A Neurobiologically Grounded Connectome
The Human Connectome Project will provide a treasure trove of information at an unprecedented level of detail. It is important that the interpretation of these connectivity data respects key organizational features of the human brain and the physical dimensions of its components.
- Cerebral cortex is the dominant subdivision. Including subcortical white matter, it occupies ~80% of brain volume but contains only ~20% of the brain’s 85 billion neurons.
- The highly convoluted cortical sheet (~3 mm thick and ~1,000 cm2 per hemisphere) contains ~150-200 cortical areas that differ from one another in connectivity, function, and architecture. These areas span a ~100-fold range (0.2 – 20 cm2) in average size.
- Each cortical area varies 2-fold or more in size across the normal adult population. Data from the HCP may reveal whether specific behavioral capabilities are correlated with individual variability in the dimensions of functionally specialized areas or networks.
- Individual variability in cortical folding and in areal boundaries relative to these folds is a major impediment to intersubject comparisons.
- Cerebellar cortex occupies ~10% of brain volume, and contains 80% of the brain's total neurons, and is a sheet 1/3 as thick and half the surface area of cerebral cortex. Its lobes and lobules differ in their function and connectivity and are also variable across individuals.
- Subcortical structures occupy the remaining ~10% of brain volume but contain only ~1% of its neurons. They include hundreds of cortical nuclei and subnuclei, most of which are too small to be resolved by conventional in vivo neuroimaging.
The HCP will generate invaluable information about connectivity, function and parcellation of the cerebral cortex, cerebellum, thalamus, and basal ganglia. Relatively little is currently known about human brain connectivity, but studies of nonhuman primates indicate that each cortical area and each subcortical nucleus typically has reciprocal connections with dozens of other areas and nuclei. Moreover, connection strengths for any given structure range over many orders of magnitude.
The core objective of the HCP is to determine the connectomes for each individual subject studied. We consider a connectome to be a connectivity matrix that encodes objective measures of the anatomical or functional connections between each pair of identified brain locations. Diffusion imaging yields evidence regarding the likelihood that two regions are directly connected. Such measures are correlated with, but not proportional to, connection strength (the number of fibers in the pathway). Functional imaging (R-fMRI) yields evidence regarding the dynamic interactions between two regions, which tends to be high if the regions are directly connected but may also reflect a history of common activation.
We consider a "dense connectome" as a connectivity matrix between pairs of gray-matter voxels at the finest resolution available (e.g., voxel size of ~1 mm for 7T data and ~2 mm for 3T data). A "parcellated connectome" is a more compact representation of connections between identified brain subdivisions (parcels or nodes). Parcellated connectomes are expected to be the datasets most widely used by the scientific community.
Individual variability in connectomes may account for diversity in many aspects of normal cognition, perception, and motor skills. Several challenges will be addressed in order to chart connectivity and test for such correlations.
- Connections. Current in vivo methods for mapping long-distance connectivity have significant technical limitations and generally yield many false positives and false negatives. Our proposed methodological refinements will greatly improve the sensitivity and accuracy of connectome mapping.
- Brain parcellation. Neighboring brain locations, especially in cerebral cortex, typically have similar connectivity patterns. Abrupt transitions in connectivity patterns often signify a boundary between cortical areas or subcortical nuclei, but can sometimes occur within a classical parcel. We will incorporate a connectivity-based parcellation approach that enables charting of functionally distinct parcels.
- Inter-subject registration. Accurate comparisons across individuals are vital for characterizing commonalities as well as individual differences, but they must cope with highly variable cortical folding patterns. We will utilize existing state-of-the-art registration methods, both volume-based and surface-based, for intersubject alignment. In addition, we will implement novel connectome-based registration methods that may substantially improve intersubject alignment.