Adaptive statistical parametric mapping for fMRI

Brain activity is accompanied by changes in cerebral blood flow (CBF) and the differential blood oxygenation that are detectable using functional magnetic resonance imaging (fMRI). The process of identifying brain activation regions can be facilitated by estimating the hemodynamic response function (HRF). There have been some remarkable new developments in statistics to handle this problem. In this paper, we introduce a novel procedure which is capable of adapting itself to any of the existing methods by improving its performance through the application of a penalized smoothing technique. Using a computer experiment and a real fMRI data set, the proposed procedure is assessed by comparing its performance very favorably to the popular SPM based method.