Listed below are talks from the OHBM 2012 "Where's Your Signal? Explicit Spatial Models to Improve Interpretability and Sensitivity of Neuroimaging Results" workshop.
Thomas Nichols, University of Warwick
A typical fMRI study is a massive endevour: 100's of man-hours to prepare paradigms and train subjects; costly scanner time; and laborious data analysis to process gigabytes of image data. Yet the crucial result of a study is a list of x,y,z atlas coordinates of activation, a dataset that can be recorded on a Post-it note. Indeed, this coordinate list is the only information incorporated into a typical meta-analysis. Given the vital importance of these x,y,z coordinates, why are they never reported with confidence intervals? Wouldn't we expect some tasks to produce activation with greater spatial variability than others types of tasks? The problem is that, until very recently, there simply were no methods to quantify the uncertainty in the spatial location of activations.
The purpose of this workshop was to review the emerging work in the explicit spatial modeling of neuroimaging data. Thomas Nichols provided a short overview, itemizing the limitations of standard mass univariate models and reviewing the potential of spatial models. Alexis Roche discussed a model selection approach to fMRI that uses a hierarchical spatial generative model, resulting in inference on a network of regions while accounting for uncertainty in location. Timothy Johnson presented point process model for metaanalysis and multiple sclerosis lesion data; in addition to be more interpretable, this model provide superior classification performance relative to univariate (naive bayes) methods. Finaly, Sam Gershman showed how cognitive scientists are using spatial models with fMRI, with work that uses latent topographic sources to decode semantic representations during a memory task.