Diff for "SensorSpm" - Meg Wiki
location: Diff for "SensorSpm"
Differences between revisions 17 and 18
Revision 17 as of 2009-01-22 13:55:09
Size: 4764
Editor: RikHenson
Comment:
Revision 18 as of 2009-01-22 13:56:57
Size: 4959
Editor: RikHenson
Comment:
Deletions are marked like this. Additions are marked like this.
Line 75: Line 75:
Finally, if you have an a priori timewindow and/or an a priori set of sensors where you expect an affect, you can use Small Volume Correction (the SVC button). With this, you can specify a mask image saved from an orthogonal contrast in the same SPM analysis, or from a previous SPM analysis of independent data. Or perhaps more commonly, you can select a 3D "box". For example, say you had an a priori hypothesis that a condition effect would occur from 100 to 200ms poststimulus, then type [0 0 150] into the coordinates on the Input window, press SVC, select "box" and enter [80 80 100] as its dimensions (corresponding to an +/-40mm range around x, +/-40mm range around y, and +/-50ms range around 150ms. Finally, if you have an a priori timewindow and/or an a priori set of sensors where you expect an effect, you can use Small Volume Correction (the SVC button). With this, you can specify a mask image saved from an orthogonal contrast in the same SPM analysis, or from a previous SPM analysis of independent data. Or perhaps more commonly, you can select a 3D "box". For example, say you had an a priori hypothesis that a condition effect would occur from 100 to 200ms poststimulus, then type [0 0 150] into the coordinates on the Input window, press SVC, select "box" and enter [80 80 100] as its dimensions (corresponding to an +/-40mm range around x, +/-40mm range around y, and +/-50ms range around 150ms). The corrected (but not uncorrected) p-values in the Graphics window will now change (get smaller) owing to the reduced search volume. Hopefully some will be <.05 corrected for height or extent!

3D Sensor SPMs

3D topography x time sensor SPMs can help you identify the latency and location of effects in sensor space. For example, here's a word-pseudoword effect (N=19) on magnetometers and gradiometers (RMS), thresholded for height (p<.001 unc) and extent (p<.05 corr):

attachment:WdNw3DSensorSpm.jpg

Creating 3D Sensor SPMs

1. Select an averaged file (m*.mat), split it into separate files for magnetometers, gradiometers, and gradiometer-magnitude (m*-mags.mat, m*-grds.mat, and m*-grms.mat, respectively). (Note you can do a similar analysis within a single subject by writing epochs to images).

clear S
S.D='mae_mydata.mat';
spm_eeg_splitFIF_grms(S);

2. Write image volumes.

% Select options (see help for spm_eeg_convertmat2ana3D):
clear S
S.interpolate_bad = 0;
S.n = 32;
S.pizsize = 3;
% Select trial types:
S.trialtypes = [1 2];
% Mags:
S.Fname = 'mae_mydata-mags.mat';
spm_eeg_convertmat2ana3D(S);
% Grad magnitude:
S.Fname = 'mae_mydata-grms.mat';
spm_eeg_convertmat2ana3D(S);

3. Smooth image volumes (optional, but good idea for grms, since distribution of RMS of gradiometer values [over voxels] is not Gaussian, even if by Central Limit Theorem, the distribution of the mean over enough subjects will be).

% Select files (or pass from stage above)
P = spm_select(Inf,'image','Select image file(s)',[],pwd,'.*img');
% Smoothness in x (mm), y (mm) and z (ms)
SmoothKernel = [5 5 10];  
for n=1:size(P,1);
           [pth,nam,ext] = fileparts(P(n,:));
           Pout{n}       = fullfile(pth,['s' nam ext]);
           spm_smooth(spm_vol(P(n,:)),Pout{n},SmoothKernel);
           Pin = strvcat(P(n,:),Pout{n});
           spm_imcalc_ui(Pin,Pout{n},'((i1+eps).*i2)./(i1+eps)',{[],[],'float32',0}); 
end
%The final imcalc step above is just to reinsert NaN's for voxels outside space-time volume into which data were smoothed

4. Then compute an ANOVA over subjects. Either use SPM's GUI, or a script like 'meg_batch_anova.m' which can be found in /imaging/local/meg_misc (or here http://imaging.mrc-cbu.cam.ac.uk/svn/meg_misc/devel/), eg:

% wd is your working directory (eg pwd)
outdir = fullfile(wd,'SensorTimeSPMs');
meg_batch_anova(Pout,outdir);

Viewing 3D Sensor SPMs

To view the results, navigate to the directory that contains the SPM.mat for your ANOVA. In an SPM window, click 'Results' (1), select SPM.mat (2), select contrast Unwhitened EOI (3), and select options: (4) don't mask, whatever title, choose your p-value and extent thresholds (I usually start with .001 and 0).

attachment:Demo3dSpm1Scaled.jpg

Click 'whole brain' (5) to view whole-volume results. The coordinates on the far right (6) show x (mm), y (mm), and time (ms). Ignore the glass brain images (7).

attachment:Demo3dSpm2Scaled.jpg

To view the cloud-in-a-box images, click 'overlays'->'sections', (8) and select mask.img (9). Lovely (10). If you want to write out the image volume, click 'save' and give it a name. You can then view it using 'Display' or 'Check Reg'.

attachment:Demo3dSpm3Scaled.jpg

If you want to correct for extent, rather than clicking 'Results', click 'Toolbox'->'ns'. As above, select SPM.mat, Unwhitened EOI, options: don't mask, whatever title, no p-value adjustment, threshold {p} .001 (or your choice), extent threshold 0. The leftmost column ('cluster-level p-corrected') shows p-values corrected for cluster extent (11). Note that this information was absent when you clicked 'Results'.

attachment:Demo3dSpm4Scaled.jpg

If you want to view or write out an image thresholded for cluster extent, determine what cluster size ('cluster-level k-voxel') needed to achieve your desired p-value ('cluster-level p-corrected'; in the example, 214 gets p=0.052 (12)). Then follow the above 'Toolbox'->'ns' directions, but this time enter the k-voxel extent threshold when prompted (eg., 215 (13)). Now the results are thresholded for both height and extent (14). View and save as above.

attachment:Demo3dSpm5Scaled.jpg

Finally, if you have an a priori timewindow and/or an a priori set of sensors where you expect an effect, you can use Small Volume Correction (the SVC button). With this, you can specify a mask image saved from an orthogonal contrast in the same SPM analysis, or from a previous SPM analysis of independent data. Or perhaps more commonly, you can select a 3D "box". For example, say you had an a priori hypothesis that a condition effect would occur from 100 to 200ms poststimulus, then type [0 0 150] into the coordinates on the Input window, press SVC, select "box" and enter [80 80 100] as its dimensions (corresponding to an +/-40mm range around x, +/-40mm range around y, and +/-50ms range around 150ms). The corrected (but not uncorrected) p-values in the Graphics window will now change (get smaller) owing to the reduced search volume. Hopefully some will be <.05 corrected for height or extent!

CbuMeg: SensorSpm (last edited 2013-03-08 10:02:45 by localhost)