Model-free optimal de-drifting and enhanced detection in fMRI data

Discriminating between active and non-active brain voxels in noisy functional magnetic resonance imaging (fMRI) data plays an important role when investigating task-related activations of the neuronal sites. A novel method for efficiently capturing drifts in the functional magnetic resonance imaging (fMRI) data is presented that leads to enhanced fMRI activation detection. The proposed algorithm apply a first order differencing to the fMRI time series samples in order to remove the drift effect. Using linear least-squares, a consistent hemodynamic response function (HRF) of the fMRI voxel is estimated as a first-step that leads to an optimal estimate of the drift based on a wavelet thresholding technique. The de-drifted fMRI voxel response is then obtained by removing the estimated drift from the fMRI time-series. Its performance is assessed using a visual task real fMRI data set. The application results reveal that the proposed method, which avoids the selection of a model to remove the drift component, leads to an improved activation detection performance in fMRI data.