Dissertation Information for Ulukbek Ibraev NAME: - Ulukbek Ibraev
DEGREE:
- Ph.D.
DISCIPLINE:
- Library and Information Science
SCHOOL:
- Rutgers University (USA) (2005)
ADVISORS: - Paul B. Kantor
COMMITTEE MEMBERS: - Tefko Saracevic - Gheorghe Muresan - Sven Josef Dickinson
MPACT Status: Fully Complete
Title: Imposing graph structures on space-time densities for indexing and retrieval
Abstract: fMRI is a brain imaging technique that allows us to see traces of neural activation caused by cognitive processes in the human brain, with a few seconds delay. The development of the fMRI technique, with its relatively high temporal and spatial resolution and non-invasive nature, opened a new era in functional neuroimaging. We can expect rapid growth of fMRI databases and with that the need to organize, index, store and retrieve this data.
The focus of this research is to explore using graph structures to represent cognitive processes in a brain and to assess their effectiveness for fMRI retrieval. The idea that it might be possible to represent dynamic brain images by graphs representing 4D space-time structures, where nodes represent activated brain areas and edges represent interactions between them, was put forward in a series of proposals by Paul Kantor and Stephen Hanson, the third of which was funded by the NSF, supporting the work reported here. In the present work we did not attempt to validate that concept in terms of the specific space time localization of the activated brain regions reported by the graph. Instead, we seek to validate it in the context of query by image content. In that context we faced two problems: extracting the graph, and assessing the retrieval performance. The author has proposed the idea of linking time slices by a correlation measure (maximum correlation graph). The validation was in terms of the ROC curves for every possible retrieval case.
The results showed that the correlation-linking approach provides better retrieval performance than an alternative based on an anatomical representation of the brain (Brodmann vector). The Brodmann vector-based approach performed better than random guessing in 85% of all cases and the correlation based approach performed better than the random guessing in 98% of all cases. This was much better than chance. Another method of link finding (minimum distance graph) was also shown to give performance comparable to the correlation-based links and performed better than chance in 98% of all cases. The results of the experiments supported the proposed approach for fMRI data retrieval.*
*This dissertation is a compound document (contains both a paper copy and a CD as part of the dissertation).
| |
MPACT Scores for Ulukbek Ibraev A = 0
C = 0
A+C = 0
T = 0
G = 0
W = 0
TD = 0
TA = 0
calculated 2008-01-31 06:30:52
Advisors and Advisees Graph
|