Evaluating Lowpass Filters for fMRI Temporal Analysis

Temporal filtering of fMRI images is an important part of the process of preparing the data for clinical examination. We are able to reduce noise accumulated from both exogenous and intrinsic sources during scanning by applying a filter to the time series. This paper summarizes the testing methods and results found from the evaluation of linear and non-linear temporal filtering methods used in functional MRI analysis. Evaluation is carried out on data sets which have been generated by combining real resting data with synthetic BOLD activations and subsequently filtered in the temporal domain. In addition to examining commonly used linear and non-linear filters, comparison is made with a temporal filter which uses the SUSAN noise reduction algorithm to achieve more robust signal recovery than conventional fMRI analysis techniques. Results show, somewhat surprisingly, that in general the matched filter used widely by default may be better than existing non-linear filters.