Activation detection in functional MRI based on non-separable space-time noise models

Detecting activated regions in the human brain by cognitive tasks is a significant task in the data analysis using functional MRI (FMRI). To create a detection statistic for activation, noise models under two assumptions; 1) spatial independence and 2) space-time separability have been dominantly used in the FMRI data analysis. In this paper, we propose a novel detection statistic derived from noise models with spatiotemporal correlation and without space-time separability. In order to obtain a sufficiently flexible class of noise models for nonseparable space-time processes, an unusual noise modeling based on truncated cepstrum expansion is suggested. Developed methods are applied to a human dataset.