Scatterplots (F,G) show median H vs. Gconn of brain voxels, for run-1 vs. run-2 (young subjects only) and old vs. young subject (averaged over runs).
Scatterplots (D,E) show median H vs. SDBOLD of brain voxels, for run-1 vs. run-2 (young subjects only) and old vs. young subject (averaged over runs).
Moreover, the R2 goodness of fit is highly consistent over all brain voxels for DFA, whereas PSD obtains higher fit in some regions (e.g., cingulate cortex, parietal lobes) and lower fit elsewhere (e.g., sensorimotor cortex).
GPC can also provide a spatial map showing the most discriminating regions (i.e., the relative importance of each brain voxel for the discrimination boundary).
This apparently was not due to elevated DirBa activation in patients; DirBa activation, as a percentage of mean activation in all brain voxels, was similar for patients and controls.
After the best choice of H, we had a large number of weights (one for every brain voxels) so we kept only the voxels for which |w| > > 0, discarding the other as less important for the classification.
For each participant, "brain" voxels in a specific image were defined as voxels with an intensity greater than 15% of the maximum value in that image.
The parameter estimates were calculated for all brain voxels using the general linear model, and contrast images comparing each condition against fixation (i.e., the baseline) were computed [Friston et al., 1995b].
To do so, we first used the AFNI program 3dLSS (Least Square Sum regression43) to estimate univariate trial-wise β-coefficients for all brain voxels from the concatenated data.
The parameter estimates were calculated for all brain voxels using the general linear model.
