The Journal of Neuroscience, August 10, 2011 • 31(32):11597–11616 • 11597 Behavioral/Systems/Cognitive Mapping Human Cortical Areas In Vivo Based on Myelin Content as Revealed by T1- and T2-Weighted MRI Matthew F. Glasser and David C. Van Essen Department of Anatomy and Neurobiology, Washington University School of Medicine, St. Louis, Missouri 63110 Noninvasively mapping the layout of cortical areas in humans is a continuing challenge for neuroscience. We present a new method of mapping cortical areas based on myelin content as revealed by T1-weighted (T1w) and T2-weighted (T2w) MRI. The method is general- izable across different 3T scanners and pulse sequences. We use the ratio of T1w/T2w image intensities to eliminate the MR-related image intensity bias and enhance the contrast to noise ratio for myelin. Data from each subject were mapped to the cortical surface and aligned across individuals using surface-based registration. The spatial gradient of the group average myelin map provides an observer- independent measure of sharp transitions in myelin content across the surface—i.e., putative cortical areal borders. We found excellent agreement between the gradients of the myelin maps and the gradients of published probabilistic cytoarchitectonically defined cortical areas that were registered to the same surface-based atlas. For other cortical regions, we used published anatomical and functional information to make putative identifications of dozens of cortical areas or candidate areas. In general, primary and early unimodal association cortices are heavily myelinated and higher, multimodal, association cortices are more lightly myelinated, but there are notable exceptions in the literature that are confirmed by our results. The overall pattern in the myelin maps also has important correlations with the developmental onset of subcortical white matter myelination, evolutionary cortical areal expansion in humans compared with macaques, postnatal cortical expansion in humans, and maps of neuronal density in non-human primates. Introduction al., 2005a). Of these, only a minority have been mapped to the Modern neuroimaging methods reveal an enormous amount of individuals’ cortical surfaces (Fischl et al., 2008), enabling information about the functional organization and structural surface-based registration with improved intersubject alignment. connectivity of human cerebral cortex. However, interpretation Available architectonic maps are almost exclusively based on of these findings is seriously impeded by inadequacies of exist- postmortem histology. In vivo MR-based methods for accurately ing cortical parcellations. Brodmann’s cytoarchitectonic areas mapping individual cortical areas would be useful, both to im- (Brodmann, 1909), though widely used, are inaccurate over prove cortical coverage and because they can be applied directly much of cortex (Zilles and Amunts, 2010). An additional imped- to living humans. iment is that most analyses have been performed using methods Myeloarchitectural features have been visualized using MRI in that do not respect the sheet-like topology of the convoluted humans, including the stria of Gennari in V1 (Clark et al., 1992; cerebral cortex. Observer-independent probabilistic architec- Barbier et al., 2002; Walters et al., 2003, 2007; Bridge et al., 2005; tonic maps only cover a modest portion of the cortex (Eickhoff et Clare and Bridge, 2005; Eickhoff et al., 2005b) and tripartite lam- ination of area 4 (Kim et al., 2009). Other studies have shown regional differences in T1 or T1-weighted (T1w) image intensity Received May 2, 2011; revised June 7, 2011; accepted June 24, 2011. in cortical gray matter, including differences between association Author contributions: M.F.G. and D.C.V.E. designed research; M.F.G. performed research; M.F.G. analyzed data; M.F.G. and D.C.V.E. wrote the paper. cortices and primary sensory and motor cortices using surface ThisworkwassupportedbyNIHGrantR01MH-60974andbytheHumanConnectomeProject(1U54MH091657- (Fischl et al., 2004; Salat et al., 2009) and volume analyses (Steen 01) from the 16 NIH Institutes and Centers that support the NIH Blueprint for Neuroscience Research. M.F.G. was et al., 2000). Several studies have directly compared MR images to supportedbyaNationalResearchScienceAward–MedicalScientistNIHT32GM007200.TheConteCenterdatawere myelin-stained sections of the same tissue. In marmosets, this provided by John G. Csernansky with assistance from Michael Harms and Lei Wang and were acquired through supportfromNIHGrantsMH056584andMH071616.TheNAMICmultimodalitydatawereprovidedbythePsychiatry approach revealed a strong correlation between T1 and T1w in- Neuroimaging Laboratory and the Surgical Planning Laboratory at Brigham and Women’s Hospital, supported in tensities and histologically measured myelin content and enabled part by grants from the National Alliance for Medical Image Computing (NAMIC-U54 EB005149), and a Neuroimage accurate delineation of several cortical areas (Bock et al., 2009). In AnalysisCentergrant(NACP41RR13218).ComputationswereperformedusingfacilitiesoftheWashingtonUniver- humans, a similar approach demonstrated a myeloarchitectonic sity Center for High Performance Computing, partially supported by Grant NCRR 1S10RR022984-01. We thank Todd Preuss for discussions that helped stimulate this project, consultations on anatomical findings, and comments on difference between areas 4 and 3a in ex vivo T1 slices and myelin- this manuscript; Joel Price for consultations on anatomical findings; and John Harwell and Tim Coalson for software stained sections (Geyer et al., 2011). Also, fibers of the perforant development. path are visible in both T2*-weighted images and in myelin- Correspondence should be addressed to David C. Van Essen, Department of Anatomy and Neurobiology, stained sections (Augustinack et al., 2010). The myelin-related Washington University School of Medicine, 660 South Euclid Avenue, St. Louis, MO 63110. E-mail: [email protected]. MR contrast largely reflects differences in lipids (Koenig, 1991) DOI:10.1523/JNEUROSCI.2180-11.2011 and free and myelin-bound water (Miot-Noirault et al., 1997) Copyright © 2011 the authors 0270-6474/11/3111597-20$15.00/0 concentration, but is also influenced by iron, particularly in T2*- 11598 • J. Neurosci., August 10, 2011 • 31(32):11597–11616 Glasser and Van Essen • In Vivo Myelin Maps of Human Cortex weighted images. However, myelin and iron are strongly colocal- 1 mm slices) were acquired. Data were downloaded from the NAMIC ized within cortical gray matter (Fukunaga et al., 2010). Thus, it is MIDAS website: http://insight-journal.org/midas/collection/view/190. reasonable to conclude that MR-based signals across the cortical Surface generation and processing of T1w volumes. The original unresa- gray matter largely reflect myelin content both directly and indi- mpled T1w volumes were processed through FreeSurfer 4.5’s default recon-all preprocessing pipeline (http://surfer.nmr.mgh.harvard.edu/), rectly. Sigalovsky et al. (2006) found an increased R1 signal (the which includes brain extraction, intensity normalization, segmentation, inverse of T1) in the posterior medial Heschl’s gyrus and sug- generation of white and pial surfaces, surface topology correction, infla- gested that this reflected the high myelin content of primary au- tion of surfaces to a sphere, and spherical registration to the fsaverage ditory cortex. Yoshiura et al. (2000) reported that Heschl’s gyrus, surface based on a measure of surface shape (Sled et al., 1998; Dale et al., particularly the posterior portion, has a lower T2-weighted 1999; Fischl et al., 1999a,b, 2001; Se´gonne et al., 2004). The most accurate (T2w) intensity than the superior or middle temporal gyri. These surfaces were obtained using unresampled T1w volumes, likely through studies suggest that the myelin content of a cortical area covaries minimization of partial-volume effects. FreeSurfer white and pial sur- with both T1w intensity and T2w intensity, but in opposite faces were converted to GIFTI format with application of a transforma- directions. tion matrix to correct for a translational offset (“c_ras”) so that the surface and volume would line up. Using Caret software (Van Essen et al., Here, we tested this hypothesis using T1w and T2w MR 2001) a midthickness surface was generated by averaging the white and images from standard 1 mm isotropic 3T protocols. The ratio pial surface coordinates. The white, pial, and midthickness surfaces with of T1w to T2w signal intensity was mapped to the cortical the original number of vertices are referred to as “native” mesh surfaces. surface using a customized algorithm. The ratio method sub- The registered spherical surface (sphere.reg) was converted to GIFTI stantially improves areal localization by increasing the con- format and resampled onto the fsaverage template spherical surface us- trast to noise between heavily and lightly myelinated areas and ing Caret’s “create deformation map” function. The resultant deforma- also by mathematically canceling the MR-related intensity bias tion map between the native mesh surfaces and the fsaverage surface was field (see methods). We demonstrate that myelin-based anal- applied to bring the native mesh surfaces into register with and onto the ysis (myelin maps) reveals part or all of the areal boundaries 164k vertex fsaverage left or right mesh (hereafter, fs_L or fs_R). The fs_L and fs_R meshes are not in register, so we used a landmark-based regis- for