Imaging brain microstructure with diffusion MRI: practicality and applications

Daniel C. Alexander, Tim B. Dyrby, Markus Nilsson, Hui Zhang
2017 NMR in Biomedicine  
This article gives an overview of microstructure imaging of the brain with diffusion MRI and reviews the state of the art. The microstructure-imaging paradigm aims to estimate and map microscopic properties of tissue using a model that links these properties to the voxel scale MR signal. Imaging techniques of this type are just starting to make the transition from the technical research domain to wide application in biomedical studies. We focus here on the practicalities of both implementing
more » ... h techniques and using them in applications. Specifically, the article summarizes the relevant aspects of brain microanatomy and the range of diffusion-weighted MR measurements that provide sensitivity to them. It then reviews the evolution of mathematical and computational models that relate the diffusion MR signal to brain tissue microstructure, as well as the expanding areas of application. Next we focus on practicalities of designing a working microstructure imaging technique: model selection, experiment design, parameter estimation, validation, and the pipeline of development of this class of technique. The article concludes with some future perspectives on opportunities in this topic and expectations on how the field will evolve in the short-to-medium term. KEYWORDS brain, diffusion MRI, magnetic resonance imaging, microstructure imaging, neuroimaging, quantitative imaging, virtual histology | INTRODUCTION The central vision in microstructure imaging is of virtual histology: estimating and mapping histological features of tissue using non-invasive imaging techniques, such as MRI. This virtual histology has several advantages over classical histology: (i) it is non-invasive, avoiding the need for tissue samples, e.g. from biopsy; (ii) it views intact in situ tissue, avoiding disruptions that arise from tissue extraction and preparation; (iii) it is non-destructive, so enables repeat measurements for monitoring; (iv) it provides a wide field of view, typically showing a whole organ or body, rather than the small samples often used in classical histology; and (v) data acquisition is relatively fast, cheap and automated compared with classical histology. Classical histology has been a lynchpin in the development of modern neuroscience including understanding the brain's macroscopic organization (see, e.g., Reference 1), the mechanisms of connectivity and communication, 2 and the pathologies underpinning neurodegeneration. 3 Such work primarily uses sliced post-mortem tissue. Clinical applications in the brain are mostly for post-mortem confirmation of diagnosis, as in vivo brain biopsy is normally justified only in aggressive diseases such as grading brain tumours. The non-invasive, non-destructive nature of virtual histology offers the potential to study the live brain in situ in healthy volunteers or patients. The relative ease of data acquisition allows population studies that provide insight into anatomical variability. Furthermore, its non-destructive nature allows repeat measurements to monitor changes during normal development or pathological processes. Clinically, virtual histology avoids biopsy and the potential side effects of the invasive procedure, and provides a window on tissue changes when the risk of side effects prohibits biopsy. Moreover, the wide field of view that virtual histology provides potentially reduces false negatives that may arise from, say, poor targeting of a biopsy. Figure 1 compares typical images from classical histology and microstructure imaging. The clear advantage of classical histology is its level of anatomical detail; its submicrometre image resolution provides vivid insight into the cellular architecture of tissue, whereas microstructure imaging FIGURE 1 Comparison of classical histology and microstructure imaging showing a range of microstructure imaging techniques in the current literature organized by target tissue feature. A-D, Imaging indices of neurite (axon or dendrite) density with classical histology from Reference 4 (A) and by model-based dMRI (B-D). Maps show the cylinder fraction from Reference 4 (B), orientation dispersion (OD), neurite density index (v ic ) and isotropic fraction (v iso ) from NODDI (C), 5 and isotropic fraction, 'stick density', and tissue mean diffusion from CODIVIDE (D). 6 E-G, Imaging fibre orientation distribution. E, Estimation of fibre directions from histology and corresponding estimates from dMRI. 7 F, In vivo fibre orientation mapping using constrained spherical convolution. 8 G, Combined mapping of microstructure and orientation by the spherical mean technique. 9 H-L, Imaging indices of axon diameter. H, Histology provides high-resolution maps enabling measurements of individual axon diameters; images from Reference 10. I, Estimated axon diameter distributions from diffusion MRI using AxCaliber in Reference 10 of the in vivo rat-brain cluster into groups reflecting corresponding diameter histograms from histology. J-L, Axon diameter indices from the monkey brain using ActiveAx (J), 11 ex vivo spinal cord (K) 12 and in vivo spinal cord using 300 mT/m gradients (L). 13 M-P, Imaging cell shape indices. M, Classical histology reveals elongated cells in a meningioma to the left and rounder cells in a glioma to the right; from Reference 14. N, Fractional anisotropy from DTI is low in both meningioma and glioma tumours, but the microscopic anisotropy (μFA) from DIVIDE is more specific to cell shape and shows high value in the meningioma only. 14 O,P, A similar measure of the microscopic anisotropy from double diffusion encoding in a rat brain (O) 15 and a healthy human brain (P). 16 Q-T, Imaging myelin density. Classical histology by luxol fast blue shows reduced myelin density in the brain of a multiple sclerosis patient (Q) and MRI-derived maps using quantitative relaxometry show similar features (R). 17 S, MRI used to track the myelination in infants. 18 T, An early example of the myelin water fraction from relaxation-weighted MRI 19
doi:10.1002/nbm.3841 pmid:29193413 fatcat:43pukyiwubb6bhsaey5gs3sizy