Prediction of postoperative EEG changes for intractable epilepsy through a multidimensional autoregressive analysis
Akiyama Y, Mori K, Baba H, Ono K
Department of Neurosurgery, Hamamatsu Rosai Hospital
Preoperative EEGs were quantitatively analyzed by means of a multidimensional autoregressive model (AR model) in order to predict postoperative EEGs. Recorded preoperative EEGs were digitized at an interval of 10 msec. The AR model fitting was executed on each digitized data. As this model described a multichannel feedback system having a peculiar activity in each site under observation, the interstructural relations could be described distinctly in the direction. Namely, the independent and the projected activities through the feedback circuits could be separately described for each brain site. Therefore, postoperative EEG could be simulated by elimination of the component in the AR model corresponding to each operative region. In this report, we presented three cases and discussed usefulness of this method. Case 1 was a patient with post-traumatic epilepsy, who was treated with focal resection. Preoperative EEGs revealed spike and wave discharges mainly in the right frontal region. Simulated postoperative EEGs based on elimination of component from the right frontal region in the AR model, corresponding to focal resection, revealed disappearance of spike and wave discharges. These findings were quite similar to observed postoperative EEGs. Case 2 was a patient with posttraumatic epilepsy, who was treated with anterior callosotomy. Preoperative EEGs revealed diffuse multifocal slow spike and wave discharges. Simulated postoperative EEGs, based on elimination of interhemispheric feedback pathways in the AR model, corresponding to anterior callosotomy, revealed marked lateralization of diffuse spike and wave discharges to the left hemisphere. These findings were qualitatively similar to observed postoperative EEGs. These results suggested that postoperative EEG changes could be well predicted by multidimensional autoregressive analysis.
No Shinkei Geka 1995 Jul;23(7):587-93