Desktop version

Home arrow Computer Science arrow Computational Diffusion MRI: MICCAI Workshop, Athens, Greece, October 2016



The cytoarchitecture of the cerebral cortex in humans and many other mammalian species was first investigated in the early 1900s using histological sections of post-mortem brains stained for cell bodies or myelinated fibers. Pioneers of this era [1-3] discovered that the microstructure of the cortex was organised into six layers, the columnar appearance of which varies throughout the cortical sheet. Roughly homogeneous modules of variable size were observed and attributed to functional specificity, starting with striate cortex whose border with V2 is easily visible to the naked eye in hand-cut unfixed tissue. The hypothesis of a mosaic of internally homogeneous areas prevails today and these classical parcellations have been widely adopted in modern studies, for example, to localise activation foci in functional imaging studies.

Despite their pervasiveness, it is evident that traditional cortical maps suffer many methodological limitations. One limitation is observer dependant bias, [4]. Another problem is that histological methods are often restricted to a single cell stain per specimen, thus requiring the observer to combine identified boundaries across differently distorted adjacent sections. These limitations may explain the variability in size, location and number of cortical areas reported by different such methods [4-7]. The labour-intensive process of histological sectioning enforces further limits on the sample size used to generate such cortical maps. Subsequent studies have demonstrated a large degree of intersubject variability with regards to the exact location and extent of several well-defined cortical areas. Given this, classical maps derived from a small sample of cadaver brains, are unlikely to accurately reflect boundary definitions for the entire population. Other considerations include the introduction of artefacts from histological sectioning (e.g., including unique nonlinear distortions in each section due to slide mounting and outright tears in the tissue), which complicate registration of data back into undistorted 3D space.

Despite their lower resolution, in vivo methods have the potential to overcome some of these limitations—in vivo analysis provides observer independent image processing, much larger samples sizes, the possibility of multi-modal studies, and completely avoids histology artefacts. Thus far, in vivo investigations of cortical microstructure have focussed predominantly on myeloarchitectonics, via quantitative T1 [8] and R1 (1/T1) mapping [9-11] mapping using multiple flip angles, the ratio of T1-weighted over T2-weighted images [12, 13], and multiple inversion times (MP2RAGE). However, myelination density provides only a partial picture of cortical microstructure. More recently, Glasser et al. extended their T1-weighted/T2-weighted methods into a multi-modal framework for cortical mapping [14]. This approach combined myelin maps , resting state, and task-based functional MRI measures of approximately 200 subjects with expert anatomical knowledge and a complex processing pipeline to produce a semi-automated, group-average, full-hemisphere, cortical parcellation. Nevertheless, the datasets in that paper do not directly measure the fine-grained structural information that is associated with the cyto- and myeloarchitecture of the cortex. For investigators wishing to acquire a detailed understanding of the structure-function relationship in the cortex, it may be desirable to include measurements of additional features that characterise grey matter (GM) micro-environments.

There is a growing body of evidence [15-22] to support the use of diffusion MRI in cortical imaging facilitated by recent technological advancements, such as multi-band excitation, magnetic field probes [23], and ultra high field imaging. In particular, investigators have demonstrated changes in the dominant diffusion direction between the primary somatosensory and motor cortices, via a measure of radiality [20]. Others have shown the relationship between cortical gyrencephaly and diffusion tensor metrics [21]. Nagy et al. demonstrated the in vivo individualised discriminative power of a feature set derived from high angular resolution diffusion imaging data by testing distinct fMRI-based regions of interest [22]. These findings suggest that dMRI in grey matter may provide an additional informative modality for replicating and possibly refining/redefining the boundary definitions of existing cortical parcellation approaches.

The dMRI signal is sensitive to several microstructural features, including but not limited to axon diameter, neurite density, and dominant fibre direction and hence may offer additional structural information beyond bulk myelination density alone. To test the idea that grey matter dMRI might provide a richer description of cortical microenvironments, we used unsupervised, surface-normal- based, group-average cortical parcellation derived from dMRI-based measures of cytoarchitecture. We applied and refined the framework initially developed by Nagy et al. [22] to a large group of subjects using surface-based and surface-referenced cross-subject averaging of dMRI to obtain a hemisphere-wide map of grey matter diffusion patterns from high resolution, 3T data. The resultant parcellation exhibits several coherent clusters that correspond closely with the locations of well-known cortical areas, despite the classifier having no prior information or non-local spatial constraints of any kind.

Found a mistake? Please highlight the word and press Shift + Enter  
< Prev   CONTENTS   Next >