|
Scientific Understanding of Consciousness |
Multi-Modal Parcellation of Human Cerebral Cortex
Nature 536, 171–178 (11 August 2016) A multi-modal parcellation of human cerebral cortex Matthew F. Glasser, et.al. Department of Neuroscience, Washington University Medical School, Saint Louis, Missouri 63110, USA FMRIB Centre, Nuffield Department of Clinical Neurosciences, John Radcliffe Hospital, University of Oxford, Oxford OX3 9DU, UK Department of Computing, Imperial College, London SW7 2AZ, UK Department of Biomedical Engineering, Washington University, Saint Louis, Missouri 63110, USA Center for Magnetic Resonance Research (CMRR), University of Minnesota, Minneapolis, Minnesota 55455, USA Donders Institute for Brain, Cognition and Behavior, Radboud University, Nijmegen 6525 EN, The Netherlands Department of Cognitive Neuroscience, Radboud University Medical Centre Nijmegen, Postbus 9101, Nijmegen 6500 HB, The Netherlands [paraphrase] Understanding the amazingly complex human cerebral cortex requires a map (or parcellation) of its major subdivisions, known as cortical areas. Making an accurate areal map has been a century-old objective in neuroscience. Using multi-modal magnetic resonance images from the Human Connectome Project (HCP) and an objective semi-automated neuroanatomical approach, we delineated 180 areas per hemisphere bounded by sharp changes in cortical architecture, function, connectivity, and/or topography in a precisely aligned group average of 210 healthy young adults. We characterized 97 new areas and 83 areas previously reported using post-mortem microscopy or other specialized study-specific approaches. To enable automated delineation and identification of these areas in new HCP subjects and in future studies, we trained a machine-learning classifier to recognize the multi-modal ‘fingerprint’ of each cortical area. This classifier detected the presence of 96.6% of the cortical areas in new subjects, replicated the group parcellation, and could correctly locate areas in individuals with atypical parcellations. The freely available parcellation and classifier will enable substantially improved neuroanatomical precision for studies of the structural and functional organization of human cerebral cortex and its variation across individuals and in development, aging, and disease. Neuroscientists have long sought to subdivide the human brain into a mosaic of anatomically and functionally distinct, spatially contiguous areas (cortical areas and subcortical nuclei), as a prerequisite for understanding how the brain works. Areas differ from their neighbours in microstructural architecture, functional specialization, connectivity with other areas, and/or orderly intra-area topographic organization (for example, the map of visual space in visual cortical areas). Accurate parcellation provides a map of where we are in the brain, enabling efficient comparison of results across studies and communication among investigators; as a foundation for illuminating the functional and structural organization of the brain; and as a means to reduce data complexity while improving statistical sensitivity and power for many neuroimaging studies. The human cerebral cortex has been estimated to contain anywhere from ~50 to ~200 areas per hemisphere. However, attaining a consensus whole-cortex parcellation has been difficult because of practical and technical challenges that we address here. Most previous parcellations were based on only one neurobiological property (such as architecture, function, connectivity or topography), and many cover only part of the cortex. Combining multiple properties provides complementary as well as confirmatory information, as different properties distinguish different sets of areal boundaries, and more confidence can be placed in boundaries that are consistent across multiple independent properties. We analysed all four properties across all of neocortex in both hemispheres, using new or refined methods applied to the uniquely rich repository of exceptionally high-quality magnetic resonance imaging (MRI) data provided by the Human Connectome Project (HCP), which benefited from major advances in image acquisition and preprocessing. Architectural measures of relative cortical myelin content and cortical thickness were derived from T1-weighted (T1w) and T2-weighted (T2w) structural images. Cortical function was measured using task functional MRI (tfMRI) contrasts from seven tasks. Resting-state functional MRI (rfMRI) revealed functional connectivity of entire cortical areas plus topographic organization within some areas. Previous parcellations typically used either fully automated algorithmic approaches, or else manual or partly automated neuroanatomical approaches in which neuroanatomists delineated areal borders, documented areal properties, and identified areas after consulting prior literature. Here we combined both approaches. For the initial parcellation, we adapted a successful observer-independent semi-automated neuroanatomical approach for generating post-mortem architectonic parcellations to non-invasive neuroimaging. We used an algorithm to delineate potential areal borders (transitions in two or more of the cortical properties described above), which two neuroanatomists then interpreted, documenting areal properties and identifying areas relative to the extant neuroanatomical literature. We then used a fully automated algorithmic approach, training a machine-learning classifier to delineate and identify cortical areas in individual subjects based on multi-modal areal fingerprints, allowing the parcellation to be replicated in new subjects and studies. We have produced a population-based 180-area per hemisphere human cortical parcellation using exceptionally high quality multimodal data from hundreds of Human Connectome Project subjects aligned using an improved areal feature-based cross-subject alignment method (MSMAll). Inspired by an observer-independent post-mortem architectural parcellation approach, we developed a semi-automated neuroanatomical approach adapted to non-invasively acquired multi-modal MRI data. Although algorithms determined the final areal borders, the multi-modal data were carefully interpreted by neuroanatomists, the properties of each cortical area were documented, and each area was named in relation to the extant neuroanatomical literature. A cross-validation showed that the areas forming the parcellation were robustly and statistically significantly different from their neighbours across multiple modalities. We identify this parcellation as HCP-MMP1.0 (Human Connectome Project Multi-Modal Parcellation version 1.0), making the version 1.0 designation because we anticipate future refinements as better data become available. Unexpectedly, we discovered that despite improved intersubject alignment, some areas have atypical topological arrangements in some subjects, which we demonstrated for areas 55b, FEF, and PEF. We developed a fully automated method for parcellating individual subjects based on a machine learning classifier that can cope with this kind of individual variability. The areal classifier detected 96.6% of individual subject cortical areas in new subjects, including atypical areas, and replicated the group parcellation in an independent sample. Though we made extensive use of the HCP’s specialized task fMRI battery when generating the parcellation, we showed that task fMRI data is not essential for future studies aiming to use the areal classifier to automatically define the cortical areas in their subjects. Instead, it suffices to acquire the same core set of MRI images needed for the rest of the HCP’s software pipelines. By generating a robust neuroanatomical map of human neocortical areas—a century-old aim of neuroscience—and providing methods for mapping these areas in any individual undergoing study with non-invasive neuroimaging, the present work represents a major advance relative to previous human cortical parcellations. The overall approach described here shows that we can produce sharp, reproducible brain images across multiple non-invasive neuroimaging modalities. We can generate a highly reproducible and generalizable cortical parcellation through state-of-the-art methods of data acquisition, preprocessing, and analysis designed to compensate for individual variability and thereby minimize blurring of images. These improvements, together with the new parcellation, make it desirable to use spatial localization methods that move beyond the traditional use of stereotaxic coordinates combined with Brodmann areal assignments to characterize centers of cortical activation in fMRI studies. [end of paraphrase]
Return to — Brain Anatomy Return to — Brodmann Areas |