- FMRI analysis defaults
- A) Non-SPM Preprocessing - moving from raw data to SPM images
- B) SPM Preprocessing - from SPM-based raw data to smoothed data ready for analysis
FMRI analysis defaults
This page describes the default analysis steps for WBIC data. These are data collected before June 2006 on the WBIC Bruker 3T scanner.
The description here is of the steps you will need to take in the GUI to analyze in SPM, with some discussion of the AnalysisPrinciples involved. Personally (MatthewBrett) I would recommend that someone new to imaging, or new to SPM, does a few analyses this way to get themselves used to the interface and the processing steps involved.
All the preprocessing steps (from data retrieval to smoothing) can now be carried out automatically, via Rhodri Cusacks AA scripts. It is highly recommended that you preprocess using this method when you are used to the analysis.
If you want to read this page on a printed copy, I suggest you also print:
(Bruker data) Getting WBIC data page
(Bruker data) Bruker format page
(Bruker data) pvconv page
Identifying the cerebral commissures
You will get useful additional information from the SPM99 manual, Kalina Christoff's preprocessing webpages (which don't precisely apply to our data, but are very well explained), and from searching the SPM mailing list archives for specific topics, such as "slice timing" etc.
A) Non-SPM Preprocessing - moving from raw data to SPM images
There are various steps required to convert the raw data once it has arrived at the CBU to images that SPM will apply preprocessing steps on. Initially, if the data have been acquired on the new standard EPI protocols (for instance, those with a 1.1s or 1.6s TR), then we also need to reconstruct the data ourselves. Following this, the reconstructed data need to be converted to Analyze format, so that SPM can read the images. Finally, the 4D Analyze images created need to be split into 3D images (without a time component per image).
Each of these stages can be accomplished by in house software at the command prompt (see below). However, there is a useful program that creates a batch file to automatically perform all the stages above (or a subset of these as the need arises), as well as a few other common steps carried out at this stage (writing appropriate directories out to place the different types of files, copy phase maps and structural images to useful locations, etc.). See Daniel Bor's MakefMRIBatch page for details about how to run this program and the batch file it generates.
Getting the data
If you have Bruker data from the Wolfson Brain Imaging Centre, see the Bruker data page. If you followed the steps in these pages, you should now have 4D (X,Y,Z, time) Analyze format images for the functional runs, and 3D images for the structural etc scans.
Making images for SPM
There are two steps needed before processing the data with SPM; you will need to convert the 4D functional images into a series 3D volumes, with one volume for each time point. Then you may need to discard a few dummy scans from the beginning of the time series. Deleting dummy scans is useful because there is a large signal change in the first few images in a functional dataset, as the tissue reaches a steady state of radiofrequency excitation. We usually discard 15 seconds of data (5 scans with a TR of 3, 8 scans with a TR of 2).You can make the 3D volumes and discard the dummies with the ana4dto3d program. If you do not use ana4dto3d to throw away the dummies, and you are going to discard the first few images, now would be a good time to move them out of the directory with the other images, or delete them if you are brave.
If you have used ana4dto3d, you might have created a mean and variance image for the functional runs. It is a good idea to inspect these using SPM or MriCro, to look for reconstruction or data acquisition problems (funny brain shape, strange effects on lower image slices, high ghost variance, radiofrequency artefact). See the FMRI diagnostics page for details.
Now, you can start SPM 99 / SPM2...
B) SPM Preprocessing - from SPM-based raw data to smoothed data ready for analysis
After creating 3D Analyze format images, a number of corrective steps need to be applied before statistical models can be applied:
See SliceTiming for a discussion of this topic.
First choose "Slice timing" which corrects for different slices of the volume being collected at slightly different times. The MRI scanner at the WBIC performs "interleaved" echo-planar images, to reduce the effect adjacent slices have on each other. First the odd slices are collected, then the even ones. In the slice timing prompts, first select the images in the order they were acquired. In the CBU, we have edited the slice timing routine to add an option to the ensuing list box that is correct for WBIC data; this will be called 'Interleaved (1 3 .. 2 4 ..). Select this option if it is there. If not, choose "user specified" from the menu. Assuming 40 slices, typing "1:2:39 2:2:40" in the text input box, and pressing return will step from 1 to 39 in steps of 2 and from 2 to 40 in steps of 2, giving the order in which the slices were collected. Choose 1 as reference slice (see SliceTimingRikEmail), type in the TR, and the TA. The TA is the Time of Acquisition. Somewhat counterintuitively, the TA is taken as the time between the start of the first slice acquisition, and the start of the acquisition of the last slice. Thus, the default TA, which assumes there is no gap at the end of the volume acquisition, is the TR - TR/(number of slices in the volume). WBIC EPI scans have a gap between the end of the acquisition of the last slice and the beginning of the acquisition of the first slice of the next volume. You need to know what this gap was for your scans. The actual time to acquire one slice is the best value to use to give the TA value (mutliply the slice acquisition time by [no of slices-1] ). The actual slice acquisition time for a Bruker / WBIC image is given in the Bruker text files; getting the repetition times is different for the Bruker sequences (as used by default until early 2002) and Christian's EPI sequences (as used by default since then).
- For the Bruker sequences, you will have an 'imnd' file, in the session subdirectory of the raw data directory. If you have used pvconv to convert your data, the text files will all have been saved with the converted image, as a file "image.brkhdr", where "image.img" is the image file name. The value you want is from the field
- which gives the time taken to acquire one slice in milliseconds. Thus, if EPI_slice_rep_time = 115, and you have 20 slices, the TA needs to be 0.115 * 19.
- For Christian's EPI sequence, it is a little more complicated; if "srt" is the slice repetition time:
# The first is the default situation: if ##CBU_SliceTiming == 'Uniform': srt = (##PVM_RepetitionTime - ##CBU_TriggerPulseLength) / ##PVM_SPackArrNSlices # This second is much less common if ##CBU_SliceTiming = 'Bunched': srt = ##PVM_MinRepetitionTime / ##PVM_SPackArrNSlices
Now you need to wait for a bit. The slice timing corrected images have the same name as the originals, but with an "a" prepended to the file name. NB This scheme may well not be correct for FMRI data acquired from other imaging centres - please ask for details.
Image orientation and headers
You may want to reorient your images before you continue the analysis. This has two advantages; first, if you do any analysis on the data from this subject without spatially normalizing, and you have oriented your images roughly to the orientation that SPM expects, then the results will be approximately correctly displayed on the SPM glass or cartoon brain displays. Second, this will give the SPM spatial normalization a good place to start from, and will therefore be less prone to error. The orientation for these purposes does not need to be very accuratem so a few millimetres error will not be important. The orientation you are trying to achieve is shown in the page describing the cerebral commissures. You may want to keep the images on this page displayed somewhere for reference.In this section we will be creating or modifying .mat files to reorient the images. Another approach is to edit the origin field in the headers (.hdr files) of the Analyze format images. I suggest you avoid this, as the origin field is much less flexible than the .mat file, and because the origin field is ignored by SPM when the image has a .mat file, which can cause confusion.
Now, click the display button, and choose the first image in the series, of the slice timing corrected - a*img images.
If you have used a recent version of pvconv to convert your WBIC (Bruker) data, you will already have .mat files for your images, which will have oriented the data to neurological (L=L) convention according to the position of the patient in the scanner. If you didn't use the new version of pvconv, or you have deleted the .mat files for the images, and you have saved the Bruker text files, you can recreate the appropriate .mat files using brk2mat.pl. If you have used pvconv, the orientation will be approximately correct when you start (if the subject's head was positioned normally).
If you haven't used pvconv, then your images will probably need some major reorientation in order get them near to the correct neurological orientation. Flip as desired, using the X/Y/Z zooms. For most current EPI images from the WBIC acquired in axial sections (standard) the images will need to be flipped left-right (X) and front-back (Y) to conform to neurological orientation. Do this by applying a -1 resize (zoom) in X and Y in the parameter boxes. On current WBIC images a small bright spot (a bottle of water) can sometimes be seen on the right hand side of the images. Now you have the orientation more or less right, you may want to tume it to better match the orientation of the SPM template. Try comparing your images to the T2 template on the cerebral commissures page. Start by setting any yaw needed (Z rotation, shaking head type movements), then set roll (Y rotation, moving head to touch ear on shoulder), and pitch (X rotation, nodding). Then use the X Y Z translation boxes to set the blue lines to cross at the anterior commissure.
Don't click on the images during this process, as you will move the blue lines, to some arbitrary point. If you need to reset the blue lines to the current origin, type 0 0 0 in the mm coordinates box in the box on the left of the graphics window, or click on the little embossed line above the coordinate entry boxes.
When you have set all the zooms, rotations and translations to your liking and not before, record the values you are using somewhere for further reference, and then choose "Reorient images" at the bottom of the graphics window. Select all the images in the slice timing corrected - a*img series. If you have used pvconv to convert the data, also select any other images for this subject collected in the same session, as these will be roughly coregistered. When you select Done, this will apply the affine transformations that you have just designed, by adding to, or creating .mat files for all the images.
Check the transformation was as you expected by clicking on "Display" again, and selecting the first image in the slice timing corrected series. The blue lines should cross at the AC, and the image should be in the correct orientation.
Realignment of subject motion
See the MotionCorrection analysis page. If you are following the recipe on the page you are reading now, you will want to motion correct the slice-time corrected images.
EPI images at 3T (e.g. from the WBIC or Oxford) are distorted relative to the structural scans. In order to be able to coregister the EPI images to the structural images properly, we need to first undistort the EPI images to better match the structural in shape. There may be advantages to doing this even if you do not want to coregister to the structural; see Rhodri's undistortion pages.
See ProcessingCoregistration. Note that, if you have used pvconv to convert your images from Bruker format, images in the same scanner session should already be coregistered using the automatically generated .mat files from the scanner orientation information.
Time to warp. See ProcessingNormalization