Dear Will,

I passed your query to Martin Havlicek, who is an expert on the haemodynamic model in DCM, who kindly gave the explanation below.



Best,

Peter.



> A) how DCM will interpret the negative BOLD ?



In standard single-state DCM, the negative BOLD response is modeled as a decrease of neuronal activity below below baseline (you can also call it neuronal deactivation, but I would be more careful with using term inhibition, because there are always two players - excitatory and inhibitory neurons -  and single-state DCM theoretically models only excitatory ones).



You can achieve negative response:

1./ Directly by having driving input scaled by a negative value in matrix C (assuming that you have negative response in one of primary areas); 2./ Or through a negative (feedforward) connection between different brain regions. For example, you might have more primary region with positive response which can cause changes in other region, resulting in decrease of neuronal activity (encoded by negative value in connectivity matrix A or B). I make distinction between negative feedforward and negative feedback connection, because negative feedback connection usually does not result in negative response (only modulates the primary response).



> B)

> how the different types of DCM (traditional, 2-state, stochastic) may 

> differ in this regard



Above I described situation for traditional single-state model. In standard implementation of 2-state DCM you cannot model negative response resulting from causal influence between brain regions (i.e. my point 2. above). Two-state model is built upon assumption that the excitatory connections can be only positive, therefore, it cannot handle decrease of neuronal response below the baseline. We are aware of this limitation and we will soon provide a new two-state model which can model model negative response in similar way as single-state model.



Note, currently there is an extra option for two-state model to model connections between excitatory node of one brain region and inhibitory node of other region, which could provide a mechanism behind the negative response. This is possible if you specify also second layer of connectivity matrix A. However, in my personal opinion this option still requires some adjustments to be really useful.



Same applies to stochastic DCM, depending which neuronal model you use (single or two-state)



Hope this help,

Martin



-----Original Message-----

From: SPM (Statistical Parametric Mapping) [mailto:[log in to unmask]] On Behalf Of Will Gibson

Sent: 29 January 2015 01:44

To: [log in to unmask]

Subject: [SPM] DCM with negative BOLD



Hi all - I understand that DCM infers hidden neuronal states from BOLD, but how does it deal with negative BOLD signal? Does it assume negative BOLD = decreased neuronal activity? I was hoping to run a DCM analysis on a set of VOIs, one of which has a strong inverse correlation with the canonical HRF. Any help understanding A) how DCM will interpret the negative BOLD and B) how the different types of DCM (traditional, 2-state, stochastic) may differ in this regard would be greatly appreciated. Thanks. -Will