Analysis of functional neuroimages using ICA with adaptive binary sources

Abstract The low signal-to-noise ratio and the many possible sources of variability makes recordings from non-invasive functional neuroimaging techniques a most challenging data analysis problem. Independent component analysis (ICA) is currently a popular—although highly controversial—approach for exploratory analysis of the extensive amount of data acquired during functional magnetic resonance imaging (fMRI) studies. Since most common algorithms for independent component analysis are computationally demanding some sort of data reduction is usually required. In this paper we present a computationally efficient mean field algorithm for noisy independent component analysis (ICA) which makes it possible to carry out fast exploratory analysis on unreduced fMRI datasets. We assume adaptive binary sources and determine the number of hidden sources using the Bayesian information criterion (BIC) in which the Thouless–Anderson–Palmer (TAP) free energy is used as an approximation to the likelihood. We illustrate the method on both an artificial data set and a set of functional neuroimages from a visual activation study.

[1]  S Makeig,et al.  Spatially independent activity patterns in functional MRI data during the stroop color-naming task. , 1998, Proceedings of the National Academy of Sciences of the United States of America.

[2]  G. Parisi,et al.  Statistical Field Theory , 1988 .

[3]  Ole Winther,et al.  Ensemble Learning and Linear Response Theory for ICA , 2000, NIPS.

[4]  E C Wong,et al.  Processing strategies for time‐course data sets in functional mri of the human brain , 1993, Magnetic resonance in medicine.

[5]  Te-Won Lee,et al.  Independent Component Analysis , 1998, Springer US.

[6]  L. K. Hansen,et al.  On Clustering fMRI Time Series , 1999, NeuroImage.

[7]  L. K. Hansen,et al.  Generalizable Patterns in Neuroimaging: How Many Principal Components? , 1999, NeuroImage.

[8]  M. Opper,et al.  Tractable approximations for probabilistic models: the adaptive Thouless-Anderson-Palmer mean field approach. , 2001, Physical review letters.

[9]  M Barinaga,et al.  What makes brain neurons run? , 1997, Science.

[10]  Ole Winther,et al.  Mean-Field Approaches to Independent Component Analysis , 2002, Neural Computation.

[11]  Karl J. Friston,et al.  Characterizing the Response of PET and fMRI Data Using Multivariate Linear Models , 1997, NeuroImage.

[12]  R L DeLaPaz,et al.  Echo-planar imaging. , 1994, Radiographics : a review publication of the Radiological Society of North America, Inc.

[13]  P. Bandettini,et al.  Echo - planar magnetic resonance imaging of human brain activation , 1998 .

[14]  Scott L. Zeger,et al.  Non‐linear Fourier Time Series Analysis for Human Brain Mapping by Functional Magnetic Resonance Imaging , 1997 .

[15]  Thomas E. Nichols,et al.  Statistical limitations in functional neuroimaging. II. Signal detection and statistical inference. , 1999, Philosophical transactions of the Royal Society of London. Series B, Biological sciences.

[16]  Marcia Barinaga,et al.  Neuroscience: What Makes Brain Neurons Run? , 1997, Science.

[17]  Thomas E. Nichols,et al.  Statistical limitations in functional neuroimaging. I. Non-inferential methods and statistical models. , 1999, Philosophical transactions of the Royal Society of London. Series B, Biological sciences.

[18]  Lars Kai Hansen,et al.  Blind Separation of Noisy Image Mixtures , 2000 .

[19]  David Heckerman,et al.  Asymptotic Model Selection for Directed Networks with Hidden Variables , 1996, UAI.

[20]  Michael I. Jordan.  Learning in Graphical Models , 1999, NATO ASI Series.

[21]  Terrence J. Sejnowski,et al.  An Information-Maximization Approach to Blind Separation and Blind Deconvolution , 1995, Neural Computation.

[22]  G. Schwarz.  Estimating the Dimension of a Model , 1978 .