How clear is our view of brain activity? : BCF researchers publish article in "Human Brain Mapping", demonstrating the large influence of filtering algorithms on results in brain imaging

In the current issue of Human Brain Mapping, Tonio Ball of the Bernstein Center Freiburg and colleagues from the universities of Oldenburg, Basel and Magdeburg demonstrate how variable the results of imaging techniques like functional Magnetic Resonance Imaging (fMRI) can be – depending on the way how the original data are filtered. The use of filtering algorithms is indispensable in order to separate meaningful information from inherent noise that is part of every data set. These filters have different “mesh sizes” or widths, and are indispensable in the first place to reveal activity patterns that span different scale sizes. In most cases, only a filter of one specific width, which differs from study to study, is employed. 

Tonio Ball and colleagues systematically investigated the influence of the mesh size of these filters on the resulting imagery of brain activity. They conducted an experiment during which test persons had to rate music by pressing a button while lying in an fMRI scanner. During this task, brain regions responsible for hearing, vision, and arm movements were active. The scientists treated the gained data with filters of different widths and found surprising results: The filters had a marked influence on the outcome of the brain scan analyses, showing increased brain activity in one region in one case, and in a different region – in the other. Even smallest changes in filter width led to areas of the brain appearing to be either active or inactive. This effect can ultimately lead to widely disparate interpretations of such a scan. Tonio Ball and his colleagues therefore stress the importance of taking into account the effect of filtering in future interpretations of fMRI studies. This way, scientists won’t run the risk of inadvertently skewing our view of the brain.

Interfering filters: In the identical original data set, a region of the brain (circled) appears active in one case, but not in the other – solely dependent on the “mesh size” of the used data filter (BCF/University of Freiburg). 

Press release by the University of Freiburg (German)