Detection of Epileptic Seizures Through Audio Classification

Patients that undergo treatment in the epilepsy clinic Kempenhaeghe in the Netherlands are being monitored with different sensory signals, including audio. In this paper a new patient monitoring system for the detection of epileptic seizures through audio classification is proposed. The proposed system enables automated detection of epileptic seizures, which can have a large positive impact on the daily care of epilepsy patients. This system includes three stages. First, the signal is enhanced by means of a microphone array, followed by a noise subtraction procedure. Secondly, the signal is analyzed by audio event detection and audio classification. When an audio event is detected, features are extracted from the signal. Bayesian decision theory is used to classify the feature vector based on a discriminant analysis. Finally, it is decided whether or not to trigger an alarm. The performance is tested with audio signals obtained from measurements with epileptic patients. The results show that, with a limited set of features, good classification results can be achieved.