Figure 8

The transition diagrams and the posterior probabilities of the reduced feature space after mRMR and AICc. (a) A bidirectional sequence of state progress is found where the HMM is not able to distinguish between tonic firing and postictal activities. (b) The mapping of the HMM output to the experimental data is shown. The tonic firing phase prior to the chronic seizure and the postictal activity are represented by the same model state. Because the state transition probability from the tonic/postictal state is extremely small (0.001), it would take much longer for the interictal activity right after chronic seizure to be classified correctly.