fMRI Research via Database Mining, Management

Gazzaniga, Michael

Abstract

Large-scale databases of complete fMRI study data from the peer-reviewed literature offer valuable opportunities for innovative research projects to obtain new knowledge about human brain function. In addition, they provide broad ranging exemplars of experimental protocols, designs, and parameters for developing leading edge technological solutions enabling exploration and examination of these rich neuroscientific data sets. This multi-center project seeks to utilize archived fMRI study data to guide novel, hypothesis-driven, fMRI experimentation into the neurophysiological correlates of cognitive function. Under this proposal, the fMRI Data Center (fMRIDC), serving as the core center, and its collaborators at Berkeley, Santa Barbara, and the University of Toronto will: i) mine data from this unique repository of neuroimaging data to examine individual differences between subjects in the characterization of group-level and 'default state' BOLD activity across multiple cognitive domains; ii) devise continuously updating representations of the current content of the fMRI study archive using multivariate and study clustering methods to visualize emerging trends in fMRI experimentation and map study, subject, and domain-specific variation in BOLD activation; iii) identify and validate summary statistical measures of BOLD activity that may be used to rapidly assess similarity across a large number of fMRI time courses; iv) research and deploy software to dynamically analyze the published fMRI literature to guide the construction of an fMRI study ontological framework for containing information relevant to how fMRI studies are performed and reported; iv) with collaborators from UC Berkeley, develop and deliver freely-distributable software tools for fMRI study data management that simplify the process of data exchange; v) with collaborators from the Rotman Institute, examine fMRI processing pipelines in order to identify those strategies best balancing the bias-variance tradeoff, thereby optimizing inferences that can be made from fMRI data given scanner, experimental, and other protocols. Functional imaging data sets will be obtained to validate methods for optimized data processing and for summary statistical calculations. With collaborators from UC Santa Barbara, using knowledge and tools obtained from these projects, we will conduct novel fMRI studies of human cognitive processes, in particular, testing specific hypotheses regarding individual differences between subjects in episodic memory function. The software tools for fMRI data management will be essential for accurately encapsulating study meta-data and processing optimization methods will be critical for accurate statistical modeling image time courses. The long-term outcomes of this project will be the extraction of new knowledge about brain function from previously published fMRI study data and its usefulness in guiding original, hypothesis-driven, neuroimaging research. Additional, valuable outcomes of the project will be i) improving collaborator and other user remote access to the archive for resource sharing and data mining by tying directly into the Dartmouth Internet2 connection; ii) enable greater remote processing capability using powerful Grid technology to assist distributed uses of the archive; iii) enhancement and extension of data base capabilities to add greater online fMRI data visualization for website users through interaction with other online neuroimaging data resources; and, iv) supporting open neuroscientific data sharing and curation. Overall, this project represents a unique synergistic effort which, in keeping with the mission of the Human Brain Project, will share computational and database resources for the purposes of advancing investigator-initiated neuroscience research and methods development.