Dear Roni

The self-connections on the A-matrix (in the 1-state deterministic DCM for fMRI) are log-scaling parameters, which scale the prior of -0.5Hz. The idea is that whatever value is tried by the estimation algorithm, the self-connections will always be negative. To convert the self-connections on A to units of Hz, you need to perform the following transform:

a_ii = -0.5 * exp(DCM.Ep.A(i,i))

To get the self-connection in the context of modulation by input k:

a_ii = -0.5 * exp(DCM.Ep.A(i,i) + DCM.Ep.B(i,i,k))

(During model estimation, this is implemented on line 105 of spm_fx_fmri.m within the DCM neuronal model.)

Apologies for any confusion that has caused – I am working on some new documentation to make all this clearer :-)

Best

Peter

From: SPM (Statistical Parametric Mapping) [mailto:[log in to unmask]] On Behalf Of Roni Tibon
Sent: 23 November 2016 20:19
To: [log in to unmask]
Subject: [SPM] Interpreting DCM results

Dear DCMers,

We hope you could help us with some difficulties in interpreting our DCM for fMRI results.

We are running DCM on 3 ROIs with a large model-space. We have 3 conditions (x, y, z), modelled as three inputs: U(1): x+y+z, U(2): y+z, and U(3): z.

The A-matrix is constant across models, and includes all possible (self and intrinsic) connections. The input (C-matrix) is also constant, and defines input to one region, for U(1) only (ie x+y+z). The models differ in their B-matrix, each with different modulatory effects in B(2) and B(3).

We ran BMS with RFX and BMA to get posterior estimates mEp. We also calculated Pp’s for each mEp. Our questions are:

1)      For one ROI (receiving the input), mEp for the self-connection in B(2) was positive, with a Pp > .95. We were under the impression that self-connections should always be negative (to ensure system settles after perturbation) – so does this positive value for B(2) make sense?  

2)      Even though the modulatory parameter B(2) was positive, we wondered whether the “actual” self-connection strength for each condition was positive, so tried to estimate these as:

X = A-B(1)-B(2)

Y = A+B(1)-B(2)

Z = A+B(1)+B(2)

Are these equations correct? If so, then we still get a positive value of mEp for the self-connection for some conditions (and for some other self-connections) – is that again a concern?

Thanks,

Roni