Robust and scalable methods in fMRI statistical analysis

E. Cabana Garceran del Vall, R. E. Lillo, H. Laniado

This work deals with the functional Magnetic Resonance Imaging (fMRI) statistical analysis. The classical approach based on GLM is not a robust methodology, a very important fact with this kind of data that presents several acquisition artifacts and signal noises. In order to gain in robustness, while maintaining an efficient computational time, we propose a method based on the detection and removal of multivariate outliers in the data, using a robust Mahalanobis distance. This measure is defined in terms of robust estimates of location and scatter based on the notion of “shrinkages”, a very well-known concept in financial statistics and portfolio optimization, that has not been applied to the neuroimaging field before. We present some good preliminary results with simulated fMRI data within a comparison with other existing methodologies in the statistical literature.

Palabras clave: robust regression multivariate outlier detection fMRI statistical analysis

S. Velilla Cerdán

G. Mateu-Figueras, P. Daunis-i-Estadella, J. Martín-Fernández

E. Boj del Val, T. Costa Cor