Anomaly Detection in Invasively Recorded Neuronal Signals Using Deep Neural Network: Effect of Sampling Frequency

Abnormality detection has advanced in recent years with the help of machine learning, in particular with deep learning models, which can predict accurately across many types of signals and applications. In the case of neuronal signals, abnormalities can present themselves as artefacts or manifestations of neurological diseases. Among the diverse neuronal pathologies, we chose to look at the detection of seizures, as they manifest as a brief anomaly in contrast to normal brain activity in the majority portion of the data during a prolonged recording. Epileptic patients benefit from portable systems, which are dependant on efficient energy consumption, and the sampling frequency of the signal is of vital importance element to its battery lifespan. In this article, the impact of the sampling rate on a deep learning-based multi-class classification model is explored via the use of an open-source seizure dataset.