Authors :
First Author: Sebastian Böttcher, PhD – Epilepsy Center - University Medical Center Freiburg
Presenting Author: Matthias Dümpelmann, Dr. – Medical Center - University of Freiburg
Nicolas Zabler, Master of Science – Epilepsy Center – University Medical Center Freiburg; Michele Jackson, MD – Division of Epilepsy and Clinical Neurophysiology – Boston Children’s Hospital, Harvard Medical School; Solveig Vieluf, PhD – Division of Epilepsy and Clinical Neurophysiology – Boston Children’s Hospital, Harvard Medical School; Elisa Bruno, MD – Department of Basic and Clinical Neuroscience, Institute of Psychiatry, Psychology and Neuroscience – King’s College, London; Mona Nasseri, PhD – School of Engineering – University of North Florida, Jacksonville; Matthias Dümpelmann, PhD – Epilepsy Center – University Medical Center Freiburg; Mark Richardson, PhD Md – Department of Basic and Clinical Neuroscience, Institute of Psychiatry, Psychology and Neuroscience – King’s College, London; Benjamin Brinkmann, PhD – Bioelectronics Neurophysiology and Engineering Laboratory, Department of Neurology – Mayo Clinic, Rochester; Tobias Loddenkemper, PhD Md – Division of Epilepsy and Clinical Neurophysiology – Boston Children’s Hospital, Harvard Medical School; Andreas Schulze-Bonhage, PhD MD – Epilepsy Center – University Medical Center Freiburg
Rationale:
Wearable non-EEG biosignal recordings captured from the wrist/ankle offer potential for seizure monitoring. However, signal quality remains a challenge impacting the reliability and value of biosignal data. Models trained for seizure detection rely on features calculated specifically from ictal and peri-ictal data. Thus, we investigate signal quality changes related to seizures to establish accurate and reliable epilepsy monitoring with wearables.
Methods:
We analyzed wearable biosignal recordings (Empatica E4, Empatica, Milan) from 238 study participants including 940 seizures acquired at three international Epilepsy Monitoring Units (EMUs). The study applied previously developed measures
1 to assess signal quality in ictal and peri-ictal data. The pre-ictal period was defined as a five minute window two minutes before the seizure onset, and the post-ictal period as a minute minute window two minutes after the seizure offset. Seizures were divided into three groups: tonic-clonic (TCS), focal motor, and focal non-motor seizures. We evaluated electrodermal (EDA) and blood volume pulse (BVP) biosignal quality, in the context of accelerometer-based physical activity (ACT). Figure 1 plots accelerometry and ACT, EDA, and BVP indices for selected seizures from each group.
1 Böttcher, Vieluf, et al. Data Quality Evaluation in Wearable Monitoring. Sci. Rep. 2022, doi:10.1038/s41598-022-25949-x.
Results: Changes in signal quality and activity level occurred across all seizure types, most notably in motor component seizures, between ictal and peri-ictal phases (Wilcoxon signed-rank test; p< 0.01; Figure 2). The largest decrease in signal quality and increase in activity were seen for TCS (change pre-ictal to ictal, EDA: -3.6%, BVP: -26.2%). Overall, EDA signals were less prone to signal quality changes (change pre-ictal to ictal, -3.6%, -3.5%, 2.7%, for the three seizure groups respectively), while BVP, especially regarding motor seizures, had the highest changes in mean signal quality (change pre-ictal to ictal, -26.2%, -20.9%, -8.8%, for the three seizure groups respectively). Mean signal quality levels during nightly seizures were overall better than during the day. Additionally, non-epileptic movements and nurse intervention may affect the signal quality during seizures classified as non-motor, but further in-depth analysis is required.