Abstracts

Candidates for epilepsy surgery are routinuely worked up using a variety of imaging modalities, including MEG and EEG. Experts are needed to examine large amounts of data to search for spikes. Once spikes are successfully identified, an inverse solution must be obtained to locate the spike on an anatomical image of the brain. The clinical reports are written based on these localizations. The above process is time consuming and error-prone. By automating these tasks, fatigue and human error are removed from the equation, and clinical reports can be produced in much less time. Five surgical candidates for epilepsy were evaluated using EEG and MEG for one hour. Anatomical MRI images were subsequently obtained. Trained experts visually inspected the raw data and wrote clinical reports for each patient. The same data was then re-evaluated using automated software. Data was filtered to 0.1-30Hz. Equivalent dipoles were obtained at increments of 10ms. Dipoles which had a goodness-of-fit [gt] 80%, and whose current * length values were between 200 and 400 nA*m were considered potential spikes. Artifacts from eye movements and QRS complexes were excluded. MRIs were mapped using Freesurfer to obtain inflated brain surfaces. The software generated images of potential spikes, as well as a written report of named regions of the brain. For example, a named area could be [quot]left hemisphere inferior temporal sulcus.[quot] The written report included an appendix with the time of each potential spike. All five images generated by the automatic spike detector included the areas identified by the human reports. The detector found, on average, 157 candidate spikes per patient. The automatically generated text report correctly identified areas which the human reports had indicated. There was an average of 6 such named areas per patient. Both the image and written report included additional areas which either were not actual spikes, or were not identified as such by the original human report. The automatic spike detector software was able to narrow down potential seizure generation sites to a short list, as well as identify times when possible spikes occurred. Because this list was relatively small, a skilled expert (human or computer) could then quickly narrow down the list to identify true spikes. This more focused search could take the place of the more exhaustive search now commonly done by clinicians. The automatic spike detector and report generation software represent a tool which has high sensitivity and low specificity. The ideal way to take advantage of this would be to use the software as a screening tool prior to final examination by a trained expert. The end result would be a faster, more reliable clinical reporting mechanism for clinical workup of epilepsy surgical candidates using EEG and MEG. (Supported by The MIND Institute)