A. Procedure : single subject
B. Source code:spm_slice_timing.m
C. Notes
D. Mail list

A. Procedure : single subject

•Select images to acquisition correct : fmri*.img (done)

•Select sequence type...

- ascending (first slice=bottom)

- descending (first slice=top)

- interleaved (first slice =top)

- user specified <x>

•Order of slices (1=bottom) : 1 3 5 7 9 11 13 15 17 19 2 4 6 8 10 12 14 16 18 20

•Reference Slice (1=bottom) : 19

Slice timing working...

B. Source code:spm_slice_timing.m

% function spm_slice_timing(P, Seq,refslice,timing)

% INPUT:

% P nimages x ? Matrix with filenames

% Seq slice acquisition order (1,2,3 = asc, desc, interl)

% refslice slice for time 0

% timing additional information for sequence timing

% timing(1) = time between slices

% timing(2) = time between last slices and next volume

%

% If no input is specified the function serves as a GUI

%

% OUTPUT:

% None

C. Notes

1. Temporal resampling of fMRI time series .

2. To correct differences in slice timing of 2D acquisitions .

3. A button has been added for correcting fMRI time series data for the differences in image acquisition time between slices. Note that there is nothing at all spatial about this procedure! It is included here because it is an optional pre-processing step that should improve results - especially for event related studies.

4. To correct differences in image acquisition time between slices. This routine is intended to correct for the staggered order of slice acquisition that is used during echo planar scanning. The correction is necessary to make the data on each slice correspond to the same point in time. Without correction, the data on one slice will represent a point in time as far removed as 1/2 the TR from an adjacent slice (in the case of an interleaved sequence).

5. This routine "shifts" a signal in time to provide an output vector that represents the same (continuous) signal sampled starting either later or earlier. This is accomplished by a simple shift of the phase of the sines that make up the signal. Recall that a Fourier transform allows for a representation of any signal as the linear combination of sinusoids of different frequencies and phases. Effectively, we will add a constant to the phase of every frequency, shifting the data in time.

6. Shifter - This is the filter by which the signal will be convolved to introduce the phase shift. It is constructed explicitly in the Fourier domain. In the time domain, it may be described as an impulse (delta function) that has been shifted in time the amount described by TimeShift. The correction works by lagging (shifting forward) the time-series data on each slice using sinc-interpolation. This results in each time series having the values that would have been obtained had the slice been acquired at the beginning of each TR. To make this clear, consider a neural event (and ensuing hemodynamic response) that occurs simultaneously on two adjacent slices. Values from slice "A" are acquired starting at time zero, simultaneous to the neural event, while values from slice "B" are acquired one second later. Without corection, the "B" values will describe a hemodynamic response that will appear to have began one econd EARLIER on the "B" slice than on slice "A". To correct for this, the "B" values need to be shifted towards the Right, i.e., towards the last value.

7. This correction assumes that the data are band-limited (i.e. there is no meaningful information present in the data at a frequency higher than that of the Nyquist). This assumption is support by the study of Josephs et al (1997, NeuroImage) that obtained event-related data at an effective TR of 166 msecs. No physio- logical signal change was present at frequencies higher than our typical Nyquist (0.25 HZ).

8. NOTE WELL: This correction should be the first performed (i.e., before orienting, motion correction, padding, smoothing, etc.). Additionally, it should only be performed once!

9. NOTE ALSO: The routine below assumes the data are interleaved. Since the images out of the vision have already been arranged in proper slice order, this routine will not work on non-interleaved slices.

D. Mail list

1. Question
The spm_slice_timing routine in SPM99b writes out a*.mat files. As I understand it, the time shifted data is written into the a*.img files.

Why does spm_slice_timing write a*.mat files?

1. Answer

When SPM writes out an image, it also creates a ".mat" file when the orientation and position of the image can not be exactly represented by the voxel size and origin fields of the ".hdr" files. When the origin field is empty, the origin is assumed to be in the center of the volume. If this is not at the centre of a voxel, then SPM will write out a ".mat" file for the image. It is just a little feature of SPM.

The ".mat" files represent the orientation of the images to SPM and encode the relative positions of the images as well as starting estimates that can be used for spatial normalization. This allows MR images to be acquired in any orientation, and still be spatially normalized, coregistered etc by SPM. If your conversion routines

from the native MR scanner format to Analyze format do not include writing out a ".mat" file, then you can re-orient your images via the "Display" button. This involves trying out different translations and rotations until you get your image in the right orientation and position (transverse image at the bottom etc, with the AC corresponding to a point near [0 0 0]mm.). The entered transformation can then be

applied to all the other images of the series via the "Reorient images..." button.

2. Question
I would like to use slice timing on an fMRI time-series with a TR of 2 seconds and 24 slices The aquisition sequence was as follows:

slice: 1 5 9 13 17 21 2 6 10 14 18 22 3 7 11 15 19 23 4 8 12 16 20 24

Which would be the best reference slice? Spatially slice 12 is in the middle of the scan, but it is the fourth last of all slices to be acquired. In the time domain, slice 22 would be in the middle, but spatially it is almost on the very top of the scan.

How much signal has a slice to contain, to be a good reference slice?

2. Answer
1. The middle slice in time (slice 22) would be best. It doesn't matter "how much signal is in the reference slice.This is because the timing is most accurate at time points close to the reference slice, so you minimize the time difference over all slices in the volume.

Note also that if you have an a priori hypothesis about a particular region, and you know which plane it is in, then I guess that you would use that plane as the reference. Of course, this is a little tricky with your data, but it should be straightforward for simple ascendingly or descendingly organised images.

3. Question
Should the slice time correction be done before or after the realignment?"

3. Answer

I think this depends. If you have simple ascending or descending data, then the time difference between adjascent slices should be relatively small. In this case, I guess you would do the realignment before the slice time correction. If you have interleaved slices, then the slice timing correction should probably be done before realignment (this assumes that most of the movement is a gradual drift and the position of one image volume should be close to the position of its neighbours. )

Another thing to consider is the length of your epochs. If you have longer epochs, then the slice timing should be less critical.

4. Question
Which slice would you suggest to choose as reference having 20 slices for each volume (block-design)?

4. Answer

Assuming negligible power above the Nyquist limit, the choice of reference slice doesn't matter (in the sense of the resulting SPMs; the choice of reference slice does of course affect the appropriate stimulus onset times required). There is some debate about whether the reference slice matters in the presence of significant supraNyquist power. Note however that in a blocked design, slice timing correction is not so critical, given that the time constants of blocks of activation are typically much larger than the TR.

5. Comment
5.1: Order of motion correction and slice correction

The question was raised as to what is the appropriate order for these manipulations of the data. I believe that the answer will always be that the slice correction step should be performed before the motion correction step. The rationale for this assertion is that motion correction disrupts the known absolute time at which a given data point was acquired. It is the fMRI slice in absolute space (and not the actual brain within that slice) that needs to be shifted in time.

5.2: Selection of reference slice

As Rik Hension noted, it doesn't matter what the slice actually contains as regards its use as a reference (the slice could cover only air!). I would disagree, however, that it is optimal to use the middle slice of the acquisition series or a slice about which one might have an a priori hypothesis. Instead, I think the most appropriate reference slice is the first slice acquired in a given TR.

I would argue that this is because the goal of slice acquisition correction is to create a data set that corresponds to what one would have obtained if all of the slices were acquired simultaneously at the moment in time that the TR was initiated. If one models in the G matrix the pattern of neural activity (and induced hemodynamic response) with regard to the onsets of the TRs, any other reference slice would result in a fixed offset (e.g., 1/2 the TR) between the model and the actual data. Of course, one could also shift the covariates of the G matrix by this fixed offset, but this seems

needlessly complicated!

Also, I don't believe that it is the case that using the middle slice affords any greater "accuracy" in making the correction. The slice correction routine makes the assumption that no meaningful power is present in the data above the Nyquist frequency. (For which there is some empirical support). Given this assumption, all shifts in time using sinc-interpolation are equally valid.