Windowless Dynamic Functional Connectivity

(c) 2019 Shih-Gu Huang, Moo K. Chung
University of Wisconsin-Madison


Dynamic correlation offers far superior modeling flexibility over static correlation in functional brain signals (Chung et al. 2019) [3]. Sliding window (SW) and tapered sliding window (TSW) methods are the most common approaches in computing dynamic correlations between brain regions. However, due to data acquisition and physiological artifacts in resting-state fMRI, the sidelobes of the window functions in spectral domain will cause high-frequency fluctuations in dynamic correlations. To address the problem, we propose to define the heat kernel, a generalization of the Gaussian kernel, on a circle continuously without boundary. The windowless dynamic correlations are then computed by the weighted cosine series expansion, where the weights are related by the heat kernel. The method is published in Huang et al. (2019) [1].

The MATLAB code will perform the SW, TSW and windowless heat kernel method in a simulated network data with the ground truth. The state space is estimated using the k-means clustering that was explained in Huang et al., 2018 [2]. If you use any part of code, please reference Huang et al. (2019) [1].


[1] Huang, S.-G., Chung, M.K., Carroll, I.C., Goldsmith, H.H. 2019 Dynamic functional connectivity using heat kernel. IEEE Data Science Workshop (DSW), in press

[2] Huang, S.-G., Samdin, S., Ting, C.M., Ombao, H., Chung, M.K. 201 Statistical model for dynamically-changing correlation matrices with application to resting-state brain connectivity. arXiv:1812.10050

[3] Chung, M.K, Lee, H., Ombao, H., Solo, V. 2019. Exact topological inference of the resting state brain networks in twins, Network Neuroscience, in press