Vaibhav Diwadkar

Recovering immediate and delayed brain network dynamics: Applications of convergent cross mapping 

We used convergent cross mapping (CCM)(Sugihara et al., 2012) to recover transient and delayed dynamics between brain regions from fMRI data acquired during a) transient motor responding (Diwadkar et al., 2017) and b) sustained working memory maintenance (Falco et al., 2019). Methods: Data (Siemens 3T) were acquired in 20 participants for both tasks. CCM was based on Takens' theorem (which gives the conditions under which a dynamical system can be reconstructed from a sequence of observations of the system’s states). We used a time-symmetric embedding, enabling relatively precise estimation of delayed dynamics. The estimated X(t) and the original X(t) time series at different delays are compared by calculating the linear correlation coefficient between them with (potentially) asymmetric directionality. CCM detected significant (p<;.05, Bonferroni) time delayed interactions (up to 6 s lag and approximately coincident with task dynamics) during sustained working memory maintenance notably between the dlPFC and the parietal cortex. Significant dynamics (between V1, M1 and parietal cortex) for the motor task were recovered at lags of zero but not beyond. It is possible to recover dynamics and task structure from imperfect fMRI signals from a deterministic, maximally dynamic but highly “noisy” system like the brain.