Correlation Analysis of Multivariate Neural Signals with Massively Parallel Wavelet Coherence

The study of the correlations that may exist between neural signals generated by different brain regions is a critical issue in understanding brain functions. There is a lack of an appropriate approach which is capable of (1) estimating the correlation between neural signals, and (2) adapting to the quickly increasing scales and sizes of neural signals. Wavelet coherence is an effective method for investigating the interaction dynamics between neuronal oscillations. Continuous wavelet transform (CWT) is suitable for analyzing neural signals which are non-stationary in nature. CWT forms the basis of wavelet coherence methods. However, the wavelet coherence method has been largely hampered by the high complexity of CWT. Aiming at this problem, this study proposed an improved wavelet coherence method with parallelized wavelet transform upon General-purpose computing on the graphic processing unit (GPGPU), namely, GPGPU-enabled wavelet coherence (GWC). The proposed method has been used to analyze 32-channel ERP recordings. The experimental results indicate that (1) the correlation between different neural signals can successfully indicate the different reactions of brain regions when people are watching videos and (2) GWC solves the performance bottleneck with the conventional counterparts.