Authors :
Presenting Author: Matthew McCumber, MS – University of Nebraska Medical Center
Srijita Das, MS – University of Nebraska Medical Center; Stephen Gliske, PhD – University of Nebraska Medical Center; Garnett Smith, MD – University of Michigan; William Stacey, MD-PhD – University of Michigan; Kevin Tyner, MS – University of Nebraska Medical Center
Rationale:
Epilepsy is one of the most common neurological disorders. For thirty percent of patients, medication alone does not control seizures. Resective surgery remains the standard of care in focal refractory epilepsy. Yet, surgery does not always result in seizure freedom. The purpose of this study was to objectively localize seizure activity in epilepsy patients using intracranial EEG data with the goal of aiding in resective surgery planning to improve surgery outcomes.
Methods:
Patients with Engel Class 1 outcomes with sEEG implantation were selected from the University of Michigan intracranial EEG database. One seizure was selected at random per patient, and channels known to be extraparenchymal or to have poor data quality were redacted. Seizures with low data quality were redacted, leaving N=5 subjects. A high pass filter was applied at 1 Hz. Individual Component Analysis was performed to reduce noise. A common average reference filter was applied. The lead channel for analysis was selected from the clinical notes. Time-frequency analysis was performed on the selected channel. A unique frequency was extracted for the analysis of each seizure by identifying the frequency with the highest power. A seizure-specific frequency window was established spanning the selected frequency +/- 1 Hz. The power of all channels was calculated in this frequency window. The baseline power mean and standard deviation was calculated for each channel from 10 to 60 seconds before the clinically marked seizure onset. One second epochs were created for the first two seconds following seizure onset. The channels with power greater than their own baseline mean +2 SD in both epochs were considered active. The active channels were then compared to the clinically documented resected volume.
Results:
In two subjects, the identified channels were contained in the resected volume (positive predictive value, PPV=100%). In the other three subjects, the PPV was 75%, 53%, and 37%. Post hoc visual analysis determined that the two subjects with lowest PPV also had the lowest frequencies selected (both < 6.5 Hz).