 |
 |
 |
|
Back To News>>
11/23/2004
Swartz Center for Computational Neuroscience
A laboratory center of the Institute for Neural Computation, UCSD
Summary of Activities, November 2004
Research Goals
The goal of the Swartz Center for Computational Neuroscience is to observe and model how functional activities in multiple brain areas interact dynamically to support human awareness, interaction and creativity. Research in the Center involves development computational methods and software, experimental methods and equipment, collection and analysis of human cognitive experiments, and collaborations to analyze data collected by other groups in such experiments. The core facilities of the Center include a network of Linux workstations and a unique state-of-the-art 256-channel electroencephalographic (EEG) acquisition system that can be reconfigured as a two-person-each-136 channel system for simultaneous data collection from two subjects performing separate or interacting tasks.
Overview of Current Research Activities
Current research projects in the laboratory include:
Modeling long-range brain interactions. We are using independent component analysis (ICA) and time/frequency analysis techniques to study a phenomenon we have detected in high-dimensional EEG data, theta band synchronization events (TSEs). TSEs appear to be produced in the EEG only in response to significant events – events with immediate implications for reshaping behavioral planning. We are studying these events using a set of continuous performance tasks involving frequent choice behavior and immediate performance feedback.
Neuroscience of social interactions. We are using two-person EEG and video face tracking to study the brain dynamics support real-time interactions between subjects in face-to-face situation such as competitive and cooperative game playing, comedy video viewing, etc.
EEG dynamics of emotional processing. We are studying EEG dynamics during 3-5 minute periods when the subject actively imagines a situation provoking a suggested emotion, attempting to recreate the bodily feelings associated with the emotion. Study of data from 15 such emotions shows that EEG dynamic changes in different emotional states are complex, requiring multidimensional modeling.
EEG dynamics of memory processing. During short term working memory, 4-8 Hz theta band activity in the EEG recorded over the frontal midline increases in mean amplitude. Our research shows that the mean increase only captures one aspect of the dynamics of the EEG changes associated with memory processing. We have isolated several dynamic modes involving at least three frequency bands in which dynamic changes occur in frontal midline EEG related to current memory load. Other brain locations show a variety of other dynamic changes that this project seeks to model and to relate to behavior, task context, etc.
Multiscale analysis of concurrent scalp and intracranial data. In this project, we are analyzing EEG and intracranial EEG (iEEG or ECOG) data collected by our collaborator Dr. Greg Worrell of the Mayo Clinic, Rochester MN, and soon, by our UCSD collaborator Dr. Robert Buchanan. We are using ICA to determine the relationship between electrical field activity recorded in the brain and on the scalp. This project will investigate the degrees to which synchronization of cortical field activity at smaller spatial scales produces activity recorded at a larger spatial scale on the scalp, and how much activity reaching the scalp can contribute to analysis of local phenomena collected from within the brain itself.
Multimodal brain imaging. We have collected 70-channel EEG data during full-rate fMRI scanning at UCSD, and intend to develop analysis methods for concurrent high-density EEG and MEG data from the UCSD MEG system expected to be installed in the coming year. Combining these dynamic brain imaging methods with structural MR scanning will open new windows into the relation of macroscopic brain dynamics to behavior and experience.
Advanced applications of statistical signal processing to dynamic brain data. Several efforts are underway in the Center to apply current developments in blind source separation to EEG and fMRI data. These include EMSICA of Arthur Tsai, working with Center associate director Tzyy-Ping Jung. EMSICA (Electromagnetic spectrotemporal ICA) simultaneously maximizes the probability of a linear multi-source model of EEG specified by a time course of activation and a map of relative activity strength on a model of the cortex itself. If successful, this may be an advance in the decomposition of high-dimensional scalp EEG data into anatomically localized sources. Other projects, in conjunction with students of UCSD engineering Profs. Rao and Kreuz-Delgado, are studying the application of sparse decomposition methods to EEG time series data. With a student of Lars Kai Hansen of the Danish Technical University, we are studying the application of convolutive ICA methods to EEG data analysis.
Independent factor analysis of static brain imaging data. With collaborators from Taiwan and the National Institutes of Health, we are applying ICA methods to factor analysis of co-registered MRI, PET and SPECT brain images from clinical subject groups and controls, to determine the brain imaging correlates of clinical disease classification.
EEGLAB and FMRLAB software development and distribution. In 1997 we began putting Matlab functions implementing the new ICA and related analysis methods we developed on our website at Salk Institute for free download. The Swartz Center web site (sccn.ucsd.edu) is now the download home for two software suites, EEGLAB and FMRLAB, for implementing a wide range of analyses on EEG and functional magnetic resonance imaging (fMRI) data, respectively. Development and maintenance of EEGLAB is proceeding under a grant from the National Institutes of Health. We recently hosted the first international EEGLAB workshop on the UCSD campus. We are building these two Matlab software suites as open source environments for nearly any advanced processing of dynamic brain imaging data.
High-dimensional computing. Currrently the laboratory has a network of dual-processor 32-bit Linux workstations interconnected by 100mbit Ethernet and Sun Grid Engine software. We do much of our development and visualization in Matlab, and code time-intensive functions in C or Fortran. This year UCSD installed a fiber cable between SCCN and UCSD. Currently, this connection uses one gigabit fiber. Many more fibers are available and UCSD has agreed to upgrade our connection as our needs increase. Our in-house 256-channel EEG and proposed UCSD 435-channel MEG/EEG systems generate too much data to be processed at one time using 32-bit workstations. We therefore are planning to acquire and exploit a cluster of 64-bit workstation blades for routine and exploratory data processing. We have had extensive discussions with SDSC about tying in to their high-capacity SRB data storage system and plan a 2-Tbyte front end to hold data retrieved from SRB for processing and visualization at SCCN and to buffer new data for transfer to SDSC.
Supercomputing. Our work shows there is a great deal of information about human brain dynamics available in high-density EEG data that is not captured by standard analysis methods. We are currently working with engineering collaborators at UCSD and Danish Technical University to implement advanced EEG signal processing algorithms on the current parallel IBM machine at the San Diego Supercomputer Center (SDSC). We hope to be early users of the advanced IBM Blue Gene machine scheduled to be installed at SDSC in January.
Helpful Links
Swartz Center for Computational Neuroscience at UC
|
|
 |
|
Saturday, December 21, 2024
|
| |
The Swartz Foundation is on Twitter: SwartzCompNeuro
|
|
|
 |
print