Abstracts

Rationale: Surgical planning for treatment of epilepsy depends on the clinician’s understanding of what will be necessary to prevent or reduce seizures without functional deficits. A representation of the probable outcome of the surgical intervention on an anatomically complex network is critical to this. Data from several streams—imaging, semiological interpretation, functional studies, and stereotactic EEG (sEEG)—contribute to the calculation of an apparent best plan. Spatially distributed Bayesian probability calculations regarding network data are implicitly involved in surgical planning, including decisions about where to optimally place sEEG electrodes. In making these calculations more explicit, our tools will aid in the generation of algorithms to assist in surgical planning. We demonstrate two uses of the sEEG electrode coordinates and recordings as the foundation for a coordinate system to study connectivity data obtained from subdural grids (Park & Madsen 2018) or sEEG (Stone et al 2021). The goal is to evaluate strategies for improving the odds of a target locus (represented by a sEEG contact) being in or out of the seizure focus as determined from ictal data.

Methods: First, a search strategy considers a resection to be characterized by the top m contacts in Granger connectivity (GC); the evaluation is in terms of sensitivity and specificity of predicting that any given electrode will be in the independently determined seizure onset zone (SOZ) with the search parameter m, searching within the set of all logical vectors of dimension n, the total number of contacts. In the second approach, the sEEG 3D geometry is included by considering boundaries outlined by the actual SOZ and the candidate resection, defined geometrically as the convex hull of a subset of contacts. All 3D points in the convex hull of all n contacts can be categorized into one of four sets: overlap between “true” SOZ and test volume (true positive), in none of the above (true negative), and in one but not the other (false negative and false positive), yielding a volumetric interpretation of sensitivity and specificity.

Results: Through these tools, sensitivity and specificity of both approaches can be calculated for each of the 20 cases, yielding a statistically significant informativeness score to compare the two SOZ localization strategies without prior knowledge of the size of the seizure onset zone. Additional tools within the set can compute properties of the spatial representation such as volume or overlap with other 3D shapes from separate data sets or calculations.

Conclusions: This computational geometry tool provides a novel technique to assess surgical planning methods by leveraging network data from monitoring procedures, aiding in resection strategies, or determining a need for further localization data—moving research towards Bayesian computation. Geometric analysis of data accounts for the physical boundaries of the SOZ and is thus vital to surgical planning. In future work, Bayesian computation will better describe the overlapping 3D spaces to provide a concrete quantitative method of SOZ localization.

Funding: Please list any funding that was received in support of this abstract.: n/a.