Rationale: To evaluate the performance of automated ESI for SOZ localization in refractory epilepsy patients. Methods: The study was performed in fifty consecutive patients with drug-resistant epilepsy (Filadelfia Hospital, Denmark). The analysis started with computation of the spectrogram of the ictal epoch from the marked onset to 3s after for all EEG channel. This shows the overall energy of ictal activity of each channel. Then, the time-frequency window of interest (WOI), which shows the maximum energy over all channels, was identified automatically in the provided frequency band of interest. For ESI, a patient specific forward model was constructed from the MRI with more than 13.000 dipoles were distributed in the gray matter (GM). The brain was segmented into 50 sublobes (25 per hemisphere). As inverse ESI technique, a modified version of LORETA was developed and the performance was evaluated with some simulations. Then, ESI was applied to the ictal epochs and with considering the WOI, a times-series was generated for each dipole. Principal component analysis was applied afterwards to the signals of dipoles belonging to each sublobe to estimate the sulobar activity. Afterwards, the spectral analysis was performed on these estimated time series and the sublobe with the highest activity in the WOI was identified as SOZ. The estimated SOZ was compared to the post-operated MRI at sub-lobar level. Based on the known surgical outcome after one-year follow-up, the performance of the method to localize the SOZ was quantified by calculating the sensitivity, specificity, positive predictive value and negative predictive value. Results: Six patients had data with too much artefacts to provide source localization. They were classified as true negative or false negative depending on their outcome. The analysis led to sensitivity, specificity and accuracy of 64.52%, 89.47%, and 74%, respectively. The Positive Predictive Value and Negative Predictive Value were 90.91% and 60.71%. Conclusions: The results indicate that automated ictal ESI, which has a similar accuracy to analysis done by human experts, has potential to be used during the presurgical evaluation. Funding: Please list any funding that was received in support of this abstract.: None
