12th MR Methods Meeting
14. September 2017, 15:30 Uhr, DZNE, Haus 64, Raum 121
Title: "A new non-parametric and threshold-free framework for statistical inference of fMRI data"
Speaker: PD. Dr. Gabriele Lohmann
Group: Max-Planck-Institut für biologische Kybernetik Tübingen
Abstract: One of the principal goals in functional magnetic resonance imaging (fMRI) is the detection of activations in the human brain. For this purpose, statistical inference methods are needed in order to separate signal from noise. Recent publications have shown that inflated false positive rates and lack of statistical power pose very serious problems in fMRI so that better methods for statistical inference are urgently needed. Furthermore, there are currently no statistical methods that specifically meet the requirements of high-resolution imaging data acquired at ultrahigh field MRI scanners (>7 Tesla). Here we propose a new general framework called LISA to address this demand. Compared to widely used other methods, we show that it has higher statistical power without inflating the false positive rate so that smaller sample sizes are required to detect true effects reliably. It is applicable in a wide range of scenarios such as standard group analysis, single subject analysis or multivariate pattern analysis. Furthermore, it preserves spatial precision making it suitable for ultrahigh-resolution imaging. LISA is non-parametric and threshold-free.
