Incremental activation detection in fMRI series using Kalman filtering

We propose a new detection algorithm for functional magnetic resonance imaging (fMRI) data. Our basic idea is to use an extended Kalman filter (EKF) to fit a general linear model on fMRI time courses, under the assumption of one-degree autoregressive noise with unknown autocorrelation. Because the EKF is designed to be an incremental algorithm, it enables us to compute activation maps on each scan time, and this at moderate computational cost. While our technique is evaluated "offline" in this paper, we believe it is potentially well-suited for future real-time applications.