PB14. Fully automated segmentation of intracerebral depth electrode contacts with subsequent separation into distinct electrodes in patients with pharmacoresistant epilepsy

Introduction The localization of the seizure onset zone, being mandatory prior to epilepsy surgery, frequently requires the use of invasive depth electrodes. The interpretation of the electrophysiological recordings then critically depends on the visualization of the electrode positions in 3D MRI scans of these patients. We developed a fully automated software pipeline to segment the depth electrode contacts and to achieve a separation into distinct electrodes. The pipeline uses pre- and post-implantation cranial computerized tomography as well as post-implantation 3D MRI. It consists of standard pre-processing routines for co-registering and masking of images and a custom-made image analysis procedure. Prior spatial information about the implanted electrodes is not required. Methods In a first step, both the pre-implantation CT (preOP-CT) and the post-implantation CT (OP-CT) were co-registered to a post-implantation 3D MRI. Then masks of brain and skull bone were extracted from the preOP-CT, and an automated search for intensity clusters was performed in the brain masked OP-CT. The detected clusters were finally separated into electrode contacts. The only requirements for this method were a voxel size of 1 × 1 × 1 mm or better for the OP-CT and an equidistant spacing of the contacts of all implanted electrodes. Results The method was tested in 28 patients with pharmacoresistant focal epilepsy (161 electrodes and 1608 contacts altogether). A comparison between automatic segmentation and visual identification of the electrode contacts showed that 1597 contacts (99.3%) were correctly detected. Only 9 contacts, which were all located within or outside the skull bone, remained undetected. In 2 patients, one single contact was falsely detected and had to be corrected manually. The pre-processing routines (co-registration and masking) took about 30 min per patient. The image analysis procedure (cluster search and electrode separation) took less than 30 s per patient. Conclusions The implemented method is, to the best of our knowledge, the first fully automated procedure for the segmentation of intracerebral depth electrode contacts with subsequent separation of the detected contacts into distinct electrodes, which can be directly superimposed on 3D MRI. The method does not require any a priori information about the implanted electrodes and is able to identify the loci of depth electrode contacts in a fast and reliable manner.