研究目的
To propose a novel method for generating surrogate graph signals that preserves correlation structure as captured by the graph Laplacian, combining the graph spectral decomposition with a sign-randomization of the graph Fourier coefficients.
研究成果
The proposed framework for the generation of surrogate graph signals based on the GFT and sign-randomization of the real-valued GFT coefficients maintains autocorrelation and can be used to generate a null-distribution of correlation coefficients to evaluate the similarity between EEG topographical maps, contributing to analyze the effects of behavioral tasks on single-trial EEG topographies.
研究不足
The amplitude histogram of the surrogate graph signal is not guaranteed to match the one of the original signal, and sign-randomization of real values has less degrees-of-freedom than phase randomization of complex values.
1：Experimental Design and Method Selection:
The method involves transforming the measured graph signal into the spectral graph domain, randomizing by permuting or randomly generating the signs of the GFT coefficients, and then applying the inverse GFT to obtain a realization of the surrogate graph signal.
2：Sample Selection and Data Sources:
A high-density EEG dataset of 235 channels was used to demonstrate the feasibility of the proposed method.
3：List of Experimental Equipment and Materials:
EEG cap with 235 electrodes.
4：Experimental Procedures and Operational Workflow:
The adjacency matrix was built based on the Euclidean distance between electrodes, and the GFT was defined using the eigendecomposition of the normalized Laplacian. Surrogate graph signals were generated and used to assess the similarity between EEG topographical maps.
5：Data Analysis Methods:
The null distribution of the Pearson correlation coefficient was obtained from the surrogate graph signals to evaluate the similarity between EEG topographical maps.
独家科研数据包，助您复现前沿成果，加速创新突破
获取完整内容
