Preprocessing in Functional MRI

Descombes X.1, Kruggel F.2

1 INRIA, Sophia Antipolis, 2 University of California, Irvine

The preprocessing stage improves the signal quality before signal detection, including: artifact detection, baseline correction, movement correction, and image restoration. Gaussian filtering is commonly applied to increase the signal to noise ratio in fMRI preprocessing. However, this filter spoils the signal and introduces artifacts (loss of detail, blurring of edges, fusion of neighboring objects, displacement of objects). Several alternatives have been proposed for signal restoration. Markov Random Fields are very popular in this context because statistical tools to estimate parameters and to optimize the model are well defined and efficient. A priori properties are modeled by interactions between neighboring pixels in time and space. Using well-adapted interactions, noise can be reduced without introducing blurring.


Comparison of standard evaluation (left), Gaussian filtering (center) and image restoration (right) on the results of a statistical evaluation of the same fMRI experiment.


Descombes X., Kruggel F., von Cramon D.Y. (1998) fMRI Signal Restoration using a Spatio-Temporal Markov Random Field Preserving Transactions. NeuroImage 8, 340-349.

Descombes X., Kruggel F., von Cramon D.Y. (1998) Spatiotemporal fMRI Analysis using Markov Random Fields. IEEE Transactions on Medical Imaging 17, 1028-1039.

Kruggel F., von Cramon, D.Y. (1999) Comparison of Filtering Methods for fMRI Datasets. NeuroImage 10, 530-543.

Kruggel F., Descombes X., von Cramon, D.Y. (1998) Preprocessing of fMR Datasets. In: Vemuri B. (ed.), Workshop on Biomedical Image Analysis (Santa Barbara), pp. 211-220. IEEE Computer Society Press, Los Alamitos.