MEDx Frequently Asked Questions - Other Issues

Q: When do I choose to use the linear temporal filter?

The linear temporal filtering module allows you to filter out specific frequencies in time-series data while preserving other frequencies. Therefore, it will work best for experimental paradigms where you can distinguish between frequency components to be preserved and frequency components to be filtered out. If you can determine either the frequency of an effect of interest (e.g., the frequency of an on-off stimulus) and/or the frequency of an effect of no interest which you wish to filter out (e.g., periodic physiologic noise), and if these two frequencies are distinct, you may be able to use linear temporal filtering effectively.

On the other hand, if for some reason you cannot distinguish between frequency components of interest and frequency components for no interest, linear temporal filtering may not be appropriate for your data. For example, if your on-off stimulus had a frequency of 0.1 Hz and some physiologic noise in your data also had a frequency of 0.1 Hz, you will not be able to use linear detreading filtering to filter out the noise, because the signal due to the input stimulus will be filtered out as well. As another example, if your input stimulus isn't periodic and is more or less random, then its frequency content will be spread out across the frequency spectrum, and filtering out any set of frequencies will also tend to filter out the signal due to the input stimulus.

With those caveats, you can use the linear temporal filtering module to remove low-frequency trends such as linear drift in time-series data while preserving the signal due to some periodic input stimulus. To do this, load in a group of image volumes, and bring up the toolbox. Select Functional --> Time Series..., and in the Time Series dialog box which appears click on the OK button; this assigns a temporal order to the image volumes. Then select Functional --> Filtering... to bring up the Filtering dialog box. Click on the Temporal Filtering tab. Set the Input Group to the name of the group of image volumes your loaded in (you can either type it in or browse MEDx pages by clicking on the Select... button on the right). Set the Filter Type to High Pass. Set the Family to Butterworth. Set the TR to the TR of your experiment, measured in seconds.

The value to type in the entry field labeled "High pass (sec)" will depend on the period of your particular on-off stimulus. As a rough rule of thumb, take the period length in seconds of your on-off stimulus and multiply it by two. You can use the result as the value to type in the "High pass (sec)" entry field. All frequency components having a period greater than this number will be attenuated relative to frequency components having a period less than this number. That is to say, high frequency components will tend to be preserved while low frequency components will tend to be diminished.

Other parameters under Options are set to reasonable defaults which you can use. Click on the Preview button to see frequency and temporal plots of the Butterworth filter you've designed, before (in green) and after (in red) windowing. The post-windowing filter is the one which is actually realizable given the constraints of sampled, quantized data on a digital computer. The Preview window is helpful in designing your filter to ensure that there isn't anything funny about it. Then click on the Apply button. The temporal filter you designed will be applied to your time series data, and the MEDx page holding that data will be replaced with the filtered data.

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