Using gradient as a new metric for dynamic connectivity estimation from resting fMRI data

The study of time varying functional connectivity between different parts of the brain (the functional connectome) has emerged as an important aspect of brain imaging studies. The most widely used approach to estimate these time varying connectivities uses sliding window Pearson correlation to estimate connectivity between different parts of brain. The choice of the window length can impact the results and interesting information might go undetected. Here we propose a new approach that evaluates the gradient (both its magnitude and phase) defined in a new space as a metric for connectivity. Using a very small window, weighted average phase of these gradient values are calculated. Here using simulation, we show that our metric is capable of estimating even very short connectivity states and also provide additional information unavailable to a sliding-window approach. In addition the proposed method is utilized to analyze a real dataset.