淡江大學數學系演講公告
主 講 人：蔡  志  鑫    教授
        中央研究院 統計科學研究所
講   題： Independent component analysis of EEG group studies by reproducibility
日    期：99年5月11日（星期二）
時    間：下午2:30 –3:20
地    點：數學系(科學館S433室)
摘    要：
Independent component analysis (ICA) extracts components from several 
channels of EEG electrodes. Extracting components of interest always involves 
using an a priori spatial or temporal template as well as expert experience. 
However, in ICA one might select a “component of interest” which might not be 
equivalent to brain functional processes expected. Reproducibility evidence can 
be estimated by assessing those independent components that are consistent 
between subjects and quantifying whether the components extracted from the 
two groups have sufficiently similar features from the remaining (unextracted) 
components. A pair of independent components from two subjects can resemble 
each other in many ways (e.g., scalp maps, power spectra, ERP, ERSP etc.). 
Finding reproducible components between subjects might be non-trivial. In the 
literature, the tensor and group ICA methods have been introduced to search for 
reproducible evidence (Calhoun et al., 2001; Beckmann and Smith, 2005). In 
this talk, we propose using a statistical test for the association between 
independent components based on information decomposition of continuous and 
discrete data. Empirical applications of the methodology suggest that 
reproducible evidence is robust to small sample sizes and sensitive to both the 
magnitude and persistency of cortical activities (Liou et al. 2003). We extend 
the methodology proposed by Liou et al (2005) for collecting reproducible 
evidence in EEG localized sources. The reproducible evidence implies those 
estimated spatial-wise sources across several sets of single trials.

EEG epochs of 2 sec in 20 healthy right-handed subjects (24±4.0 years) during 
stop-signal paradigm task were analyzed by means of EEGLab software 
(Delorme et al. 2004). Trait anxiety level was measured using the Chinese 
version of the State Trait Anxiety Inventory (Spielberger et al., 1970; Shek, 
1993). For each subject, we computed their 122 ICA components as well as 
corresponding ERSP. The ROC curves (x-axis: false alarm, y-axis: sensitivity) 
by reproducibility analysis for threshold selection were plotted where the 
optimal threshold on the curve was selected by maximizing the Kappa index 
(shown in green line) due to Cohen (1960). The reproducibility index is 
therefore obtained by the selected threshold which shown 14 components has 
strong reproducibility (>90% of total number of subjects). Since significant tests 
are identified between similar components between subjects, it therefore leads 
to suggesting reproducibility evidence for research findings in EEG experiments.


           
歡 迎 參 加             敬 請 張 貼
淡江大學數學系 敬啟