Rest state fMRI extraction from task-driven data using independent component analysis

Human brain mapping has been given a special emphasis in the neuroscience field. Functional magnetic resonance imaging (fMRI) of the brain is one of the widely used neuroimaging modalities because it is non-invasive and it provides a good spatial and temporal resolution. fMRI research has grown drastically during the last two decades. Therefore, a fairly large number of fMRI data sets has been collected. These data sets allowed researchers to investigate different hypotheses depending on the studied map. The hypothesis driven data collection has lead to task-driven fMRI. Recently, the research community has developed an interest in studying non-task driven resting state fMRI. Consequently, it is important to find a way to extract resting state data from task-driven datasets. In this study we propose using Independent Component Analysis (ICA) to analyze box-car task-driven fMRI data to get its source components. We then identify the components correlated with the performed task and reconstruct the original fMRI signal after removing these components. Thereafter, we conduct the typical fMRI statistical GLM analysis to assert the accuracy of the reconstruction. We also discuss methods to evaluate the reconstructed data and compare it to real rest state fMRI. Our preliminary results show that the rebuilt resting state data is very similar to real rest state data, and we anticipate the differences to be due to physiological reasons related to the inhibition of the brain’s default network during activation. This finding will help researchers draw more results from the vast number of already collected fMRI data.