COMPSTAT 2018: Start Registration
View Submission - COMPSTAT2018
Title: Decision thresholds using Hilbert-Huang transforms in fMRI applications Authors:  Po-Chih Kuo - Institute of Statistical Science, Academia Sinica (Taiwan) [presenting]
Michelle Liou - Institute of Statistical Science (Taiwan)
Philip Cheng - Institute of Statistical Science (Taiwan)
Abstract: Statistical decision on the brain activation maps in functional magnetic resonance imaging (fMRI) time series requires two steps: (1) calculating the statistics in brain regions; (2) thresholding the statistics using a criterion. In Step (2), parametric methods usually make an assumption on the asymptotic distributions of statistics. Alternately, nonparametric procedures determine thresholds using surrogate distributions. For example, a surrogate data approach randomly permutes the phases of raw time series based on Fourier transform which preserves the stationary structure of data. However, spontaneous and induced brain responses in the real world are non-stationary. To incorporate possible non-stationarity in fMRI time series, a randomization procedure based on the Hilbert-Huang transforms is proposed. Two fMRI datasets with either stationary or non-stationary properties were used in our experiment. The significance of individual voxels was determined by comparing the empirical data against the surrogate data distribution. When compared with the results using conventional phase-randomization and wavelet-based permutation methods, the proposed method provided activation maps revealing essential brain regions while filtering out noises in the white matter. Our work shows the importance of considering the non-stationary nature of fMRI time series when selecting resampling methods for probing brain activity or functional networks in real-life fMRI experiments.