Single-trial variable model for event-related fMRI data analysis

Most methods for fMRI data analysis assume that the hemodynamic responses (HRs) across similar experimental events are same. This assumption is not appropriate when HRs vary unpredictably from trial to trial. Here, we introduce a new method for fMRI data analysis. The main features of the proposed method are as follows: 1) The trial-to-trial variability is modeled as meaningful signal rather than assuming that the same HR is evoked in each trial; 2) Since the proposed method is a constrained optimization based general framework, it could be extended by utilizing prior knowledge of HR; 3) The traditional deconvolution method can be included into our method as a special case. A comparison of performance on simulated fMRI datasets is made using the general linear model, the deconvolution method and the proposed method with receiver operating characteristic (ROC) methodology. In addition, we examined the effectiveness and usefulness of our method on real experimental data.