Mapping effective connectivity by virtually perturbing a surrogate brain

Luo et al. proposed a deep learning based method using next-step-prediction to train ANNs for fMRI signals. With the well-trained ANNs, a virtual perturbation scheme can be adopted to infer a brain-wide effective connectome (EC). The inferred EC was found as correlated with the empirical EC (i.e., CCEPs from F-TRACT dataset). They first provided an proof-of-concept of using next-step-prediction with ANNs for fMRI signals, however, there is still unsolved issues about how to identify the noise amplitude in resting-state simulation of the ANN model.