Patient-specific rigorously methodological test of the mean phase coherence

Purpose: Due to a confined amount of recordings in existing EEG databases, only few papers have described a patient-specific rigorously methodological test of features for seizure prediction. By extraction of intracranial EEG from epilepsy patients with at least 6 seizures, this study has conducted such a test using the mean phase coherence (MPC) feature. Methods: The MPC was implemented as described by Mormann et al., 2000. It was tested in a generic rigorously methodological way on the FSPEEG database, Winterhalder et al., 2003. 10 patients were used for training of the algorithm’s optimal settings, while the remaining 11 patients were used for testing. With new iEEG extractions from Copenhagen University Hospital, it was possible to obtain training and test sets with adequate amounts of seizures from individual patients. Preictal data from 4 patients was extracted from 11, 7, 7, and 6 seizures respectively together with at least 24 h of interictal data from each individual. With a variant of the leave-one-out training and test method, the optimal amount of training seizures was determined. Results: The generic test on the FSPEEG database resulted in a {sensitivity, false prediction ratio} of {0.55, 0.3/h}. For the 4 new patients the generic results were {0.91, 0.55/h}, {0.85, 0.71/h}, {0.29, 0.49/h}, and {0, 0.36/h} which is comparable to the FSPEEG database. The patient-specific approach yielded {0.81, 0.17/h}, {0.57, 0.13/h}, {0.81, 0.13/h}, and {0.60, 0.50/h} respectively, meaning that the patient-specific approach resulted in a mean improvement of {0.19, 0.30/h}. It was found that 4 seizures were optimal for training. Conclusions: By making the seizure prediction algorithm patient-specific, great improvements were obtained in this preliminary study. 3 out of 4 patients showed clinically applicable results. It is therefore expected that many epilepsy patients can benefit from patient-specific seizure prediction, although there is still room for further improvements.