Automated simulation and replication of fMRI experiments

We present a free online tool that predicts brain activity from text, images and videos that users supply to the system. The tool is based on models that were trained using data from experiments in which subjects process complex, uncontrolled stimuli: they listen to hours of stories and watch real movies. The models can then predict brain activity for any new stimulus, effectively simulating the results of an fMRI experiment that was not performed. Our system allows users to run a variety of common analysis tools used in imaging such as computing contrasts between conditions, running statistical tests, and interactively visualizing the obtained brain maps in multiple views. This can be useful while planning a new fMRI experiment, and could be crucial for encouraging replicability of fMRI results: published results and activation maps can be easily compared against our predictions for the same stimuli.