When is fMRI data independent?
To get independent chunks of data, the experiment itself has to be designed to allow for this partitioning. Ideally, an experiment is split into several chunks, where each chunk is like a mini-version of the whole experiment. It contains all conditions, i.e. experimental manipulations of the independent variables. Another run of the experiment is basically another version of the mini-experiment, preferrably with a different randomized trial-order, to minimize this type of signal artefacts in the data.
At ISMRM 2009, Stephen LaConte presented some guidelines when fMRI data can be considered independent