Detecting Tonic-Clonic Seizures in Multimodal Biosignal Data From Wearables: Methodology Design and Validation.
Sebastian Böttcher1,2, Elisa Bruno3, Nikolay V Manyakov4
1Epilepsy Center, Department of Neurosurgery, Medical Center - University of Freiburg, Freiburg im Breisgau, Germany.
JMIR Mhealth and Uhealth
|November 22, 2021
Summary
Wearable sensors can accurately detect tonic-clonic seizures (TCSs) using multimodal biosignals and machine learning. This technology offers a promising solution for continuous epilepsy monitoring outside clinical settings.
Area of Science:
- Biomedical Engineering
- Neurology
- Machine Learning
Background:
- Video electroencephalography (EEG) is the gold standard for seizure monitoring but is not feasible for daily life.
- Wearable devices offer a potential solution for continuous, real-world epilepsy seizure logging.
Purpose of the Study:
- To evaluate supervised machine learning for detecting tonic-clonic seizures (TCSs) using multimodal wearable biosignals.
- To assess the performance of the detection model on a new dataset during both daytime and nighttime.
Main Methods:
- Utilized a multimodal watch recording accelerometry and electrodermal activity from 10 participants with 21 TCSs.
- Extracted 10 accelerometry and 3 electrodermal activity features, analyzed with a gradient tree boosting algorithm.
- Employed leave-one-participant-out cross-validation for model optimization and out-of-sample testing.
Main Results:
- Achieved 100% sensitivity in detecting 10 training seizures with a low false alarm rate (0.46/day).
- The model detected 10 out of 11 TCSs in the test set with zero false positives.
- Specificity analysis on data from 28 additional participants yielded a false alarm rate of 0.19/day.
Conclusions:
- Supervised machine learning robustly detects TCSs from multimodal wearable data.
- High sensitivity and low false-positive rates are achievable even with limited training data.
- This methodology shows promise for wearable-based nonconvulsive seizure detection.
