Multimodal seizure detection: A review.

Abstract

A review is given on the combined use of multiple modalities in non electroencephalography (EEG)-based detection of motor seizures in children and adults. A literature search of papers was done on multimodal seizure detection with extraction of data on type of modalities, study design and algorithm, sensitivity, false detection rate, and seizure types. Evidence of superiority was sought for using multiple instead of single modalities. Seven papers were found from 2010 to 2017, mostly using contact sensors such as accelerometers (n = 5), electromyography (n = 2), heart rate (n = 2), electrodermal activity (n = 1), and oximetry (n = 1). Remote sensors included video, radar, movement, and sound. All studies but one were in-hospital, with video-EEG as a gold standard. Algorithms were based on physiology and supervised machine learning, but did not always include a separate test dataset. Sensitivity ranged from 4% to 100% and false detection rate from 0.25 to 20 per 8hours. Tonic-clonic seizure detection performed best. False detections tended to be restricted to a minority (16%-30%) of patients. Use of multiple sensors increased sensitivity; false detections decreased in one study, but increased in another. These preliminary studies suggest that detection of tonic-clonic seizures might be feasible, but larger field studies are required under more rigorous design that precludes bias. Generic algorithms probably suffice for the majority of patients.