Abstracts

Rationale:
Epilepsy is associated with disability, comorbidity, and reduced quality of life, resulting in a 3X higher risk of premature death in the U.S.None of the newer antiseizure drugs developed over the past decades show improved efficacy compared to first-generation drugs, leaving 30–40% of patients with refractory epilepsy. A major barrier to the development of epilepsy drugs is a lack of advanced tools capable of automatically detecting and analyzing seizures in EEG data from preclinical epilepsy models. While manual data review is the most reliable approach, it is highly labor-intensive, expensive, and time-consuming. Moreover, subjectivity and human error often lead to inconsistencies and poor reproducibility, with at best only moderate agreement between ratings by board-certified epileptologists.
Method:
Expesicor is developing PrestEEG, a secure cloud-based EEG analysis tool that automates seizure detection in preclinical EEG data. This platform is unique among existing approaches for preclinical seizure detection and analysis; the user-friendly platform is intuitive enough to be used by non-scientists and incorporates advanced functionality to accommodate experienced preclinical epilepsy researchers. The platform’sproprietary methodologies automatically mark and analyze seizures. A machine learning algorithm runs a secondary validation to identify false positives and provide support for identification of seizures or other neurological biomarkers. Users can scroll through EEG data and review the identified seizures and view summary statistics for the entire data set . Our machine learning techniques are partnered with advanced data management to enable seizure detection with a validation system that provides reliable results. Additionally, our platform offers a secure and large-scale location to store data accessible around the globe.
Results:
Preliminary data has demonstrated that PrestEEG is able to detect seizures and differentiate non-seizure artifacts in one rodent epilepsy model—the rat kainic acid and lorazepam (KaL) model of temporal lobe epilepsy1. More than 20 hours of data from the KaL model have resulted in seizure detection with approximately 85% accuracy and within +/- 5 seconds of the validated start and finish of the event. Machine learning will increase accuracy and assist in artefact detection for a much larger data set in transit from an Expesicor collaborator.   1..        Kienzler-Norwood F, C, et al. A novel animal model of acquired human temporal lobe epilepsy based on the simultaneous administration of kainic acid and lorazepam. Epilepsia 2017;58:222-30.  
Conclusion:
Expesicor’s versatile, cloud-based electroencephalogram analysis tool reliably identifies epileptiform activity for preclinical epilepsy models. In particular, the ability to reduce false positives will greatly improve the accuracy, pace, and reproducibility of preclinical research. Optimization of PrestEEG across additional models will be followed by further development to adapt the platform to additional epilepsy models, diverse collection methods, and clinical data.
Funding:
:Private seed funding and indirect funding from an NIH grant