Hello,
we have a similar problem than described in the email by Meng (see below) Our experiment consists of 2 sessions. One experimental condition P is the same in both sessions, but additionally the cognitive state is manipulated in the two sessions (cognitive state: R1 in session 1 and R2 in session 2). So in one session we have a condition P with a state R1 and in the other session condition P with a changed cognitive state R2. Now we want to do a DCM and see the difference of changing connectivity between areas due to P+R1 compared to P+R2. We wanted to concentrate the two sessions and have P+R1 (which would be the onsets of P in session 1 only) and P+R2 (the onsets of P during the session 2 only) as modulatory inputs that both change the connectivity from region A to region B, and then compare the parameter estimates of the modulatory inputs. Is this a valid procedure? If not, how could you implement a changed cognitive state in one session as an modulatory input in DCM?
Thanks for your help,
Marie-Luise
Meng’s question:
Let's say we have two different tasks (L and E) performed in two different sessions (therefore, resulting in two separate datasets). Task L needs a certain connection, but task E doesn't need such connection. …
I want to test if only task L (but not task E) needs a certain connection in a predefined brain network or both tasks require this given connection. I think there are several different ways to do the DCM analysis/comparison but not sure which is correct.
One way is to compare two models, one model includes this connection (model X) and the other model doesn't (model Y), on task L dataset and task E dataset, separately. If task L data show preference to Model X but task E data show preference to Model Y, then it would suggest that only task L needs this connection. Otherwise it would suggest both tasks require this connection.
The second way is still comparing two models on the two task datasets, but this time both model include this connection but one model allows this connection to be modulated by the task but the other model doesn't. So the difference of interest between the two models is parameter B rather than A.
The third way is to combine these two task datasets and then perform model comparison, i.e., compare models on a single, combined, dataset. There are three models to be compared, all including this connection, but first model allows the modulation by task L, the second model allows the modulation by task E, and the third model allows the modulation by both task L and E.
Which is the correct way to do?
|