FMRI BASELINE DRIFT ESTIMATION METHOD BY MDL PRINCIPLE

This paper introduces a new method for estimating and removing baseline drift in the fMRI signal. We propose the technique of minimum description length (MDL) to the problem of fMRI drift analysis. The proposed method is based on the iterative estimation of the activation level and the drift component, using least square estimation and the concept of MDL denoising. Simulation results show that the proposed algorithm can estimate the drift term without having any prior knowledge or assuming an overly restrictive model. The algorithm is tested on both simulated and real fMRI data. The performance is compared with that of polynomial detrending method used in AFNI.