Data acquisition was performed on a 3 Tesla whole-body MR scanner (MedSpec S300, Bruker Biospin, Ettlingen, Germany) using single-shot gradient echo--planar imaging (T_R of 1000 s, effective T_E of 35 ms slice thickness 4mm, field-of-view of 23 times 23 cm^2)
The simulated one is a fully artificially constructed mathematical fMRI phantom with known activation. It consists of a time series (35 time courses) of a transversal brain slice (128 times 128 pixels) with a time invariant texture of 4809 pixels (gray/white matter, ventricles) and three regions of activation (49 pixels). The relative signal increase upon ``activation'' in all activated pixels was 5%, 7%, and 9% with 3% of baseline level Gaussian noise of zero mean value added, resulting in three simulated fMRI data sets with functional CNR (contrast-to-noise ratio) of 1.33, 1.66 and 2, respectively, common values in fMRI of the human brain.
The hybrid data sets were constructed using a baseline in vivo MRI data set with the activation added artificially. It consists of a time series of 140 images with a matrix size of 64 times 64 pixels. The slice chosen was overlaid with 25 pixels of activation (a square of 5 times 5) with a contrast-to-noise ration of 1.33, 1.66 and 2 where the noise was calculated inside a region within the brain.
It is in principle possible to provide the baseline dataset for your simulations under the conditions that you :