Article Notes: D'Alessandro et al. (2005)

• Probabilistic Neural Network
• Seizure Prediction
• Intracranial EEG
  • Two studies of 2 and 4 seizures each
• Baselines have to be at least 3 hours from seizures
• Post-processing: Figure of Merit (FOM)
  • 1 - 1.1P(FN) - 0.9P(FP)
    • P(FN) probability of false negative
    • P(FP) probability of false positive
• Preprocessing: Curve length, energy, nonlinear energy, 60 Hz filtering, bipolar montaging for each channel
  • A genetic algorithm chooses which of these stats to use for any particular patient
• Any data that came from a seizure cluster was discarded --- they only used the beginnings of the clusters, because seizure clusters have different properties than ordinary seizures.
• 10-minute data snippets
  • These are divided into sub-blocks
  • The length of the sub-blocks is chosen “empirically” for each patient