Multistage seizure detection techniques optimized for low-power hardware platforms

Abstract

Closed-loop neurostimulation devices that stimulate the brain to treat epileptic seizures have shown great promise in treating more than a third of the 2 million people with epilepsy in the United States alone whose seizures are currently nonresponsive to pharmaceutical treatment. Seizure detection algorithms facilitate responsive therapeutic intervention that is believed to increase the efficacy of neurostimulation by improving on its spatial and temporal specificity. Translating these signal processing algorithms into battery-powered, implantable devices poses a number of challenges that severely limit the computational power of the chosen algorithm. We propose a cascaded two-stage seizure detection algorithm that is computationally efficient (resulting in a low-power hardware implementation) without compromising on detection efficacy. Unlike traditional detection algorithms, the proposed technique does not explicitly require a "training" phase from individual to individual and, instead, relies on using features that result in distinct "patterns" at the electrographic seizure onset. We tested the algorithm on spontaneous clinical seizures recorded using depth electrodes from patients with focal intractable epilepsy and annotated by epileptologists at the University of Freiburg Medical Center, via the Freiburg database. The algorithm performs with a specificity and sensitivity of 99.82 and 87.5%, detecting seizures in less than 9.08% of their duration after onset. The proposed technique is also shown to be computationally efficient, facilitating low-power hardware implementation. This article is part of a Supplemental Special Issue entitled The Future of Automated Seizure Detection and Prediction.