Combined correction for geometric distortion and its interaction with head movement in fMRI

C. Hutton, R. Deichmann, R. Turner, J. Andersson Wellcome Department of Imaging Neuroscience, UCL, London, United Kingdom, MRC, Karolinska Institute, Stockholm, Sweden Introduction The introduction of an object (head) in a magnetic field causes geometric distortions, the specifics of which depend on the object and its orientation. This means that the apparent shape of the brain will vary with its position in the scanner, leading to movement related variance in fMRI time-series that cannot be corrected by rigid body realignment [1]. Correcting this by acquiring a field-map at each time point would lead to increased repetition time and may, due to errors in the measured field-maps, not always decrease variance [2]. Previous work has shown that the rate of change of the field with respect to subject movement can be estimated directly from the variance of the time-series [3], alleviating the need for measuring the field at each time point. This method explicitly minimizes variance under constraints given by the physics of the imaging process. Therefore variance will always be reduced or remain unchanged. This can have particular impact on fMRI time-series with task-correlated head motion because motion artefacts can be corrected without removing true activation, which can happen when using movement parameters as regressors of no interest in the statistical model [4]. However, using the method presented in [3] means the images will be undistorted to some average distortion over the whole time-series rather than to the true anatomy. Here we present a method that combines geometric distortion correction using a single measured EPI-based field-map with the model-based approach that estimates change in field at each time point. This method yields an unwarped time-series corrected for both geometric distortions and the interaction of head movement with distortion. Theory We represent the field at each and every time point as a first order Taylor expansion with respect to the movement parameters. The movement parameters and the corresponding component fields of the expansion (the rate of change of the field w.r.t. head movement) are estimated from the time-series. The zeroth order term, i.e. the static field is measured at the beginning of the time-series using dual echo-time images followed by phase unwrapping [5]. The resulting field map for each time-point is then calculated by combining the measured field-map with the estimated change in field (figure 1) and used to unwarp the time-series.