MEDx Frequently Asked Questions - Other Issues

Q: When to Use Spatial Smoothing

Spatial smoothing should be applied to data that is locally noisy. A good example of this is O15 PET data. I have found that smoothing can also be useful in analyzing fMRI data. This of course will reduce the effective resolution but may lead to better statistical results. In terms of the FWHM, typically I use a 10mm FWHM but this depends on how noisy the data is. For example, with PET data you should probably use a 15-20mm FWHM. A general rule of thumb for the kernel size is

(2*FWHM)/(voxel dimension)

This will set the kernel dimensions to be approximately 1 standard deviation or roughly 95% of the filter will be inside the kernel.

Temporal smoothing is sometimes useful but greatly depends on the characteristics of your data across time. You can use the voxel surfer (Temporal Display) to examine how stable your data is within task epochs after performing a statistical operation such as the Mean under Within Group. Smoothing in the temporal domain can actually hurt the statistical power by blurring the transitions between task epochs. Ideally you would like to apply the smoothing within task epochs rather than across the entire series. The wavelet based temporal filtering is actually a way of doing this. The default Type - "Visu Shrink" is a good filter to use.

There is of course an interaction between the above smoothing operations and intensity normalization. It is probably best to perform Spatial smoothing first, followed by intensity normalizaiton, and then the wavelet filtering. The idea being that you want to correct for local (both in the sense of spatial and temporal) spurious variations in intensity using the spatial smoothing first. This is then followed by removing global variations or signal drift based on data that is already locally congruent. The wavelet filtering which works on the data from a single voxel across time is the applied to data in which global or linear variations across time have been removed.

Other FAQ Topics

If you have a question that you would like to see addressed in our list of Frequently Asked Questions, please contact customer support.

Back to FAQ Home