Show simple item record

dc.contributor.authorLavoie, Michael Rob
dc.date.accessioned2015-02-04T05:30:19Z
dc.date.available2015-02-04T05:30:19Z
dc.date.issued2014-08
dc.identifier.otherlavoie_michael_r_201408_ms
dc.identifier.urihttp://purl.galileo.usg.edu/uga_etd/lavoie_michael_r_201408_ms
dc.identifier.urihttp://hdl.handle.net/10724/30976
dc.description.abstractDICCCOL is a process that identifies common connectivity in the brain. It was developed to show that the cortex has a common structure thereby identifying functional correspondence. The tool compares the connectivity in a subject brain against a reference library of structural correspondence. A set of bundles is processed for comparison against this library. The result is a subject’s fiber bundle that most closely matches the libraries reference bundle. The data set is relatively small but the processing is extensive. A single thread approach to the process is very time consuming. This task is better suited for a parallel processing approach. I show how the work can be accomplished more efficiently with GPU hardware and CUDA’s parallel programming, resulting in a speedup factor of better than 6.
dc.languageeng
dc.publisheruga
dc.rightspublic
dc.subjectDICCCOL
dc.subjectCUDA
dc.subjectGPGPU
dc.subjecteigenvector
dc.subjecteigenvalue
dc.subjectmedical imaging
dc.titleA parallel approach to DICCCOL
dc.typeThesis
dc.description.degreeMS
dc.description.departmentComputer Science
dc.description.majorComputer Science
dc.description.advisorTianming Liu
dc.description.committeeTianming Liu
dc.description.committeeThiab R. Taha
dc.description.committeeThiab Taha


Files in this item

FilesSizeFormatView

There are no files associated with this item.

This item appears in the following Collection(s)

Show simple item record