Dear Jose,
during model setup SPM uses so called "stick functions" to model responses. These have a value of "1" for some time points, and "0" everywhere else. If you specify zero duration SPM will place just one "1" at the time where you specified the onset. If you specify a longer duration, SPM models a sequence of "1"s beginning at your onset and ending at onset+duration. Later on, these stick functions are convolved with your basis function (set) and the resulting time series are placed in your design matrix.
You observed that a strictly event related design does not capture the time course in your data, while a design using prolonged events does. There might be technical, physiological or neuropsychological reasons for this.
Technical: your timing is slightly off (e.g. due to incorrect synchronisation between scanner and experiment, wrong calculation of timing offsets if you manually discarded some scans at the beginning of your MR measurement). In that case, fixing the timing should get you a good chance to analyse the data as planned.
Physiological or neuropsychological: The timing is correct, but you are observing responses that have longer or delayed BOLD responses compared to a canonical hemodynamic response (e.g. reaction time experiments). These differences may even be region specific, and you should model them accordingly. Some possible solutions include using temporal and dispersion derivatives together with the hemodynamic response function or explicit modelling of responses as separate experimental conditions. The latter will only work if stimulus presentation and response are far enough separated in time or e.g. variable according to reaction times.
Best regards,
Volkmar
|