Functional MRI (fMRI) is a leading noninvasive technique for mapping functionally connected networks in the human brain. The fMRI hemodynamic response is slow and noisy, and is sampled far more slowly (seconds) than the timescale of neuronal spikes (milliseconds). fMRI data are, therefore, considered unsuitable for mapping directed, time-lagged functional connectivity among brain regions. Here, we apply machine learning to fMRI data from 1000 human participants and show that directed connectivity, estimated with Granger–Geweke causality from fMRI data, accurately predicts task-specific cognitive states, and individual subjects’ behavioral scores. Moreover, directed connectivity robustly identifies network configurations that may be challenging to identify with conventional correlation-based approaches. Directed functional connectivity, as measured with fMRI, may be relevant for a complete understanding of brain function.
Ajmera SA, Jain H, Sundaresan M & Sridharan D* (2020). Decoding task-specific cognitive states with slow, directed functional networks in the human brain. eNeuro 7(4). doi: 10.1101/681544.