Switching Multiple Models for the Segmentation of Sleep EEG Data

Tracey A. Cassar, Kenneth P. Camilleri, and Simon G. Fabri

Keywords

Switching Multiple Models, Jump Linear Systems, Sleep EEG

Abstract

A jump system is characterized by multimodal dynamics which switch from one mode to another in time or space. This work investigates the application of switching multiple models (SMM) assuming linear Gaussian dynamics to segment sleep EEG data which is known to be temporally multimodal. Ad hoc approximations applied to the SMM framework as well as techniques using Interacting Multiple Models (IMM) to handle the large number of possible mode sequences are both shown to give satisfactory segmentation results, comparable to those obtained using Hidden Markov Models in [8]. This work also extends the framework to not only identify windows in the time series containing background EEG or sleep spindles but also those with K-complexes.

Important Links:



Go Back