Multinomial inference on distributed responses in SPM

In this work, we propose statistical methods to perform inference on the spatial distribution of topological features (e.g. maxima or clusters) in statistical parametric maps (SPMs). This contrasts with local inference on the features per se (e.g., height or extent), which is well-studied (e.g. Friston et al., 1991, 1994; Worsley et al., 1992, 2003, 2004). We present a Bayesian approach to detecting experimentally-induced patterns of distributed responses in SPMs with anisotropic, non-stationary noise and arbitrary geometry. We extend the framework to accommodate fixed- and random-effects analyses at the within and between-subject levels respectively. We illustrate the method by characterising the anatomy of language at different scales of functional segregation. ►Given a fixed partition of an SPM (eg anatomical altas/independent ROIs), which regions are relatively active?►This method identifies 'active' regions as containing more events (e.g. blobs) than expected by chance►The method is sensitive to different features of an SPM, relative to conventional analyses