Spatial Smoothing

Here, we're filtering in the spatial rather than temporal domain. Also, we'll be using a low-pass Gaussian filter rather than a high-pass Butterworth filter.

(1) In the MEDx folder, select Toolbox --> SPM Modules --> Smoothing.... This pops up the SPM Smoothing dialog box.

(2) Set Input Group/Volume to New Group, which has now been temporally filtered.

(3) Under Gaussian FWHM, set X, Y, and Z to 8. These are smoothing parameters (full-width at half-maximum) in millimeters.

(4) Click on the OK button. This will smooth the 84 images, creating a new group (unlike Linear Temporal Filtering, which overwrote the previous data).

(5) The MEDx folder is getting crowded again. To keep MEDx from getting too slow, and to give you another chance to practice writing out a group of images as AVW format, use the SaveNSetAVW.tcl script to write out the spatially smoothed data to the tutorial directory. (If you don't remember how to do this, review steps 9 - 11 of Motion Correction.) Set the Root Name of these images to

to indicate that the images have now been spatially smoothed, temporally filtered, and motion corrected. Use the same X Scale, Y Scale, and Z Scale as before (because the voxel sizes have not been changed yet). Watch the images as they are displayed during the disk write, to make sure that they look "okay". They should look blurrier than the original images, because they have now been spatially smoothed.

(6) After the images have been written out to disk, exit MEDx.

Note that there is another way to do spatial smoothing in MEDx, under Toolbox --> Functional --> Filtering... --> Gaussian Smoothing; see Section 31.7.1 of Chapter 31. This alternative method is what is actually used in HS2. I believe that the SPM96 implementation is more efficient, since it uses a Fourier trick (Convolution Theorem) to do the smoothing. After this tutorial, why don't you try this alternative method on a fresh MEDx folder and see which method really is faster? Let me know the results of your comparison!

You are learning, grasshopper! Proceed to Spatial Normalization!

See also this stand-alone MEDx tutorial. Also see Section 32.5 of Chapter 32 of the MEDx User's Guide.

Return to tutorial main page