PROJECT SUMMARY
The field of human neuroimaging has long been suffering from a problem of low power, in which low signal-to-
noise ratio, multiple comparisons, and small sample sizes result in insufficient statistical power for many studies.
For studies attempting to reveal brain-behavior relationships via functional and structural connectomes, which
are a matrix representation for statistical and physical relationships between brain regions, the story is the same.
From a scientific perspective, this issue reduces the number of findings presented in the literature while also
lowering the replicability of any findings. Since connectomics research strives to ultimately be clinically relevant
by, for example, predicting risk for mental health conditions or informing personalized treatment approaches for
those with existing illness, this problem of low power greatly hinders progress. Fortunately, recent work has
introduced a framework shift in statistics whereby an emphasis on brain networks, rather than individual
connections or edges, improves overall statistical power in functional connectivity analyses. Further work has
shown that using information of the relationships between edges in the connectome to construct the networks
results in even greater power increases, but it is not yet known whether deriving networks within-dataset is more
effective than using independent networks. Therefore, this proposal will investigate whether using within-dataset
networks in network-level statistical procedures results in further power increases. It will also test these network-
level approaches that were developed in functional datasets on the structural connectome. In Aim 1, I will use
data from 3 large functional connectivity datasets spanning several phenotypes to examine if creating the
networks in the same dataset that undergoes statistical testing will offer power improvements over deriving
networks from an independent dataset. In Aim 2, I will evaluate the utility of network-level statistical procedures
in the structural connectome, and in Aim 3, I will extend the methodology from Aim 1 to the structural connectome
to comprehensively determine whether the results seen in the functional connectome extend to the structural
connectome. This work will improve understanding of which variables we can manipulate to achieve higher
statistical power in human connectivity studies and lead the field towards eventual clinical relevance by improving
the neuroimaging tools available to mental health researchers.