table of contents
LIPSIA     Spatial and temporal filtering
vpreprocess

Spatial and temporal filtering can be applied to fMRI data using the program 'vpreprocess'. Spatial filtering is often used to improve the signal-to-noise-ratio (SNR), and to reduce the inter-subject variability. Temporal filtering is used to remove low frequency drifts in an fMRI time series. Temporal filtering is implemented in 'vpreprocess' computing the Fast Fourier Transform (FFT). Specified frequencies are removed in the frequency domain. Finally, an inverse FFT transforms the image back into the original space.

Spatial filtering
'vpreprocess' can be used to perform spatial smoothing using a Gaussian filter. For example:

vpreprocess -in AC1T.v -out bAC1T.v -fwhm 7

This call performs a spatial Gaussian smoothing using a filter size of 7 mm FWHM. Gaussian smoothing should especially be used for multi-subject studies. Per default, spatial smooting is disabled. You have to specify a standard deviation (with parameter fwhm) to enable it.

Remark: If the input file contains 3D functional images (e.g., obtained by 'vfunctrans'), spatial filtering is applied in 3D space using a 3D Gaussian kernel. Otherwise, 2D spatial filtering is performed for each slice separately.

Temporal filtering: High pass filter

If '-high' is set to a positive number, 'vpreprocess' performs a high pass filter. Note that the cutoff is specified in seconds and not in Hertz. In the example below, a full cycle of stimulation lasts 48 seconds. Therefore, a cutoff of 60 seconds (60 > 48) can be used as follows:

vpreprocess -in AC1T.v -out bAC1T.v -high 60

The following figure shows the effect of a high pass filtering:


Temporal filtering: Low pass and band pass filtering
For longer block designs, it is sometimes useful to additionally perform a low pass filter to remove high frequencies that might be due to noise. A rule of thumb is to use a cutoff of less than 3 seconds (half the length of the delay of the hemodynamic response). The low pass filter is specified by the command line option '-low'. In our example, we use a low pass filter of 3 seconds. Frequencies higher than 1/3 Hz are suppressed.

vpreprocess -in AC1T.v -out bAC1T.v -low 3

Band pass filtering can be performed using both options '-high' and '-low'. Frequencies outside the interval '-high' and '-low' are suppressed.

vpreprocess -in AC1T.v -out bAC1T.v -high 60 -low 3

Caution: Low pass filtering should only be used if the experiment consists of very long trials.

Instead of a band pass filter, a band stop filter can be applied using the option '-stop'. Using that option, frequencies between '-high' and '-low' are suppressed. This can be used for removing all frequencies within a special frequency window. The following command-line option can be used to remove all frequencies between 1/10 Hz and 1/8 Hz:

vpreprocess -in AC1T.v -out bAC1T.v -high 10 -low 8 -stop true

'vpreprocess' can be called using a signal threshold with the option '-minval'. As long as there are no negative values in the images, it is safe to use the default value '-minval 0'. Computation effort can be reduced choosing an appropriate 'minval' that depends on the scanner. Time series showing an average less than 'minval' are excluded from the analysis.

Parameters of 'vpreprocess'
-help
Prints usage information.
-in
Input file. Default: (none)
-out
Output file. Default: (none)
-fwhm
Spatial filter: FWHM in mm. Default: 0
-high
Temporal Filter: Cutoff for high pass/stop in seconds. Default: 0
-low
Temporal Filter: Cutoff for low pass/stop in seconds. Default: 0
-stop
Temporal Filter: Stop instead of pass filter. Default: false
-minval
Signal threshold. Default: 0


Max Planck Institute for Human Cognitive and Brain Sciences. Further Information: lipsia@cbs.mpg.de
Copyright © 2007 Max Planck Institute for Human Cognitive and Brain Sciences. All rights reserved.