subject may be scanned over a period of time—and in population studies, where multiple subjects are scanned, often at different sites. To make meaningful comparisons between subjects or to ‘pool’ data from multiple subjects and make inferences about them, a processing pipeline that assures data quality must be developed, starting at the experimental-design stage and moving through data acquisition, image processing, and analysis. Such a framework allows a radiologist's clinical interpretation to be based on scientific criteria, making population or even epidemiological findings more likely to be valid.

Many elements of this quantitative pipeline are common to other MRI methods, but several are unique to diffusion MRI. Common elements include noise remediation, motion-artifact removal, and identification and removal of artifacts specific to the MRI-acquisition method (e.g., echo-planar MRI). Unique elements of this diffusion-MRI process are raw-image registration⁹ as well as segmentation, registration, statistical analysis,¹⁰ and data mining of multi-dimensional diffusion-tensor-MRI data or data produced by more advanced diffusion-MRI methods.

Many groups have focused on using diffusion-MRI data to help establish anatomical connectivity patterns and pathways between different brain regions.^11,12 This is an important and challenging problem because the underlying topology of white-matter pathways is quite complex. It is not clear whether it is possible to reconstruct these elaborate pathways and connections using diffusion-MRI data alone, particularly at the level of nerve axons. However, our laboratory has taken a different approach. For several years, we have focused on ‘drilling down’ within the image voxel to explore structures and dynamic processes at a finer length scale than previous diffusion-MRI methods allowed. For instance, one application we have developed, AxCaliber MRI,¹³ measures the axon-diameter distribution in white-matter tracts within each voxel: see Figure 2. This distribution is important, as it affects the transmission of nerve impulses, which changes in disease, degeneration, aging, and development.

Another new diffusion-MRI application under development should allow us to segment cortical gray matter according to different microstructural motifs or features present in distinct brain regions.¹⁴ A century ago, Korbinian Brodmann developed cytological-staining techniques to identify distinct gray-matter areas within the cerebral cortex in fixed, excised brain tissue (Brodmann parcellation). We are attempting to perform similar cytoarchitectonic parcellation, but using advanced diffusion-MRI approaches in vivo, and over the entire brain.

Diffusion MRI has a bright future. Improved analysis reveals ever finer-scale structural and anatomical features. Improved MRI hardware permits more images to be acquired in shorter periods with higher resolution and quality. Better MRI sequences improve image quality and resolution. Improved visualization technology provides radiologists and neuroscientists with new ways to depict the high-dimensional data coming out of diffusion MRI. Finally, sophisticated physical models that relate MRI data to tissue microstructure and in vivo dynamic processes are leading to a new generation of diffusion-MRI methods that will reveal architectural features of the brain that, to date, have been invisible to clinicians and neuroscientists. Our group is attempting to discover new and important microstructural features that effectively characterize changes occurring in normal and abnormal development, disease, or degeneration.