image: David Wipf is a recent graduate of the electrical and computer engineering Ph.D. program at UC San Diego's Jacobs School of Engineering. Thanks to Wipf's theoretical work, functional brain imaging practitioners should be better able to assess the relative strengths and weaknesses of competing Bayesian approaches for the source localization of neuroelectromagnetic fields. view more
Credit: UC San Diego
December 06, 2006 -- David Wipf, a recent graduate of the electrical and computer engineering Ph.D. program at UC San Diego’s Jacobs School of Engineering, has won a 2006 Outstanding Student Paper Award at a prestigious conference for his work on human functional brain imaging. The conference -- the 2006 Neural Information Processing Systems Conference or NIPS is being held in Vancouver this week.
"With this work, functional brain imaging practitioners should be better able to assess the relative strengths and weaknesses of competing Bayesian approaches for source localization," said David Wipf, who performed the research and wrote the paper while at UCSD.
"NIPS is a premier conference and this is quite an achievement," said Bhaskar Rao, an ECE professor at UCSD’s Jacob’s School and David Wipf’s Ph.D. dissertation advisor.
The new work, which is largely theoretical, may also lead to improvements of existing algorithms that attempt to determine what parts of the brain are producing the electromagnetic fields that are measured by functional brain imaging techniques such as magnetoencephalography (MEG) or closely-related electroencephalography (EEG).
MEG and EEG use an array of sensors to take electromagnetic field measurements from on or near the scalp surface with excellent temporal resolution. Using this information to create accurate maps of neural activity with the highest possible spatial and temporal resolution, and relating these time-and-space activity patterns to behavioral, perceptual, cognitive and motor processes is one of the ultimate goals of human functional brain imaging. However, determining exactly what parts of the brain produce the electromagnetic fields is a difficult and unresolved issue.
Researchers often use Bayesian statistical methods and algorithms to try to determine the source, within the brain, of recorded neuroelectromagnetic fields. Trying to answer the source localization question requires the incorporation of prior assumptions -- and Bayesian methods are useful in this capacity because they allow these assumptions to be quantified. There are, however, many different Bayesian approaches for researchers to choose from; and it is difficult to know which approach is right for any given experimental situation.
In their new paper, David Wipf and colleagues at UCSD’s Signal Processing and Intelligent Systems Lab and UCSD’s Swartz Center for Computational Neuroscience stepped back and studied many of these Bayesian approaches. Their work yielded a generalized framework that encompasses many of the established Bayesian techniques for imaging neural activity.
The paper took one of three 2006 Outstanding Student Paper Awards at NIPS, chosen from a pool of 62 nominations. David Wipf is the first author and his coauthors -- Rey Ramírez, Jason Palmer, Scott Makeig and Bhaskar Rao -- are all from UCSD. The paper is entitled "Analysis of Empirical Bayesian Methods for Neuroelectromagnetic Source Localization."
Wipf and colleagues conclude their paper with the following: "By developing a general framework around the notion of automatic relevance determination (ARD), deriving several theoretical properties and showing connections between algorithms, we hope to bring an insightful perspective to these techniques."