The structural and functional intraconnectivity of the human brain is determined indirectly, leading to significant uncertainty and limiting the adoption of connectomics when researching various neurological disease states. Furthermore, even though a structural white matter connection is needed for distinct brain regions to functionally communicate, direct measurement of either connectome does not produce complimentary results.
In this report I evaluate various proposed linear models capable of linking the structural and functional connectome. I propose a model that incorporates second and third order indirect connectivity to allow functional data to be predicted at edges where no direct structural connection exists.
This neuroscience study evaluates models that predict functional brain connectivity from structural connectivity, using data from diffusion MRI and resting-state fMRI.
The brain is made up of complex networks that connect different regions to one another. Scientists often study these networks using two types of “maps” called connectomes: the structural connectome, which shows the physical wiring (like white matter pathways), and the functional connectome, which maps how different parts of the brain communicate during activity.
Figure: Map of the human brain where each colour corresponds to a different region.
This project focused on understanding the relationship between these two types of brain maps. Using MRI data, I tested several models to see how well functional connections could be predicted from the brain’s structural layout - including not just direct pathways, but also indirect ones that link regions through multiple steps. I found that adding these indirect connections, and applying a technique called LASSO regularization to avoid overfitting, significantly improved model accuracy. These results show that even when two brain regions aren’t directly connected, their functional relationship can still be explained through the brain’s broader network structure.
Figure: Functional connectivity maps generated by fitting the five linear models to direct and indirect structural connectivity maps. The connectome in the top left shows the ground truth determined using rsfMRI.
