Automatic Monitoring of High-Frequency Autoimmune Disorder Related Seizures with Wearable Devices

The detection of Faciobrachial Dystonic Seizures (FBDS) in patients with LGI1-IgG associated autoimmune encephalitis is challenging due to their high frequency and the absence of a clear signal on EEG. This study explores the use of wearable devices as a noninvasive and convenient method for recording these events, thereby providing an objective seizure diary. We investigated the feasibility of using a wrist-worn device to reliably identify and distinguish FBDS from normal arousals. Seven patients with LGI1-IgG autoimmunity (including two who had recordings before and after immunotherapy) and four control subjects were monitored with Empatica wristbands in both hospital and ambulatory environments. We developed a two-stage semi-supervised learning approach, utilizing a Support Vector Machine (SVM) classifier to detect abnormal events based on a small set of labeled events identified by a proprietary algorithm. Significant differences in wearable signal characteristics were found before and after treatment. Putative FBDS events in pre-treatment signals were significantly more frequent than nocturnal arousals in the control group (P1-a: 13.63 event/hour, P2-a: 24.88, vs Control: 3.51, p < 0.001), lasted longer, and were associated with increases in both tonic and phasic Electrodermal Activity (EDA) during events. These findings affirm the potential of wearable technology in providing an automatic and objective measure of FBDS, particularly during sleep, with minimal burden on patients for maintaining a seizure diary.