A Bayesian approach for epileptic seizures detection with 3D accelerometers sensors.

In this paper, an algorithm able to detect epilepsy seizure based on 3D accelerometers and with patient adaptation is presented. This algorithm is based on a Bayesian approach using hidden Markov models for statistical modelling of moves signals. A particular focus is set on the learning procedure and in particular on its initialisation to ensure a good learning and to avoid numerical instability. Numerical simulations show that, without inhibition of the detection algorithm when the person is standing up, the algorithm is able to detect close to 90% of seizures when false alarms are 25% of alarms.