Over the last several years, statistical learning methods have become mainstream in the analysis of Functional Magnetic Resonance Imaging (fMRI) data, spurred on by a growing consensus that meaningful neuroscientific models built from fMRI data should be capable of accurate predictions of behavior or neural functioning.
Two years ago, the NIPS workshop New Directions on Decoding Mental States from fMRI Data reflected on progress so far and future directions. Most of the open questions discussed considered how to advance beyond single-subject, single-task, voxel-by-voxel, static analysis to better uncover the true underlying activation patterns and thus better characterize brain functioning.
Two years later, the field has continued to see great success in predictive modeling, as the results of the 2006 and 2007 Pittsburgh Brain Activity Interpretation Competition demonstrate, convincing most neuroscientists that there is tremendous potential in the decoding of brain states using statistical learning. Along with this realization, though, is a recognition of the limitations inherent in using black box methods for drawing neuroscientific interpretations. The primary challenge now in the field is how best to exploit statistical learning to answer scientific questions by incorporating domain knowledge and embodying hypotheses about various cognitive processes.
Further advances in the field will require resolution of many open questions, including the following:
1) Variability/Robustness: to what extent do patterns in fMRI replicate across trials, subjects, tasks, and studies? To what extent are processes that are observable through the fMRI BOLD response truly replicable across these different conditions? How similar is the neural functioning of one subject to another?
2) Representation: the most common data representation continues to consider voxels as static and independent, and examples are i.i.d.; however, activation patterns almost surely do not lie in voxel space. What are the true, modular activation structures? What is the relationship between similarity in cognitive state space and similarity in fMRI activation space? Can causality be inferred from fMRI?