Kriton Konstantinidis
Faculty Mentor: Emery N. Brown, M.D., Ph.D.
Department of Anesthesia, Critical Care and Pain Medicine, Massachusetts General Hospital
Project Title: Fitting adaptive autoregressive models for frequency analysis of non-stationary EEG data via hybrid Kalman filtering
Department of Anesthesia, Critical Care and Pain Medicine, Massachusetts General Hospital
Project Title: Fitting adaptive autoregressive models for frequency analysis of non-stationary EEG data via hybrid Kalman filtering
Project Summary: High quality frequency spectra are crucial in anesthesia related procedures that demand real-time high-resolution estimation of the frequency content of the Electroencephalogram (EEG) for efficient brain state tracking.
The main target of Kriton’s project is to develop a hybrid adaptive autoregressive model framework to fit non-stationary EEG data.
Kriton is working on a state space approach, used together with a hybrid Kalman filter (HKF) and an Expectation-Maximization (EM) algorithm to estimate dynamically the parameters of the autoregression and by extension the time-varying frequency spectrum of the underlying brain process.
Hybridity arises from the fact that the state vector, which includes the autoregressive parameters, evolves in continuous time while the observation equation is discrete, to account for the fact that the observations (EEG data) are indeed discrete points in time. In other words, while the autoregressive model itself is discrete, its autoregressive parameters are continuous variables. Their adaptive estimation by means of HKF and EM algorithm can lead to robust, high quality estimation of non-stationary EEG spectrograms, outperforming other purely discrete parametric methods as well as various other non-parametric approaches.