Abstracts

Rationale: About one-third of epilepsy patients do not respond to any medication, and surgical removal of the epileptogenic tissue is a viable option to stop seizures. The precise determination and removal of the epileptogenic zone are critical for obtaining a seizure-free surgical outcome. The precise determination of the seizure onset zone (SOZ) is critical for pre-surgical guidance or intracranial grid placement. In this study, we aim to image the epileptic activity of the brain noninvasively from scalp EEG with an objective and computationally efficient technique. We have tested this method by imaging the inter-ictal activities in focal epilepsy patients who have undergone surgery and became seizure free during their post-surgical follow-up. Methods: We have collected and analyzed pre-surgical MRI and high-density EEG recordings from 10 medically intractable epilepsy patients (9 focal temporal and 1 frontal-parietal), from which an individual and realistic head model was made for each individual patient and interictal spikes were extracted. More specifically, the EEG recordings were first pre-processed to reject bad channels and artifacts. Then, the pre-processed EEG recordings were passed through a matching-filter based semi-automated detector to label out all potential candidates of the inter-ictal spikes, which were then averaged to improve the signal-to-noise ratio. Solving the inverse problem using our recently proposed IRES method (Sohrabpour et al., NeuroImage, 2016), we estimated the spatial location and extent of the inter-ictal activity for each patient. The results were compared to clinical findings such as surgical resection indicated from post-surgical MRI and/or SOZ determined from intracranial studies, for validation. Results: The estimated results of our proposed IRES method match well with the location and extent of the clinical findings defined by surgical resection and/or SOZ. Our initial results suggest that a compact and focal patch source can be robustly estimated with the proposed method from the scalp EEG measurements. The results indicate a localization error of about 5 mm (averaged distance between the estimated source and the surgical resection volume). The estimated source patch falls well inside the surgical resection volume and covers around 80% (on average) of the volume size; the surgical procedure is usually conservative to cover enough tissue to result in seizure-freedom and could be larger than actual epileptic tissues. Our results also indicate that the estimated source is consistent with the SOZ determined from invasive studies, achieving 10 mm localization error in terms of averaged distance between the estimated source and SOZ electrodes. Conclusions: We have validated our iterative reweighting and edge sparsity strategies to provide an efficient and robust estimation of both location and extent of the epileptic electrical activity from dense-array noninvasive surface EEG. The capability of this algorithm to accurately localize and objectively determine the extent of the underlying epileptic brain sources is of crucial interest in studying epileptic brain activities and pre-surgical planning. This algorithm has the potential to be applied as an investigational tool to noninvasively study epileptic brain activities and make an impact on patient populations by providing a precise estimate of the SOZ from the surface measurements, which is currently achieved by invasive intracranial studies. It can also be helpful to guide the placement of the invasive electrode grids and provide additional insight to clinicians for the pre-surgical diagnosis. Funding: This work was supported in part by NIH R01 NS096761 and EB021027.