Skip to main content
. Author manuscript; available in PMC: 2022 Apr 1.
Published in final edited form as: Nat Neurosci. 2021 Aug 5;24(10):1465–1474. doi: 10.1038/s41593-021-00901-w

Figure 3: Clinical complexity and our experimental paradigm.

Figure 3:

(a) Schematic of the difficulty of different epilepsy etiologies that might arise in DRE patients. Since there is no biomarker for the EZ and it is never observed directly, the network mechanisms that cause seizures are complex. Case clinical complexity ordered by increasing localization difficulty: lesional (1), focal temporal (2), focal extratemporal (3), and multi-focal (4) that are present in the dataset. These four categories simplify the possible epilepsy presentations, but provide a broad categorization of simple to complex cases observed in the clinic. (b) Schematic of our experimental design. (bottom row) Shows a simplified analogous workflow that clinicians take to evaluate their confidence in a proposed SOZ localization resulting in a surgery. During invasive monitoring, clinicians identify the SOZ from iEEG patterns (e.g. spiking/rhythmic activity). When possible, subsequent surgical resection or laser ablation, generally including the SOZ along with a variable extent of additional tissue, is performed. Post-operatively, patients are followed for 12+ months and categorized as either success, or failure as defined in Methods - Dataset collection, resulting in an Engel or ILAE score. (top row) We evaluate various representations of iEEG in the form of spatiotemporal heatmaps, creating a partitioned summary of the clinically annotated SOZ around seizure onset, feed them into a Random Forest classifier and compute a probability of success (i.e. a confidence score) in the clinically hypothesized SOZ. The probability was then compared with the actual outcome of patients. These predictions can then be stratified based on clinical covariates, such as the actual surgical outcome. For a feature to be an accurate representation of the underlying epileptic phenomena, the following assumptions are made. As a result of seizure freedom, assume that the clinically hypothesized SOZ was sufficient, and the probability of success has a high value. In contrast, if seizures continue, then the SOZ was not sufficient and the probability should have a low value.