Plenary Lecture
Exploratory Analysis of Functional Brain MR Multimodal Imaging Sleep Data
Professor Radu Mutihac
Department of Physics
University of Bucharest
Romania
E-mail: mutihac@gmail.com
Abstract: Analysis of event-related biomedical functional time series is a complex task due to temporal dispersion of the hemodynamic response and aliasing of physiological noise. The present contribution addresses the exploratory analysis of EEG-correlated functional MR (fMRI) brain imaging simultaneously acquired data during deep sleep. Fused EEG-fMRI in imaging neuroscience features noninvasively capture of electrophysiological and hemodynamic activity of the intact human brain with optimal spatiotemporal resolution. While exploratory analysis like independent component analysis (ICA) is meant to generate testable hypothesis (models), the goal of confirmatory analysis like general linear modeling (GLM) is to test their statistical significance. A dynamic interplay between hypothesis generation and hypothesis testing maximizes chances of successfully dealing with the increasingly complex experiments resulting in large data sets.
Common sleep loss effects like slowed response times, a narrowing of attention, and an increased propensity to initiate sleep were initially studied with cognitive/psychomotor measures. With the advent of functional MR brain imaging, sleep loss effects have been correlated with brain deactivation, especially in the prefrontal cortex, inferior parietal/superior temporal cortex, thalamus, and anterior cingulate [Thomas et al., 2000], so that the interest has gradually shifted towards higher-order cognitive functions like problem solving and moral reasoning. The precise neurophysiologic mechanisms driving such deactivation, and ultimately the function of sleep itself, remain unclear though. Present challenges in sleep studies consist in delineation of the cognitive abilities affected by specific cognitive tests run during sleep loss.
To this end, we introduced group multispectral ICA of EEG data to produce the relevant time courses of activation during rest/sleep in a group of 29 healthy subject in the range of 0.005 to 8 Hz, which were subsequently used as regressors in a framework of general multilinear statistical analysis of fMRI data (SPM). Data model order selectionorder was estimated by a weighted sum of the Minimum Description Length (MDL), Schwarz Bayesian Information Criterion (SBC), and Akaike’s Information Criterion (AIC).All parametric linear and non-linear models can be validated by checking the behavior of the model residuals.Since no validation data sets were available, we searched for any form of structure present in residuals employing a structure measure [Friedman, 1987], subsequently related to mutual information [Mutihac and Gillen, 2004]. Hypnograms were overlaid on the time courses of activation for each frequency spectrum chunk, and the associated brain activity maps were determined by group SPM.
Our novel approach allows to run 2nd level mixed-effects SPM analysis of fused EEG-fMRI time series when no experimental paradigm is available (e.g., rest and sleep), and zooming in the frequency power spectrum at any resolution (i.e., as many channels as needed). Exploratory analysis and data mining techniques applied to sleep data will presumably come out with new hypotheses regarding the physiological basis of sleep mechanisms and recovery during sleep. Such results may bring exploratory methods closer to medical practitioners as an exquisite and friendly approach in biomedical investigations and computer-aided diagnosis.
Brief Biography of the Speaker: Professor Radu Mutihac got his PhD in Physics at the University of Bucharest in 1994 and became full professor in 2000. His main research fields have been signal and particularly image processing, microelectronics, and artificial intelligence. As postdoc/research associate/visiting professor/full professor he run his research and didactic activity at the University of Bucharest, 1981-on, at the International Centre for Theoretical Physics (Trieste, Italy, 1993-2004), EcolePolytechnique (Palaiseau, France, 1993), Institut Henri Poincare (Paris, France, 1998), K.U. Leuven (Belgium, 2000-2001).
In 1993 through 1994, he was appointed Deputy General Director in the Higher Education Department of the Romanian Ministry of Education and Research (MEC) acting as the representative of the MEC in the Organization for Economic Co-operation and Development (OECD).
Data mining and exploratory analysis of biomedical signals were the subjects which he dealt with during two Fulbright Grants in Neuroscience: at the Yale University School of Medicine (New Haven, CT, 1999-2000) and at the University of New Mexico (Albuquerque, NM, 2010-2011). Most of his significant research in fused biomedical imaging modalities was carried out at the Johns Hopkins University and Kennedy Krieger Institute, F.M. Kirby Research Center for Functional Brain Imaging (Baltimore, MD, 2003-2005), National Institutes of Health (Bethesda, MD, 2011-on), and Walter Reed Army Institute of Research (Silver Spring, MD, 2011-3013).
Professor Radu Mutihac is member of the ISMRM, ESMRMB, OHBM, Romanian US Alumni Association, and fellow of Signal Processing and Neural Networks Society IEEE, as well as referee for several journals of the Institute of Physics (London, UK), Neural Networks (Elsevier), IEEE Transactions on Image Processing, and evaluator/expert for the ISMRM, OHBM, CORDIS, ARACIS, CNCSIS, UEFISCDI, and the Romanian – U.S. Fulbright Commission.