Abstracts

Rationale: Different functional connectivity measures have been used in previous electrocorticography (ECoG) studies to investigate functional brain networks. However, the optimal connectivity measure and its configurations are still unclear. In this study, we compared performances of three functional connectivity measures, i.e. coherency [1], phase locking value (PLV) [2], and debiased weighted phase lag index (dwPLI) [3], to investigate their robustness against volume conduction (VC) and their detection sensitivity in time and frequency using simulated signals. We found an optimized configuration for each connectivity measure. The results of this study can help researchers to utilize an appropriate connectivity measure for investigating the brain networks. Methods: We simulated electrocorticographic (ECoG) signals in four subdural electrodes considering the VC effect and using four sinusoidal signals oscillating in specific frequencies/time intervals, with/without frequency mismatches and time lags with respect to each other. Then we computed time-frequency functional connectivity (using coherency, PLV, and dwPLI measures) between all pairs of these signals by various configuration of three parameters (i.e. windowing method, length of time window, and time window shift). After that, the connections between these signals were estimated for each configuration of parameters. For each electrode pair, we calculated the mean square error (MSE) between the estimated connectivity matrix and its true value. Then we calculated an average MSE across all pairs of electrodes to find optimal configuration of three parameters. Results: The MSEs corresponding to two parameters, i.e. length of time window and time window shift, for two windowing methods are shown in Fig.1. We found that dwPLI, Digital Prolate Spheroidal Sequence (DPSS), 95 ms, and 8 cycles of corresponding frequency were optimal choices for the functional connectivity method, method of windowing, shift of time window, and length of time window, respectively. The estimated connectivity using the optimal configuration of parameters is shown in Fig.2. Conclusions: Our results revealed that in existence of VC, time lags, and frequency mismatches between cortical sources, the dwPLI with an optimal configuration can provide an accurate representation for the brain network. This configuration set can be used in future studies in order to achieve more accurate results. Funding: This study was funded by the Children’s Foundation Research Institute, Memphis, TN.