Precise localization of subdural electrodes (SDEs) is essential for the interpretation of data from intracranial electrocorticography recordings. Blood and fluid accumulation underneath the craniotomy flap leads to a nonlinear deformation of the brain surface and of the SDE array on postoperative CT scans and adversely impacts the accurate localization of electrodes located underneath the craniotomy. Older methods that localize electrodes based on their identification on a postimplantation CT scan with coregistration to a preimplantation MR image can result in significant problems with accuracy of the electrode localization. The authors report 3 novel methods that rely on the creation of a set of 3D mesh models to depict the pial surface and a smoothed pial envelope. Two of these new methods are designed to localize electrodes, and they are compared with 6 methods currently in use to determine their relative accuracy and reliability.
The first method involves manually localizing each electrode using digital photographs obtained at surgery. This is highly accurate, but requires time intensive, operator-dependent input. The second uses 4 electrodes localized manually in conjunction with an automated, recursive partitioning technique to localize the entire electrode array. The authors evaluated the accuracy of previously published methods by applying the methods to their data and comparing them against the photograph-based localization. Finally, the authors further enhanced the usability of these methods by using automatic parcellation techniques to assign anatomical labels to individual electrodes as well as by generating an inflated cortical surface model while still preserving electrode locations relative to the cortical anatomy.
The recursive grid partitioning had the least error compared with older methods (672 electrodes, 6.4-mm maximum electrode error, 2.0-mm mean error, p < 10−18). The maximum errors derived using prior methods of localization ranged from 8.2 to 11.7 mm for an individual electrode, with mean errors ranging between 2.9 and 4.1 mm depending on the method used. The authors also noted a larger error in all methods that used CT scans alone to localize electrodes compared with those that used both postoperative CT and postoperative MRI. The large mean errors reported with these methods are liable to affect intermodal data comparisons (for example, with functional mapping techniques) and may impact surgical decision making.
The authors have presented several aspects of using new techniques to visualize electrodes implanted for localizing epilepsy. The ability to use automated labeling schemas to denote which gyrus a particular electrode overlies is potentially of great utility in planning resections and in corroborating the results of extraoperative stimulation mapping. Dilation of the pial mesh model provides, for the first time, a sense of the cortical surface not sampled by the electrode, and the potential roles this “electrophysiologically hidden” cortex may play in both eloquent function and seizure onset.