Multimodal Brain Imaging Fusion for the White-Matter Fiber Architecture in the Human Brain

Uncovering white-matter fiber architecture, particularly that of the human brain, is one of the major and most recent interests in the neuroscience community [1]. Recently, a novel analysis method for quantifying white-matter axonal orientation at a resolution of a dozen microns has been reported [2]. The proposed method is based on Nissl staining, one of the most common tools for studying postmortem brain tissue. For a long time, Nissl staining has been used extensively to investigate cytoarchitectonic features in the gray matter; but meanwhile has been almost entirely ignored for studying the features of white-matter fiber architecture. Therefore, the proposed method aims to reconstruct white-matter pathways in the human brain using the abundance of existing Nissl-stained datasets.

It is well known that an axon is usually wrapped by a myelin sheath along its whole length. In the human brain, the myelin sheath is produced by oligodendrocytes. When postmortem histological slices are stained for Nissl, the cell nuclei across the brain are visible, including neurons and oligodendrocytes. More interestingly, oligodendrocytes stained within the white matter cluster specifically in short rows aligned with the axons that they wrap. Therefore, the authors hypothesized that by measuring the local orientation of the oligodendrocytes within the white matter, one could infer the underlying axonal architecture. To test this hypothesis, the authors developed a “Nissl-ST” analysis method to estimate the orientation map of the fibers from the Nissl-stained slices, and further compared their results with a published image of postmortem polarized light imaging (PLI) and with the orientation map derived from in vivo diffusion magnetic resonance imaging (dMRI). They found that the proposed “Nissl-ST” method can reveal the local orientation of fibers, including the major white-matter tracts (for example, corpus callosum) and the fine crossing fiber bundles (for example, Edinger’s Comb and the angular bundle). Nevertheless, it is to be noted that the “Nissl-ST” analysis is an indirect observation of axonal fibers, and the current implementation is two-dimensional in nature, so the orientation of through-plane crossing fibers remains unresolved.

Besides the above novel “Nissl-ST” method based on the Nissl staining, many imaging techniques with different modalities are now available for the exploration of white-matter structural architecture in the human brain at multiple spatial scales. We summarize the features and challenges of some typical techniques in Table 1. At the macro-scale level, dMRI is the unique imaging technique to non-invasively provide information related to the microstructure of the brain white-matter tissue in vivo. Using standard clinical MR scanners and protocols, the dMRI is often acquired at a spatial resolution of 2–3 mm, which is generally insufficient to disentangle the fanning, bending, and crossing of fibers. Fortunately, recent technical developments, especially the use of high-performance gradients, have allowed for an increase in resolution to ~1 mm [3]. On the other hand, dMRI can also be used for human post mortem whole brain or excised blocks (i.e. either post mortem or from resection surgery). Using an ultra-high field and strong gradients, ex vivo dMRI can boost the spatial imaging resolution to the level of a few hundred micrometers, which has revealed complex fiber architecture in the human brain [4].

Table 1.

Features and challenges of typical brain imaging modalities for the white-matter fiber architecture in the human brain.

Imaging modality	Contrast	Features	Challenges
Macro scale (millimeter level spatial resolution)
Diffusion MRI (dMRI)	Mobility of water molecules in brain tissue	(1) Currently the only method for in vivo mapping of human white-matter fibers (2) Can also be used for human post mortem whole brain or tissue blocks	Limited spatial resolution, thus challenges to disentangle the fanning, bending, and crossing of fibers
Meso scale (micrometer level spatial resolution)
Histologic staining and optical microscopy (OM):
Myelin staining/neurofilament labeling	Dedicated labeling of axonal myelin sheaths and neurofilaments	(1) Gold standard for studying fiber anatomy in human brain (2) Collection of myelin staining approaches (e.g., silver staining and antibody-based immunocytochemical staining) (3) Integration of histological staining/labeling with tissue clearing to enable mapping fiber tracts in centimeters-thick human brain tissue blocks	(1) Cumbersome brain tissue fixation, embedding and sectioning (2) Unpredictable signal-to-noise, significant time input, and complex staining procedure
Nissl staining	Labeling of cell nuclei across the brain	(1) Classic post mortem human brain imaging, traditionally for identifying cytoarchitecture across brain structures (2) Measurement of local orientation of oligodendrocytes within white matter to infer underlying axonal architecture	Solely estimates 2D fiber orientation in brain sections
Polarized light imaging (PLI)	Phase shift in light transmittance of myelin sheaths surrounding axons	(1) Label-free and stain-free (2) Uses the intrinsic birefringence of myelin sheaths (3) Demonstration of 3D fiber orientations in serial microtome sections of entire human brain	Thin brain sectioning (typically 100 μm) for light transmittance, thus requiring enormous numbers of sections and imaging to cover entire brain
Polarization-sensitive optical coherence tomography (PSOCT)	Intensity and polarization shift in light reflectance of brain tissue	(1) Label-free and stain-free (2) Can be used to study both cytoarchitecture and myeloarchitecture (3) Typical 1–2 mm penetration depth in brain tissue, making serial sectioning and continuous imaging possible	(1) Optic axis orientation for in-plane fiber orientations, and the gradient of the retardance image to estimate the through-plane fiber orientations (2) Field of view of a single scan typically covers several millimeters, and thus much image stitching for human brain tissue
Small-angle X-ray scattering tensor tomography (SAXS-TT)	Signal anisotropy of the constructive interference of scattered photons within white matter	(1) Label-free and stain-free (2) Field of view covers typically beyond 10 mm	(1) Very long scanning time (2) Possible sample damage due to radiation dose imparted by X-rays
Micro scale (nanometer level spatial resolution)
X-ray nanoholotomography (XNH)	Phase shift as X-rays penetrate brain tissue	(1) Label-free, but preferred to heavy metal staining (2) Image millimeter-scale volumes with isotropic voxel sizes <100 nm	(1) Big data management (2) 3D reconstruction, including segmentation and identification of dendrites and myelinated axons
Electron microscopy (EM)	Morphological signatures of neurons, glia, and other cells	(1) Nanometer-level resolution of (sub)cellular structure (2) Label-free, but sample preparation depends on specific imaging techniques and research aims	(1) Time-consuming and labor-intensive (2) Big data management (3) 3D reconstruction and image interpretation

At the meso-scale level, a great many imaging techniques based on optical microscopy (OM) have been developed to target the neuronal structure of white matter. However, since the fluorescent protein transgenesis and tracer injections are not ethically feasible in humans, a limited number of techniques can be applied to human brain. Further, nearly all of these techniques are based on postmortem human brain tissue, using either histological stains or intrinsic contrast imaging. For example, a collection of myelin staining approaches (for example, silver staining and antibody-based immunocytochemical staining) can visualize the fiber orientation in thin histological sections at a thickness that typically ranges from a few to dozens of microns. These myelin staining approaches are still the gold standard for studying fiber anatomy in the human brain. In recent years, multiple serial-sectioning imaging techniques have been developed to enable automatic brain imaging in three dimensions [5]. More excitingly, the advance in tissue clearing provides a new opportunity to histologically map fiber projections in the human brain tissue blocks several centimeters thick [6]. Using the CLARITY tissue clearing method, a recent study estimated the three-dimensional neuronal fiber orientations from fluorescently labeled neurofilament images and directly compared ex vivo dMRI and optical imaging in the same human brain tissue block [7]. However, the labeling-dependent histological techniques, including myelin staining and neurofilament labeling, are prone to artifacts due to tissue fixation, embedding, and sectioning [8]. It is difficult to deal with the resultant non-uniform staining and tissue distortion. Therefore, some imaging techniques, for example, PLI [9], polarization-sensitive optical coherence tomography (PSOCT) [10], and small-angle X-ray scattering tensor tomography (SAXS-TT) [11], have been developed to utilize the intrinsic birefringent and anisotropic properties of myelinated fibers as the contrast mechanism. In these techniques, no staining or labeling is required to detect and visualize fiber structures, which is a huge benefit compared with classical histological fiber studies. On the other hand, PLI, PSOCT, and SAXS-TT are able to image fibers at a spatial resolution of tens of microns across the whole brain, which is important for resolving regions of crossing or kissing fibers. It is notable that the intrinsic physical contrast-based imaging techniques still face challenges of their own, such as a relatively low signal-to-noise ratio and complicated image processing.

At the micro-scale level, it is increasingly attractive to illustrate fine structural architecture at nanometer spatial resolution. Electron microscopy (EM) has sufficient resolution (i.e. a few nanometers and even better), but obtaining three-dimensional EM volumes of even small human brain sample requires collecting millions of EM images across thousands of thin sections and, therefore, can be time-consuming and labor-intensive [12]. In comparison, X-ray holographic nano-tomography (XNH) can image millimeter-scale volumes with sub-100-nm resolution, which bridges a key gap between OM and EM [13]. By applying image stitching, XNH might provide the optimal voxel size to ensure sufficient resolution for microstructural architecture in a relatively large human brain tissue sample.

Taking all the benefits and challenges of these brain imaging techniques into account, it can be concluded that revealing an overall picture of white-matter structural architecture in the human brain, from large fiber bundles to the single axon, will require a sophisticated combination of available imaging techniques and dedicated fusion of data from the different scales. Confirming the reliability of the various approaches and integrating multimodal evidence, from both in-vivo and ex-vivo studies, provides more productive and reproducible neuroscience research.