Activation Detection on fMRI Time Series Using Hidden Markov Model
Figure 2
(a) HMM log-likelihood map—2-state HMM model is applied to all voxels. The active voxels with obvious 2-state on/off patterns will have relative high likelihood given this model; (b) Gaussian log-likelihood map—“off”-state Gaussian model is applied to all voxels. The nonactive voxels with obvious 2-state on/off patterns will have relative high likelihood given this model; (c) HMM likelihood ratio test map—subtraction of HMM log-likelihood map and Gaussian log-likelihood map equivalent to general likelihood ration test, and it enhances the contrast for activation map.