Generalized fMRI activation detection via Bayesian magnitude change point model

In the human brain mapping field, virtually most existing fMRI activation detection methods, such as the general linear model (GLM), have assumed that the fMRI signal magnitude should follow the alternations of baseline and task periods. However, our extensive observation shows that different brain regions or networks exhibit quite dissimilar temporal activation patterns. Inspired by this observation, we develop a novel Bayesian magnitude change point model (BMCPM) that simultaneously considers the group-wise fMRI signals of corresponding cortical landmarks across a population of subjects and optimally determines the change boundaries. Then, these detected group-wise consistent magnitude change points are clustered into various patterns of temporal and spatial activations, which are named generalized activations here. The methods have been applied on a working memory task-based fMRI dataset and revealed complex and meaningful generalized brain activation patterns.