Automated detection of tonic seizures using 3-D accelerometry

Abstract

A first approach is presented for the detection of accelerometry (ACM) patterns associated with tonic seizures. First it is shown that during tonic seizures the typical ACM-pattern is mainly caused by change of position towards the field of gravity and that the acceleration caused by movement is negligible. To this end a mechanical model of the arm and physiological information about muscle contraction during tonic seizures are used. Then six features are computed that represent the main characteristics of ACM-patterns associated with tonic seizures. Linear discriminant analysis is used for classification. For training and evaluation ACM-data are used from mentally retarded patients with severe epilepsy. It was possible to detect tonic seizures with a success rate around 0.80 and with a positive predictive value (PPV) of 0.35. For off-line analysis this is acceptable, especially when 42 % of the false alarms are actually motor seizures of another type. The missed seizures, were not clearly visible in the ACM-signal. For these seizures additional ACM-sensors or a combination with other sensor types might be necessary. The results show that our approach is useful for the automated detection of tonic seizures and that it is a promising contribution in a complete multi-sensor seizure detection setup.