An Evaluation of SMILE on the TUSZ Corpus

An open source system that enables rapid labeling of seizures and other seizure-like types of brain activity known as “ictal-interictal-injury continuum” (IIIC) patterns [1] was recently released by Jing et al. [2]. At the heart of this system, often referred to as SMILE, is a Seizures, Periodic and Rhythmic Continuum patterns Deep Neural Network (SPaRCNet) model. SPaRCNet is a PyTorch model that aims to classify IIIC events with accuracy that exceeds that of clinical experts. According to the authors, SPaRCNet was trained on “50,697 labeled EEG samples from 2,711 patients and 6,095 EEGs that were annotated by physician experts from 18 institutions.” The system identifies seizures (SZs) and seizure-like events, known as ictal-interictal-injury continuum (IIIC) patterns, in EEG signals [2]. The system outputs labels for SZs, lateralized and generalized periodic discharges (LPD, GPD) and lateralized and generalized rhythmic delta activity (LRDA, GRDA). From a functional point of view, this system reads an EDF file, performs a classification of IIIC patterns, and presents the user with a GUI that enables rapid annotation of large amounts of data.