Class Distributions

We need to specify $p(y_i\vert x_i=k,\theta_k)$ for all classes. This is application specific. In brain segmentation tasks the observations, $y_i$, might be intensity levels in a brain image, and the mixture might typically be made up of Gaussians (Zhang et al., 2001). In this paper we intend to use the mixture models proposed for classifying statistical parametric maps (SPMs) in the anlysis of FMRI data.