Frontiers in Neurology (Oct 2021)
Digital Semiology: A Prototype for Standardized, Computer-Based Semiologic Encoding of Seizures
Abstract
Video-EEG monitoring (VEM) is imperative in seizure classification and presurgical assessment of epilepsy patients. Analysis of VEM is currently performed in most institutions using a freeform report, a time-consuming process resulting in a non-standardized report, limiting the use of this essential diagnostic tool. Herein we present a pilot feasibility study of our experience with “Digital Semiology” (DS), a novel seizure encoding software. It allows semiautomated annotation of the videos of suspected events from a predetermined, hierarchal set of options, with highly detailed semiologic descriptions, somatic localization, and timing. In addition, the software's semiologic extrapolation functions identify characteristics of focal seizures and PNES, sequences compatible with a Jacksonian march, and risk factors for SUDEP. Sixty episodes from a mixed adult and pediatric cohort from one level 4 epilepsy center VEM archives were analyzed using DS and the reports were compared with the standard freeform ones, written by the same epileptologists. The behavioral characteristics appearing in the DS and freeform reports overlapped by 78–80%. Encoding of one episode using DS required an average of 18 min 13 s (standard deviation: 14 min and 16 s). The focality function identified 19 out of 43 focal episodes, with a sensitivity of 45.45% (CI 30.39–61.15%) and specificity of 87.50% (CI 61.65–98.45%). The PNES function identified 6 of 12 PNES episodes, with a sensitivity of 50% (95% CI 21.09–78.91%) and specificity of 97.2 (95% CI 88.93–99.95%). Eleven events of GTCS triggered the SUDEP risk alert. Overall, these results show that video recordings of suspected seizures can be encoded using the DS software in a precise manner, offering the added benefit of semiologic alerts. The present study represents an important step toward the formation of an annotated video archive, to be used for machine learning purposes. This will further the goal of automated VEM analysis, ultimately contributing to wider utilization of VEM and therefore to the reduction of the treatment gap in epilepsy.
Keywords
