By causing increasing network disconnection, white matter damage may in fact accentuate rather than destroy the main feature of www.selleckchem.com/products/ABT-888.html the model, viz separate persistent modes of atrophy. Nevertheless, we intend to investigate nonlinear models in our future work. We will also investigate network models based on neuronal excitability (Santos et al., 2010) rather than proteopathic transmission. The model should apply to other dementias like Huntington’s, corticobasal syndrome, semantic dementia, and posterior cortical atrophy, but this aspect will require more data. We expect the multiple-comparisons

problem will be exacerbated and may become statistically untenable (e.g., 36 comparisons for six dementias). Estimates of both higher eigenmodes and rarer dementias are going to be noisier, and establishing their equivalence may require more accurate brain networks than current technology allows. Several technical challenges are inherent in our processing pipeline. Spatial and angular resolution of current HARDI data is poor, sometimes making co-registration with T1 MRI difficult. Highly atrophied subjects

sometimes fail to co-register properly. These problems necessitated CH5424802 cell line manual inspection of co-registration outcomes and rejection of problematic cases. SPM- and FreeSurfer-based volumetrics are known to be noisy, with less-than-perfect test-retest reliability. Although we have mitigated these effects by choosing a relatively coarse network with only 90 large-sized ROIs, they cannot be completely ruled out. Tractography is limited by a “distance bias” and lack of spatial and angular resolution (Behrens et al., 2007). Conventional tractography fails to capture many important but short-curved U-shaped fibers, whereas probabilistic tractography sometimes leads to

unrealistic fiber tracts having little anatomic justification. Finally, brain network statistics are liable to vary with the choice and definition of nodes; hence, we have used anatomically defined parcellations to define nodes—an approach else that we feel has more physical basis than arbitrary choice of nodes. Although we showed that our results are largely unchanged under two quite different parcellation schemes (SPM and FreeSurfer), the effect of other choices of network nodes remains untested. We model dementia progression as a diffusion process on a hypothesized brain network GG = VV,EE whose nodes vi   ∈ VV represent the i  th cortical or subcortical gray matter structure, and whose edges, ei  ,j ∈ EE, represent white-matter fiber pathways connecting structures i and j. Structures vi comes from parcellation of brain MRI, and connection strength, ci,j, is measured by fiber tractography ( Behrens et al., 2007). Consider an isolated population of fibers from an affected (R2) to an unaffected (R1) region.