Results - noise

In this section, we examine the characteristics of some of the noise parameters in the model. The middle (top to bottom) plots in figure 8 show the first order temporal autoregressive coefficient $\bar{\alpha}_{i1}$ obtained, where $\bar{\alpha}_{i1}$ is the mean of the marginal posterior of $\alpha_i(p=1)$. The temporal autocorrelation can be seen to be spatially non-stationary. EPI slices are shown for comparison and it can be seen that there is increased temporal correlation in grey matter compared with white matter. The bottom plots in figure 8 show the average spatial autoregressive coefficient $\sum_{j\in {\cal N}_i} \bar{\beta}_{ij}/\sum_{j\in {\cal N}_i} 1$ obtained, where $\bar{\beta}_{ij}$ is the mean of the marginal posterior of $\beta _{ij}$. The spatial correlation can also be seen to be spatially non-stationary, reconfirming the finding of the DIC that a non-stationary spatial noise model is required. It also indicates that, like the temporal autocorrelation, the spatial autocorrelation is spatially correlated with tissue type. There is also clearly increased spatial correlation in grey matter compared with white matter.