Functional MRI is commonly used to map between brain regions and cognitive functions. Traditional analyses have focused on testing putative associations between large, known brain regions and broad cognitive tasks; however, machine learning techniques applied globally at the scale of 3D voxels have been used to produce fine-scaled predictive associations. Extracting meaningful insight into brain structure-function relationships from these techniques remains a challenge, though, in part because voxels are arbitrary discretizations of the underlying activation signal, leading to spatial auto-correlation, and interfering with model interpretation. Unfortunately, the optimal spatial granularity in which to model is unknown and likely varies across regions and tasks.

I will present three methods for achieving the sensitivity of global, voxel-based methods while considering spatial properties to produce more reliable, meaningful results. The first is an application of a hybrid regularized regression technique, the Elastic Net, developed to handle correlated predictors in genomics applications. The second approach maps the data onto overcomplete bases in 3D space and uses High Performance Computing to perform sparse regression on the resulting large feature sets. The third method pinpoints localized sources of activation in a graphical model framework.

Carroll, M.K., Cecchi, G.A., Rish, I., Garg, R., Rao, A.R. (2009). Prediction and Interpretation of Distributed Neural Activity with Sparse Models. Neuroimage, 44(1), 112-122.