|
Abstract:
|
We apply ideas from statistical process control and change point theory to model slowly varying processes with uncertain onset times and durations. The analysis uses activity collected during a baseline period to estimate noise characteristics in the fMRI signal response and make inferences on whether, when, and for how long subsequent activity deviates from the baseline level. We tailor existing exponentially weighted moving average (EWMA) models to individual fMRI time courses, then develop a group analysis using a hierarchical model, which we term Hierarchical EWMA. Once a systematic deviation from baseline has been detected in the group, change point methods are used to estimate the time of change and recovery time. Variations across the brain in the onset and number of change points and in the duration of a shift away from baseline activity are used to classify brain regions.
|
