Exhibit Hall | Forum 9
Purpose: Time-varying Granger causality refers to patterns of causal relationships that vary over time between brain functional time series at distinct source and target regions. It provides rich information about the spatiotemporal structure of brain activity that underlies behavior. Current methods for this problem fail to quantify nonlinear relationships in source-target relationships, and require ad hoc setting of relationship time lags.
Methods: We propose deep stacking networks (DSN), with adaptive convolutional kernels (ACKs) as component parts, to address these challenges. The DSN use convolutional neural networks to estimate nonlinear source-target relationships, ACKs allow these relationships to vary over time, and time lags are estimated by analysis of ACK coefficients. The proposed model was validated on synthetic data and simulated data by the STANCE fMRI simulator, whose time-varying causal relationships were programmed into the data a priori. Also, the identified time-varying causal relationships from the proposed method were compared with those from previous methods, generalized vector autoregressive model and particle filtering. In addition, we applied our model to a real-world task fMRI dataset.
Results: The proposed method correctly identified the true Granger causality in both synthetic and simulated datasets. In addition, the time dependence of the causal relationships was correctly tracked by the ACK coefficients that quantified each time lag. Both competing methods failed to estimate Granger causality coefficients accurately. In the real-world data, lag 0 Granger causality coefficients were estimated to be nonzero among four ROIs known to be task-related.
Conclusion: Our DSN-ACK architecture that characterizes time-varying nonlinear conditional Granger causality identifies time-varying causal relationships programmed into synthetic and simulated fMRI data. When applied to real-world task fMRI data, the method identifies plausible causal brain functional relationships among brain regions that prior methods were unable to identify. Our method is promising for modeling complex functional relationships within brain networks.
Funding Support, Disclosures, and Conflict of Interest: Funding for this work was provided by NIH grants R01AG041200 and R01AG062309 as well as the Pennington Biomedical Research Foundation.
MRI, Functional Imaging, Brain