Dear Tali,

I do not think that DCM is more sensitive to timing errors than a conventional event-related GLM analysis.  As Karl showed in his 2003 DCM paper, DCM is robust to timing errors up to plus/minus one second.  The reason for this is that the hemodynamic model can absorb minor temporal misalignments.  This is roughly the same temporal scale of sensitivity that you would expect for a GLM-based event related analysis. 

On the other hand, Stefan Kiebel showed in his 2007 NeuroImage paper that you can improve DCMs (in terms of the log model evidence) if you carefully control for timing errors, e.g. by using a generative model of slice timing which is now included in DCM and accessible through the GUI. 

Concerning acquisition order of slices in MRI, interleaved acquisition orders are problematic for various applications, not only for DCM.  The reason is that with interleaved sequences neighboring slices are 0.5 TR apart.  This means that whenever you have to interpolate or combine signals from neighboring slices (e.g. during realignment, normalization or during extraction of characteristic time series from a set of voxels spanning more than one slice) you are mixing signals that are half a TR apart.  For example, using a TR of 3 seconds, you are mixing signals that are acquired 1.5 seconds apart from each other.  This is clearly not ideal and will impact on the quality of your analyses.  From a physical perspective, interleaved sequences are attractive because they avoid a “spilling over” of excitation from one slice to another.  From a signal processing perspective, however, you should always use a continuous sequence even though this means you need to include gaps between slices. 

Best wishes,

Von: Tali Bitan <[log in to unmask]>
An: [log in to unmask]; Klaas Enno Stephan <[log in to unmask]>
CC: Tali Bitan <[log in to unmask]>
Gesendet: Freitag, den 3. Juli 2009, 05:32:19 Uhr
Betreff: Re: DCM - interleaved sequence

Dear Klaas

Can you please explain why would effective connectivity analysis be more
sensitive than the conventional GLM to temporal errors?

Also - why does the interleaved sequence require more interpolation than a
continuous one?

The main question is: how critical is this issue, since our regions of interest
are rather thin, so inter-slice gaps could be problematic.


Tali Bitan

Haifa University