ReviewUsing high-resolution quantitative mapping of R1 as an index of cortical myelination
Introduction
A basic goal in neuroscience is to map out the functional landscape of cerebral cortical areas identified by structural characteristics (cyto- and myeloarchitecture — Clarke and Miklossy, 1990, Flechsig, 1920, Förster, 1934, Hopf, 1951, Hopf, 1955, Smith, 1907, Vogt, 1906, von Economo and Koskinas, 1925), response preferences, or sensory or motor mapping (Zilles and Amunts, 2010). Identifying homologous areas across species helps us understand how areal function itself has evolved (cf., reptilian jaw bones evolving into inner ear ossicles in mammals). In humans, in-vivo identification of cortical areas generally relies on fMRI mapping of representations of the sensory surfaces (e.g., retinotopy/cochleotopy/somatotopy), motor effectors, or higher level equivalents (attention-o-topy, intention-o-topy). This has been an extremely productive approach, and has shown that much of the cortex is tiled with maps (Graziano and Aflalo, 2007, Schreiner and Winer, 2007, Wandell et al., 2007). These maps not only tend to have a generally consistent location and orientation on the cortical sheet, but also show non-trivial individual differences in size, shape, and possibly even neighbor relations (Sereno and Tootell, 2005) that may have interesting functional consequences (Schwarzkopf et al., 2011).
However, identifying cortical areas through sensory and motor mapping is time consuming; it takes an hour to accurately map one modality (e.g., polar angle and eccentricity mapping to establish visual areas). The robustness, reliability, and extent of maps are strongly dependent upon the participant's level of directed attention to the stimulus over this long period of scanning (Saygin and Sereno, 2008, Silver and Kastner, 2009) making mapping more difficult in children, elderly, and clinical populations. Some cortical areas cannot be defined on the basis of functional maps alone (e.g., primary auditory areas A1 & R — (Dick et al., 2012, Hackett, 2011). And in some individuals (e.g., auditory areas in the deaf (Karns et al., 2012, Ressel et al., 2012), visual areas in the blind, and somatomotor areas in participants with hemiparesis, deafferentation, or amputations), defining inputs or outputs may be absent.
One means of estimating an individual participant's cortical areas is by probabilistic, postmortem cytoarchitectonic atlases, which are provided in a standard anatomical MRI volume or MRI-based cortical surface space (through affine or non-linear transformations (Eickhoff et al., 2005, Fischl et al., 1999, Tahmasebi et al., 2009). This requires no additional scanning time, can define a large number of cortical areas at once (12 Brodmann areas are defined in the current Freesurfer distribution), and facilitates easy comparisons across experiments, scanner sites, and labs. Unfortunately, the degree of inter-individual variation in areal size and shape is considerable (with three-fold areal differences even in V1 — (Schwarzkopf et al., 2011)), making precise definition of any area – particularly smaller and more variable ones – very challenging. This is of particular concern when localizing areas for surgical implantation (e.g., neurostimulators, drug delivery vehicles, or electrode recording grids) or excision (e.g., temporal lobectomy).
A better solution would be to use the signal and tissue contrast information in the MRI anatomical volume itself to identify an individual participant's cortical areas. Indeed, over the last decade much progress has been made in mapping individual cortical areas in-vivo by taking advantage of the sensitivity of the MR longitudinal relaxation time T1 to myelin content, an in-vivo assay of myeloarchitecture (Barazany and Assaf, 2012, Bock et al., 2009, Bock et al., 2013, Clark et al., 1992, Dick et al., 2012, Geyer et al., 2011, Glasser and Van Essen, 2011, Sánchez-Panchuelo et al., 2012, Sereno et al., in press, Sigalovsky et al., 2006, Walters et al., 2003). These myelin mapping methods use different combinations of high resolution images, including high-resolution proton-density (Clark et al., 1992), T1-weighted images (Barazany and Assaf, 2012, Walters et al., 2003), T2-weighted images (Trampel et al., 2011) and volumes derived by taking a ratio of T1- and T2-weighted volumes (Glasser and Van Essen, 2011), synthetic, high-contrast images derived from multi-angle FLASH (Hinds et al., 2008), and quantitative R1 (1/T1) images (Dick et al., 2012, Sereno et al., in press, Sigalovsky et al., 2006).
The recent achievements illustrating the advent of in-vivo histological studies are notable given the technical challenges posed. First, the degree of cortical myelination is strongly cortical-layer-dependent (for examples, see Annese et al., 2004, Braitenberg, 1962). Thus, MRI scans must be of sufficiently high resolution to resolve laminar differences to some degree. Second, interareal differences in myelination are comparable in magnitude to the differences in myelination across layers; upper layers of cortex are often quite lightly myelinated (see whole-brain-slice Gallyas stain of macaque, shown in Fig. 2 of Bridge et al., 2014). Therefore, even minor local inaccuracies in cortical surface reconstruction can significantly distort or obscure estimates of areal differences in myelination. Third, because of the subtlety of the cross-areal differences in myelination, even fairly gentle spatial biases in overall signal intensity and contrast can swamp myelin-related signal changes. In particular, transmit-field (B1+) inhomogeneities affect image contrast and can bedevil widely applied post-hoc histogram-based normalization methods (e.g. Dale et al., 1999) and also ratio methods based on normalizing scans (Glasser and Van Essen, 2011). Myelin mapping methods that combine different kinds of scans must also often contend with vessel artifacts and local spatial distortion that differ between scan types. (However, such ratio-based methods have the advantage of potentially broader application as they rely on widely-available clinical pulse sequences).
In addition to the need for high resolution and spatially unbiased data for myelin mapping, it would be very useful to be able to measure differences in myelination in the same area either over different subjects, or over time in the same subject. However this is not possible with cortical myelination measurements derived from post-hoc normalized contrast-weighted images or from image ratios, in that their numerical values are inherently arbitrary, and unstable even on a single scanner. The ability to make quantitative cross-scan, cross-individual, and cross-site comparisons of cortical myelination would allow for the establishment of norms across development, populations and disease stages (e.g., in multiple sclerosis, Alzheimer's disease and focal dystonias).
To address these challenges, we have developed and refined a method for measuring cortical myelination that takes advantage of recent advances in high-resolution, quantitative MR imaging. Here, we first lay out in some detail the underlying theory and recent advances in MR physics that made high resolution quantitative imaging possible, and discuss the advantages and drawbacks of different quantitative imaging schemes. We then present results from several initial studies using these techniques to measure cortical myelination in visual (Sereno et al., in press) and auditory areas (Dick et al., 2012).
Section snippets
Myelin mapping using R1 mapping methods
T1 is the time constant governing the recovery of the longitudinal component of the magnetization following radio-frequency (RF) excitation, and crucially, an MR parameter closely related to tissue myelination (Koenig et al., 1990) and in particular, the cholesterol that is bound to myelin (Koenig, 1991). Ex-vivo studies using white matter slices from patients with multiple sclerosis or controls have directly compared quantitative MRI measurements of these slices with microscopic
Beyond R1 mapping — challenges for quantitative mapping techniques
Large discrepancies exist between R1 estimates obtained from different methods (Cheng and Wright, 2006, Clare and Jezzard, 2001, Deoni, 2007, Deoni et al., 2005, Ethofer et al., 2003, Gelman et al., 2001, Preibisch and Deichmann, 2009, Wansapura et al., 1999, Wright et al., 2008, Zhu and Penn, 2005). Such differences might in part be due to imperfect correction of B1+-inhomogeneities (as noted above), partial volume effects between different tissue types at low image resolutions, and of
Adapting existing processing pipelines to better but different input data
As cortical surface reconstruction processing pipelines have been made more robust, they have at the same time become more sensitive to and reliant upon the exact statistics of typical T1-weighted input images. Thus, even though quantitative R1 maps are intrinsically preferable to T1-weighted images because the image intensity at each voxel is much more closely correlated with underlying tissue properties, they are more difficult to reconstruct using a pipeline highly optimized for T1-weighted
Cortical myelination and visual areas
Our first study compared R1 maps and retinotopic maps in visual cortex using surface-based methods (Sereno et al., in press). As an initial verification of the technique, we measured six participants' averaged R1 values (sampled at 8 depths in each area) in three probabilistically-defined regions-of-interest (ROIs) known to differ in myelination density — namely the angular gyrus, which is lightly myelinated, and visual areas MT and V1, which are highly myelinated (see Fig. 2a).
First, we found
Conclusion
Recent in-vivo histological studies using quantitative MRI have demonstrated the validity of the MR parameter R1 as a biomarker for myelin concentration. The subtle changes in R1 across the cortical surface require specially-dedicated MR pulse sequences that allow for accurate, precise and efficient image acquisition with sufficient image resolution to resolve the laminar distribution of the cortical layer. We demonstrated that high-resolution R1 mapping could be used for the in-vivo
Conflict of Interest
The authors declare no conflict of interest.
References (135)
- et al.
Radiofrequency map of an NMR coil by imaging
Magn. Reson. Imaging
(1993) - et al.
A myelo-architectonic method for the structural classification of cortical areas
NeuroImage
(2004) - et al.
Visualizing the entire cortical myelination pattern in marmosets with magnetic resonance imaging
J. Neurosci. Methods
(2009) - et al.
Optimizing T1-weighted imaging of cortical myelin content at 3.0 T
NeuroImage
(2013) - et al.
Delineating extrastriate visual area MT(V5) using cortical myeloarchitecture
NeuroImage
(2014) - et al.
Cortical surface-based analysis. I. Segmentation and surface reconstruction
NeuroImage
(1999) - et al.
Quantification of T1 values by snapshot-FLASH NMR imaging
J. Magn. Reson. Med.
(1992) - et al.
Regional specificity of MRI contrast parameter changes in normal ageing revealed by voxel-based quantification (VBQ)
NeuroImage
(2011) - et al.
A new SPM toolbox for combining probabilistic cytoarchitectonic maps and functional imaging data
NeuroImage
(2005) - et al.
Sequence-independent segmentation of magnetic resonance images
NeuroImage
(2004)